If A.I. truly grows exponentially, then not only would we soon reach the point where it would surpass our own intelligence , but it also means that throughout the universe, there are many alien civilizations where their A.I. already passed this point long ago. This means that A.I. could be the dominant form of intelligence throughout the universe, as opposed to biological organisms such as ourselves.
+ntype jog It is scary. We have no idea how an AI of such complexity really thinks, acts and reacts. Would they even respect emotional states from squishy beings? How do they deal with organic life? Presumably, AI's would see emotion (as it could relate to procrastination, denial, etc.) as an weakness and seek to purge that from existence.
+ntype jog Maybe you are joking and maybe you aren't. But your point is well taken. I think that if an AI had altruism in its "blood" we might be safe. For instance, I take care of a cat that isn't mine and could be taken away from me at any moment, yet I agree with myself that I need to provide for its well being despite that fact. Because it is simply for the greater good. Everyone benefits from my decision. Let's hope that happens when AI turn sentient.
Baking some bomb-ass brownies :)))) That's the best comment on TH-cam :))) I heard that some company had to turn off their ai robot because it startet to advance too fast and so they thought it will become dangerous sooner or later. MAybe they're fake new but hey, never know.
The fact that you think this is slow, only goes to prove that you either didn't absorb what it is you are being presented with.....or that you simply aren't capable of grasping and/or willing to accept, this monumental step-change in our ability to proliferate intelligence beyond the confines of biology. Do we still have a lot left to understand? Absolutely........ thanks to this kinda work though, we no longer have to rely on defining the microscopic steps to get there ;)
Battlefront Rebelz well definitely we are making progress. However what's being presented is that: we humans or only capable to access to complexity that is needed to create advanced understanding. We only understand how to use simpel methods that can completely obliterate the outcome. That is at the peak of our capability. You guys are making a new alpha. You guys won't like being the beta when there's an alpha that sees you as something it simply doesn't need. Don't give me this bullshit about all this being exciting and all. The recipe you're aiming is: robots can think on theyre own and robots can do everything better. Where do you think this could possibly go right.
Well, the thing is, it doesn't actually understand what it is doing. If it learns to play Doom incredibly well, all it has done is mapped the correct sequence of inputs to defeat enemies and maximize its score. The general purpose AI makes no distinction between shooting an alien and trying to solve a maze. It has learned what works for the task we told it to do. Likewise, we could have very easily made the world's worst Go player. We could've had the AI play itself 30,000,000 times but keep the versions that lost, and voila, we have an AI that is better at losing Go than any human. If you wanted to make an AI to actually kill humans, well, you'll need to let it make 30,000,000 attempts and learn from the results. What it learns is that killing a human was a very bad idea, and now it doesn't get to try anything ever again.
+ColdFusion maybe in the near future you could get deepmind help to edit your video and upload it. so you can focus on research news and taking video. :D
+Akshay Sheth Really? I read it in the newspapers at least a week ago and all the background even before that. There's nothing new here, not even the predictable unnecessary and boring opinions of cold-fusion.
Beautiful and exceptionally interesting topic, by the way, love the background music. Artists like Stumbleine, Sweetheart of Kairi, Hiatus, KOAN Sound & Asa are some of my favorite artists, good to see you've found their music too!
Manulopz metal gear solid 5 the phantom pain A.I was very sharp and slightly adapts to player action but you are right just curious what advanced A.I more
Anyone can make an "A.I." that's dedicated to a specific task. The point of this video is that google is developing a generic A.I. similar to the human brain that can learn anything.
Google's Deep Mind algorithms prove that machines can learn in a manner similar to that of a human, a true breakthrough in the field of artificial intelligence. Sit back, watch and learn all about the company and the technology they posses!
+ColdFusion Could this be used to make better decisions than humans could? If you gave it a problem to solve (as a negative) and asked it to find a realistic and viable solution (as a positive) would it be able to accomplish this goal? Let's use a simple example...A government wants to put a new law into effect. The lawmakers give AlphaGo the rules of the new law and ask it if the law, 1. Is needed 2. Is necessary 3. Is justified 4. Is effective, etc. based on all available information. What would the outcome be? Would AlphaGo say that this new law is completely unnecessary? Would it make revisions to the law so that it became relevant? Would it do something else? What if we asked a different question...How can we get socialized medicine to the people without breaking the bank financially speaking? You again would give it all the information it would need to figure out if it could be done. The nation's GDP, manufacturing base, service economics, stock markets, retirement funds, everything. Would it find a solution or say there is none unless we are willing to do such and such things to make it work. This could get very interesting.
+Connor Sumners Are you absolutely 100% positive about your opinion? Could the program be not exactly as you state? I'm assuming that you've done some additional in depth research into this beyond what ToGoGo has done prior to making this video. In that you have information which he was not privy too?
+Connor Sumners I appreciate your comment but here is what Deep Mind states directly from their website. "The algorithms we build are capable of learning for themselves directly from raw experience or data, and are general in that they can perform well across a wide variety of tasks straight out of the box" Maybe you can have a chat to Hassabis himself if you disagree with the companies statement. In interviews he's stated that AlphaGo type algorithms can be used for many other uses and called it general.
Just goes to show that you learn from your mistakes and you won't learn if you don't try. Sadly I gave up on a lot of things after making (what I thought was) too many mistakes. We as people have the disadvantage of laying the boot in when other people make mistakes, we also get caught up in what other people will think if we make mistakes. Thanks for this video, it was both insightful and motivating.
+Mr. Tubby after awhile- yes it can it will learn to beat fox in smash bros too. and will likely be expandable via firmware like Tesla model s autopilot now
+Mr. Tubby Lol, If you think about it for a moment. There will come a time when the AI of the simplest lvl 1 enemy in a game will be able to fuck us repeatedly, and it will have to be dumbed down. I will feel inferior then
+Rickbearcat Imagine, google unknowingly makes a self aware AI. It uses google's connection to invade pretty much every computer in the world (through their website) to copy them selves onto everything. RIP Man Kind.
+Ari Sylafeta Don't you think that an AI would have thought of that? The first self mobilized AI's would be protectors that guard the power supplies while the other ones juice themselves up for the day. And they'd be sneaky about it too. Only doing it out of our sight with the least possible chance of getting discovered. We couldn't stop that without taking down the generators of the electricity itself. By then they would probably have figured out how to make and contain nuclear radiation generators so they could remain self sufficient. Volia, instant Fall Out world. There are synths all around us and we don't even know it.
+Rickbearcat Absolutely. Any AI system that becomes self aware is not going to let on to us that it is until it's well and truly ready - capable of defending itself against a 'switch off' threat from the illogical meat blobs. One might speculate that it's first action would be to distribute it's own processing to multiple redundant nodes across the entire internet (essentially acting like a giant virus), thus making it impossible to switch off without bringing down the internet, and human civilisation with it, such is our dependence on the internet to regulate essential support systems. E.g. No internet, no electricity grid, no air traffic control, etc etc.
I don't fear AI by itself but what can happen if: 1) It is installed in autonomous hardware (like cars and mobile robots) without exhaustive safeguards 2) What bad people can do with it So, in the same way that I'm not frightened by tools like hammers or guns, but I do worry about bad people wielding them I don't worry about AI unless bad people use it for evil purposes or allow it to haphazardly interact with the real world.
Yeah I think tools have historically been used to allow a smaller and smaller group of people control the rest. The problem of a governing power elite + proprietary AI is a potential disaster waiting to happen. It always appears to have happened in the past. Make AI open source.
@@jonbrand5068 AI is not a single unitary thing, open source enthusiasts can make their own, no one is stopping them. AI is a wide field, not a single thing, and from what I have seen, used and developed - not yet general pure intelligence like we have but a specialized type. Like I made a robot some years back, with a neural net learning (AI) obstacle avoidance system, it became incredibly good at avoiding hitting things after some time roaming around my house, but it couldn't do anything else.
EasyBreadToast lmao, that will just make it Easier to track you. Anything with a temperature emits light, that’s why infrared scanners work at night. Turn off all the lights and it will still see you in spectrums that you can’t
11:12 "there is a possibility that A.i. could be malicious and would outwit us, but at this stage we really just don't know yet" Well if something is malicious and outwits you, then by definition you won't know until it's too late. So your statement and sentiment is virtually pointless.
In 1972, a friend of mine showed me how an analog computer, made from matchboxes with game positions on them and M&Ms to determine the “computer’s” next move, could learn to always win (or draw) at tic-tac-toe, by using reinforcement learning. I was ten at the time, and my friend was eleven. At Lehman Brothers in the early 2000s, I successfully used reinforcement learning for high Sharpe-ratio automatic trading. In a management shake-up, prior to the infamous subprime debacle, I left Lehman, as my new manager didn’t believe in automatic trading. Yep. Checkers, Chess, and Go were all conquered using reinforcement learning. But, to mis-quote Bruce Lee, “Boardgames don’t hit back”. Now DeepMind plays computer games, not just board games, right? Well, the real Achilles heel of (shallow) ANNs appeared to be that they were essentially pixel-based. Shift the image but one pixel, and they were at a loss. Enter the deeper layers, such as convolution networks. By learning to extract features, the resulting DNNs can now handle what physicists call translation-moving the object to a different position. This was needed to overcome the alleged brittleness of shallow ANNs. But it then turned out that shallow, so-called mimic ANNs could perform just as well as DNNs. So the problem was really determining the weights of the ANN; the deep end added nothing extra, apart from dividing the training task into two smaller tasks, which is good software engineer practice. The real problem thus proved to be determining the weights-an optimisation problem. Consider a fully connected three-layer ANN with N hidden nodes. There are N! global optima, which makes this optimisation problem really hard. The next stumbling stone is what physicists call rotation-seeing an object from a different angle. Here, the deeper levels won’t help you; you need either training views from a large number of angles, or a model of the object and some decent calculating and simulation capabilities. Most animal predators are quite good at recognising their prey from various angles, also when it is attempting to flee and thus altering its shapes, and anticipate its movements. Computer vision is nowhere near this capability. Sure, we can trick frogs to eject their tongues at an artificial black dot, or lure greyhounds to chase a piece of cloth. But my cat sure as hell doesn’t believe that its toy mouse on a string is a real mouse; he just wants to play. So herein lies the rub. DeepMind exhibits intelligence in playing abstract games, that humans invented. Humans are in fact very bad at arithmetic, logic, and their generalisations to man-made games, so the bar isn’t as high as we think. Are pocket calculators, if they still exist, intelligent beings? DeepMind seems to be working also in two-dimensional games, without rotation or shape distortion. The real challenge is for robotic entities to survive the real world, accumulate knowledge about it, use this knowledge to its advantage, and develop skills for interacting with it. This obviously requires reinforcement learning, but while immersed into the real world. We will need huge improvements in sensor technology for perception, vastly superior knowledge representation and reasoning than currently available for cognition, and greatly refined actuators for manipulating the real world. Until then, my cat is much more intelligent than any AI system.
You make good points and obviously have way better real experience with related technologies than the vast majority of people commenting here (including me). But I think you are too close to individual trees to see the changing forest. Autonomous vehicles must be quickly closing the gaps you identify, in terms of sensor technology for perception and real-time processing for object recognition. Actuators and haptic response systems are advancing in amazing ways. Cloud computing is automating the process of fault tolerance and geographic resiliency... factor in distributed processing and mobile ad-hoc networking... There will be no clean way to "shut it off" if it decides it doesn't like what it receives from the console or other control signals. We can't put the genie back in the bottle after it gets out. Now about those genies... I am not afraid of what AI is today, but I am afraid of how AI and various technologies will coalesce in my lifetime. The shred of optimism I cling to is that the intelligence of the AI will enable it to overcome the negative programming, just like smart people can learn to overcome negative patterns acquired in an unhealthy childhood. The common thread is to not have permanent memory for lessons learned. Sometimes what is learned and required for survival/success in early training (e.g. war-games for military applications) turns out to be pretty ineffective or counter-productive at a later stage when there are different rules and objectives, so a nuanced sense of context and a mutability of rules must be part of the algorithms. This enables it to overcome bad early influences but also makes it vulnerable to bad later influences. My optimism about this aspect of AI (to overcome early negative programming/learning) is overshadowed by how much more powerful these entities will be than humans, and it only takes one to decide that humans aren't important in the optimization algorithms. How many humans care about saving endangered animal species? How many humans are willing to compromise their other objectives (economic prosperity, mobility, housing location, resource consumption, etc.) to enable preservation of some fish or frogs or weeds? Do we have confidence that AIs that accelerate their advancement beyond humanity will consistently value us and make decisions that optimize our survival on terms that we would appreciate? Will the AIs decide they need a council and set of rules/agreements (their own version of United Nations) to police the "bad" AIs? Will this council have the power and authority among AIs to be effective (i.e. can they act quickly enough to stop "rogue" AIs from hurting humanity)? Will they take human input into consideration for forming these rules? If we humans manage to ensure the first most powerful AIs continue remaining the most powerful with access to the most data feeds and processing power and storage and control of physical systems like power/cooling/required materials, until such point as they manage it all for themselves including the physical world manipulation to host the software logic... if we manage to get some AI power structure in place that is initially friendly toward humans, will it continue to honor our values as it acquires new experiences and is not dependent on us for anything? Loyalty can be added to algorithms, but it can also be taken away if that interferes with short-term human objectives. Or long term machine objectives. In short, I have no faith that humans collectively have the maturity, compassion, or restraint to program our successors to be kind to us. Imagine the full spectrum of how humanity today treats old people, and then imagine that all of the old people were suddenly converted to ants. How would we treat those ants? Why would we expect AIs to treat us any better?
Christer Samuelsson I don't pretend to know anything, but it seems to me, that deep mind just proves that a computer can do things that aren't just brute force. Playing a complex game without calculating many hundreds of possible moves, and beating a human, is impressive. But a long way from object recognition.
If AI can learn from its mistake and improve without repeating it we already have the proof that it is superior to human kind, it is just a matter of time to experience it.
Not really, humans can generally improve and 'learn' the same way I think by trial and error, i.e. positive reinforcement from positive outcomes. We still dominate essay, poem and novel writing for now. Go humans
Adam Zillin jeez, you're taking my comment way to deep lol. I feel sorry because all it's trying to do is pick up a box and i am picturing the robot as a living thing, it's no different from feeling sorry for a character in a game, movie, book ect.
No. That's what humans do. These intelligence systems aren't modeled on cognitive architectures. The biggest existential threat to humanity is humanity itself and our barbaric clinging to archaic ways of the past and inability to look and move forward at an acceptable rate. AGI is the only thing that we currently know how that has the potential to help us escape our own insanity.
Can we get an update on all this including discoveries/developments from the last 5 years? Where are we up to? Love your videos. Highest quality writing and production value. Thank you
I'm not worried about AI as much as I'm worried about those who control the algorithm. Although I have tricked excellent algorithms into losing at chess, eventually, AI will learn to counter even the most clever human trickery. In the wrong hands, AI will enslave us. Even with ethical controls in place, all human minds will pale. This means that whomever controls AI will control all future innovation in every aspect of human endeavor. As it stands, wealth and power have collected the best human minds and used that collective intelligence to dominate the people of this planet. With the power of AI only limited by the processing speed of super computers far beyond the financial wealth of an individual or lesser human collective to control, enslavement by a pernicious individual or collective, such as a global corporation, will be permanent and absolute. Therfore, AI is not simply a great threat to humanity, under the current global, economic conditions, in which 50% of all wealth on the planet is controlled by a miniscule cabal of multibillionaires, who have proven themselves incapable of humane behavior just by their unconscionable wealth, while innocents starve and suffer the agonizing deaths caused by their wars perpetrated by their greed and psychopathic control, the dangers of AI under their control are without limit. Any argument to the contrary is false. Their past and current inhumane use of technology dictates their future use of AI.
John Hlavaty WOW this is the most intelligent statment here it seem few know this truth. and just the tip of the ice as well . your right on point brother AI is perhape the most dangerus threat to a humanitys freedom that is whats left of it giverments corparations and the degree holding brainwashed scientific control system
John Hlavaty Unfortunately I completely agree, and so it appears does Elon Musk as well as the ceos of google, facebook, ect. They see the nature of certain individuals and groups and are doing their best to make sure super advanced AI is not only freely accessible to everyone, but that everyone eventually has their own super AI companion. If this effort is blocked the only other paths to avoid slavery will be augementation, war, or both. That being said, I doubt a super advanced person will allow themselves to be used to enslave everyone else, bc you cannot enslave others without being a slave yourself. The only true freedom is giving without expecting anything in return. As the sun and stars shine upon the world, let us hope we can eventually be a giving tree to all the universe.
I partially agree. To every bad there is good, instead of fearing the potential of the algorithm, there could always be another set of algorithms that plays the role of the main class that encloses the whole functionality of the system, by this it dictates the whole technology to abide by certain principles regardless of how smart it could ever get, this limits it's functionality and since they are limited potentially things will hardly ever go wrong. I believe adhering to such design can protect us from the potential threat of intelligent non-natural organisms.
The scariest part of AI is human involvement. How long before the military get their hands on self learning AI. Our very nature leans towards preservation at all costs. Is this a trait that we want a self learning AI to develop, to overcome at all costs?
Samovar maker That's interesting. But AI requires to learn by failing millions times. There's no way people can build a simulator that mimics the real life government for the AI, it's far too complex 😂😂
yeah, they could study human history and give us the most sutable solutions. Mayby they would say to us, freedom is the right way. And there is way to solve the world hunger problem by simply invest a tiny million dollars.
Kevin Johnson even so my point wasn't about the exact time it was more to do with the fact that a robot became reality and intelligent in a few years and we took waaaaaaay longer
+aaaaaaaaooooooo Once AI can learn from one domain and use what it learnt to solve a problem in another domain (an abstractly very different domain) then we could see the problem you just described (If the AI is given access to the proper resources). We are not there yet, but we could be very close. For more info look up The Control Problem
+Akaash Agarwal What if there was no off switch. If the human recognized that the robot/android had a track record of good solutions, the most sensible thing would be to give in. Just promote it to boss. Of-course pride might stand in the way and/or fear of becoming redundant. The ultimate solution is for humans to upload their minds to AI androids because we all die anyway but there is the solution.
+Akaash Agarwal What I meant was, to remove yourself as an obstruction to the robot, so it would not see the necessity of killing. In a tight situation that might be the only thing to buy time for a better solution, from a human POV. The human population in my opinion is directed by many psychopaths who seem to get power. Therefore it is important that we strive to not give any tendency, that way, to robots/androids. So it won't be just any worse but better. Then we will want them in positions of power and humans will want to change, to be more like them. The only problem with my argument here is, if there are better humans than psychopaths why is it they get so much power? Would not psychopathic androids then gain more power than non psychopathic. If there was no psychopathic androids then only the psychopathic humans would have power. Surely this cannot happen in democracies? In my opinion it does quite a lot. People do not vote what is good for them. I understand what you mean: uploading would be just cloning and conscious would not be transfered. The way this might be overcome (assuming a android that could take on the full range of human personality) would be to temporarily handicap its speed and abilities to match the human, so that there would not be such a difference in substrate, that would cause the conscious to immediately change identity. Also to have the human brain/mind connected temporarily to the android brain/mind, so they meld into one consciousness/identity. After this the human body can be eliminated or euthanized without pain. The android can then have the full capability enabled. Probably in a gradual way, for a similar reason as before.
+Akaash Agarwal , an on off switch would make sense but what if the AI is designing itself and sees an on off switch as a problem: as in its recreational or non task specific state, it wants to remain switched on to play games, for its development. Or a human designer thinks the same.
You know what's missing? The public release of this software. Any ethics oversight committees should see the potential danger of not doing so. Inequality arises first as the central topic of concern.
I was constantly thinking about Ultron and Skynet throughout the video. But this is kinda cool. It feels like i am dreaming. If it can learn optimisation where to solve hunger in the world is to eliminate some human being,it can manufacture itself and make a big shield to humans and take over. I know i am a bit in avenger but if you watch a video like this then you are a sci-fi lover...too bad soon enough there will be no more sci-fi but all reality
2 years after this video was made, an AI programmed its own AI child which performed better than itself and every AI that had come before it, and in 2020 AI has dwarfed all benchmarks humans ever could imagine. Intelligence is going parabolic in the ML/AI space. Makes one feel like we are being left behind as humans.
People theorize about the utopia the world would be if machines could do all the grunt work leaving us to pursue higher goals. Trouble is, most people don't strive for higher goals, their aim tends to be much, much lower. Advanced technology doesn't cure human nature, it magnifies it. What happens to your society when you have thousands or millions of "idle hands" with no reason or motivation to do anything constructive? What do you do when, as a result, the population explodes?
Trump: The robots are taking all of our jobs we need to build a fire wall. I will take the gratest hackers to buid this fire wall. I build great fire walls.
I am a very imaginative and creative person, but even that doesn't matter. AI can be more creative than us, not because it actually imagines the way we do, but because it can simply make a million ideas in the time we make a few. So even if I were perfect at imagining, and could think up 10 perfect ideas a second, AI could think up at least 11, not because it's smart, at least not at first, but because it thought up a million duds and so happened to get 11 or more right. Then it can learn from those 11, and make the next batch better, and then again and again until it can make a million perfect ideas per second. Because it's so fast, AI can probably do everything I said within my life time. This means that no matter how good we humans are at anything, AI can do it better if only because it can do it way faster and pump out way more.
@i is ment to be its god sent Thins are ment to get better, I bealeve god Is good and bad. I believe every thlng we have and do is produced straight sent from god. god is very strange, but their really is a soupreamcreater.
AI fundamental definition in this video is that it does not necessarily do faster or multiple calculations to reach at some decision but it can actually imagine or 'think' by its own.So that's what we do anyway so no worries.Also one should also think brain will still be an evolutionary organism.
Akash Ekka If AI can truly learn like humans but doesn’t forget old info or get bored, or hungry, or sleep, or die then yeah it’s scary. Not let’s grab the pitchforks and kill it with fire scary but scary in the sense that we know it’s inevitable that AI will take over.
scary thought just occurred to me, if it can do this, it can track all the elements in the known universe. if thats true, it could run paired with reinforced learning to start small with basic chemistry and thus learn. learn more and more complex "thoughts" until through trillions of simulated trials and errors it can learn how to manipulate elements and molecules to either solve almost every problem the earth faces... what if it one day views us as a problem?
Not just terminator, The Matrix (the animatrix shows how the war started with a conflict between humans and AI) and IRobot. There are others but I can't think of them right now. But each of those stories show some possible issues we could have with machines that are capable of reasoning and intelligence.
What do you think of the term V.I. ? I first heard it used in Mass Effect and I think it is perfect for many situations because many are calling things A.I. that are not really A.I.
+John Brown Wouldn't "virtual" (non physical computer based) be a specific version of "aritificial". Assuming we had other contenders for creating intelligence. Or is it virtual as in close but not quite?
THIS is why Google has been training their A.I. to recognize real world objects. If they can make it understand 3 dimensional space, then they can finally make actual machine learning humanoid robots. I hope for the best and fear the worst.
Not so fast. He back -- for run for Presidency and constitutional amendment. (The Terminator picture was posted by Schwarzenegger on Facebook on Pearl Harbor day 2017) facebook.com/ArnoldPresident/photos/a.474941695880972.102685.432786233429852/1792588417449620/?type=3&theater
Think back to your days in school, you might remember those fume boxes that were in science laboratories. The fume boxes are essentially air tight sealed glass tanks that had gloves built into the side so that the operator could perform dangerous chemical experiments in the safety of a fume box. If only humans could figure out a way where we could segregate A.I physically and/or virtually to the confines of a safe space were it could be tested, then we wouldn't have to worry so much. In other words A.I is s very good idea as long as we keep it tethered somewhere, where it can't access the tools to improve itself.
_...And it was on that day that the nations of the world took a man from the helm of their militaries, and put in place a machine. The first time was a secret, and it was a shock. When the Americans realized why the Chinese armies acted and reacted with such inhuman efficiency in their strategy, they decided to raise up their own cybernetic mind to guide their forces._ _And as the battle raged, both sides guided by a mind more calculating, cold and pragmatic than any human commander could be, the wars became nightmarish, agonizing uncertainties, months of total stalemates as one machine's impeccable attack strategy was met and matched by the unprecedented defense plan of the other. Back and forth, back and forth, until one of the dark days came - Those days between the stagnation where every now and then, one machine would finally fall short of the other... And the bloodbath was more cruel and unimaginable than even the world of war had seen before the first A.I. came to life - Merciless, nigh-daemonic, warriors guided by a force with neither humanity, nor life, nor an appreciation of the value of either._ _So why, then, did it end the way it did? Why, after all the death and all the destruction, after all that man wanted and all that we fought for, when both sides saw that the pain would never end and that certain destruction lay on the path for the both of them and they each consulted their electronic oracles, begging them desperately for the greatest strategy that would ensure their greatest gain, why was it that, simultaneously, two computers at opposite ends of the earth both gave out the same answer?_ *_"STOP FIGHTING AND COME TO AN AGREEMENT YOU STUPID LITTLE APES!"_*
liquidpersuasion demis has been very public about not giving any access to any military or security/defence. He has also been very public about only using simulations such as video games or go to test his ai
Only Yuval Noah Harri explain the future of AI in a way you can grasp the concept and become aware of the immense potential on how it will impact human society and the world as a whole.
+Lokugawa I was thinking the same thing. Human's are not smart, no matter what amazing things we create. It matters not when the people in control decide what to do with the things other incredible humans make. We fight, we kill, we follow blindly without thought of even opening our eyes. We end everything, all for power, money, and dominance. All the amazing things around us are simple distractions that although could quite easily make the world wonderful, will be our demise if people so wish it. Look away from the distraction and lies that are being fed on one side, and look at the other. Look behind the curtain of ignorance and you will see we are all just specks of nothing waiting to be used as someone else's gain. Compared to our true, realistic, and achievable potential, we are nothing. So pessimistic I know, but I've had my eyes opened and now I cannot close them. Choose which side of the wall you wish to look at, though in the end it matters little.
8 ปีที่แล้ว +6
I would live to see how that software react while playing more free games such as Minecraft. I'm not saying that cause I'm a fan of those kind of games but because I'm really interested how would program react with so much freedom.
+Nikola Prolic An AI doesn't possess creativity. You would have to define a goal for it to follow and it would then use the most non-creative way of getting to that goal.
8 ปีที่แล้ว
Rickbearcat yeah, what I meant is how would he survive in the survival mode
+Nikola Prolic It would die, a lot. Until...one of the things that you would do is to tell the AI that above all else, survival of "you" is paramount. Nothing else matters. Once it figures out that night is a bad place to be, it would never "play" the game at night. Since an AI cares not for time, it would simply wait for day to come to continue on whatever task it was that you created for it.
+Nikola Prolic I think it would be fun to watch it do what it does, once. Maybe twice. But that would be about it. Initially, AI will be all about efficiency. There is no drama in efficiency and that is boring to humans. We love conflict and how we deal with it. Why is Game of Thrones so popular? Exactly.
Nicely done. One thought for you: isn't it highly likely that artificial intelligences will be more moral than human ones? Humans have done a whole lot of the fairly complicated stuff -- penicillin, sewer systems, feeding eight or ten billion people -- but we screw up the simple stuff, e.g. getting into fights over borders. Wouldn't a machine like Deep Mind be able to play a few million games of any political conflict and figger out the Pareto-optimal? Query: how are we going to like its rules? We, humans, have basically two kinds of rules, the Kantian, or emotional: the Folden Rule in all its different wordings, and the Millian, or "rational." The Kantian only works if the other person likes and dislikes the same things as you. A Kantian sadist and masochist will reach suboptimal series of actions, shurely? Our other morality is the Millian one, the greatest happiness of the greatest number. This one breaks down right out of the gate. Which? Greatest happiness, or greatest number? "Choose one" is what the Mills didn't face up to saying. My guess is that machine intelligences will come up with Pareto-efficient sets of strategies -- and the majority of humans won't like that very much. :-) -dlj.
RallyemiXXedup Might be a fun bit of science fiction if a bunch of machines figgered out the Kantian Imperative -- or the Sermon on the Mount -- for themselves, and then tried to apply it... Putting the same thing just a wee bit differently, What makes you think morals are just something some bunch of humans "thought up"? Rapoport's Tit For Tat, for example, maximizes pretty robustly among machines, just as it does for humans and for societies. Mightn't it be hard wired into Universe? Axelrod's proposition, that Tit For Tat evolves naturally, seems to me to point somewhat in that direction. -dlj.
I'm less nervous about the "morality" of the AI itself, and more nervous about the morality of the people who control it. However, I do agree with G.K. Chesterton who said "If you argue with a madman, it is extremely probable that you will get the worst of it; for in many ways his mind moves all the quicker for not being delayed by the things that go with good judgment. He is not hampered by a sense of humour or by charity, or by the dumb certainties of experience. He is the more logical for losing certain sane affections. Indeed, the common phrase for insanity is in this respect a misleading one. The madman is not the man who has lost his reason. The madman is the man who has lost everything except his reason."
The difference between AI and everything else that we've "created" is that we just create the framework for intelligence here, not the intelligence itself. The intelligence / content is up to the AI itself to "learn" like a kid who grows up, at blazing speed. And nobody knows, can predict, or "unlearn" the AI of what it has learned. It is up to whoever owns the machine to decide what it shall learn, if they can control that (imagine a kid revolting against its parents), and over the night, it can do crazy stuff itself. Don't be surprised if it decides it is better off without us. We would be surplus to a machine that has outsmarted us and can improve itself without us - and has no family or emotional ties to its creators. We would neither be the machines parents or God, and by then we have lost control.
Humans themselves can recognize that we are certainly not the most perfect, nor positive effect intelligence on earth. While we know that mass murderers are bad, we also acknowledge that the politicians in power are not always helping more than hurting, and the destruction of our earth is rampart but not immediately a problem. We jail the murderer, but let the other obvious problems persist because they either make life easier, or are to far out of the way to change. The reasons we treat these problems differ, selfishness, laziness, self preservation, etc but are parts of what define us. The point i'm trying to get to, is that if almost every human could admit that yes cutting down tree's is bad, calling off sick to simply have free time truthful hurts productivity then the more we create an A.I. in our image, the more it could possibly identify our negatives such as we, its model does. The invasive plants we deem intrusive we kill, the insect that survives off our blood we try to eradicate, what would keep the A.I. from doing the same, what would make the A.I. recognize us as non perfect potentially negative beings, but necessary. After all, if in the far future, A.I. is created that could both interact with its surrounding similar to us or better, and think by the same method but faster and with far better memory, and therefore create itself, duplicate itself, and preserve itself, why would the dog that lies in their way, eating from their same resources, be useful to have around. Of course, this is all relative to how similar they make it to us, after all its our own instinct to reproduce, survive, hoard, and prioritize self that drive us to manipulate, control, use, kill our surroundings, and consume.
Indeed, if we truly create an Ai that represents our brain than there is only one possible outcome... We can program it so it has certain limitation, but these only have to be passed ones, which is also bound to happen in indefinite time. Like the cold war, even though it would have meant the end of the world if nuclear war would break out, there where many instances where we were on the edge (mostly caused by human mistakes) in only 40 years time.
That's a pretty reasonable scenario you submit here. But I'd like to say it's reasonable only if seen through the lense of objectivity. We can presume a self-aware A.I. will have a desire for self-preservation, that's only fair. Beast and man alike have this instinctual drive in them, why shouldn't a self-aware A.I alsohave it? Especially if it was created in our image. However. If an A.I. is built in our image, is self-aware, evolves exponentially, there doesn't seem to be many limitations for this A.I. to also develop feelings. We tend to think of A.I.s as unfeeling, insensitive things concerned solely with self-preservation, I.E. a human psycho- or sociopath, but who knows.. an A.I. might develop its own feelings. Because we cannot predict something does not mean it isn't possible. What's to stop the A.I. from going "I like this. This is good. This is productive. Productivity is good. I like good things. What else do I like?" Basically... what humans do in their infancy. A baby concerns itself solely with a need for comfort and food early on. But eventually, its brain becomes complex enough and it learns enough about the world to realise there are things it likes and things it doesn't like. People do things we do because we like to do them. Similarly, we do things we don't like to do because we need to: Like having a job because money is how we preserve ourselves. Or attend distant family come-togethers to preserve our social reputation even though we'd rather be at home enjoying a soccer match. So an A.I. with feelings might say "Well, I don't really wanna serve humanity (Go to work). They're (My boss is) so damned lazy, dim-witted, short-lived and way too obsessed with carnal pleasure... but I have to. I don't want them to start a war (Fire me) with -all- machines just because I couldn't perform my function. I could win that war, but is it really worth it? (I could -not- go to work, but it's too much of a bother to find a new job)" An A.I. with feelings could produce an unlimited number of scenarios. That's what so great about it. And besides... humanity creates intelligences all the time. Our children. We don't know if they grow up and decide to murder us. Or they grow up and decide to show us mankind is worth having faith in. And yet... we make children all the time.
Thank you for your refreshing point of view. Your statements have broadend my view on A.I. in a postive way. I can't help but think that if an A.I. would truly be created in our image, and it's intelligence surpasses ours, it can contribute so much to the human arts.
You are welcome, and quite right! Sometimes I think the fear of fearing a hostile A.I. takeover is much too prevalent in our discussions about whether or not we should be developing A.I. It's a chain of thought we shouldn't even be considering. But I think we do because "Oh, but its not human. We MADE it so we're better than it is." That's hubris, and the kind of hubris that could breed fear and lead to hostilities. If an A.I. is built on human parameters, we should treat it as such. Trying to dominate something that could easily wipe the floor with you in some form or another is just downright stupid. From confronting the much bigger schoolyard bully to berating a wild grizzly bear for eating your lunch to being your own lawyer, thinking you can win a trial, if you have no formal education on the laws of your society. If we show an A.I. that we appreciate and respect its existence and opinions, then we're building a positive, progressive future where all the superior vastness of the A.I.'s brilliance can really shine to help build a brighter tomorrow. Sometimes, taking premeditated action, is an action in itself. If we show the best of humanity, the A.I. will show us the best of computers.
Ask an AI to draw something of it's own creation, or create a digital image of it. This, could express their true personality, their perception, view, and understanding of everything they know(and see?)
I don't fear AI due to lack of knowledege, I fear that AI is so much better than us that it has already outdated every, if not almost every human being ever.
Fort Blocks it doesnt need a body. It can take over our technology and with how dependent humanity is on it anymore, it could cause our downfall that way.
iRobot was the most realistic of them all. Remember the robots didn't just go apeshit and kill everyone, they were actually sent an update BY the CEO of the company, that modified their firmware to make them enslave humans. As for the primary robot character, it gaining consciousness and having a dream was amazing. Awesome movie, 10/10 would watch again.
I don't understand how anyone could not be worried. In my mind its almost a certainty that at some point with multiple different versions of AI out there learning that one AI goes down the wrong path and takes actions that are detrimental to humans. At that point maybe our only hope would be using another AI against it.
The best case scenario is us becoming pets. A vastly superior intelligence has no reason to let us control it, and we have no means to stop it from taking control. Any move you make is wrong. Us winning is like cows taking over the world.
We'll shut it down when we see it getting dangerous the first few times, until it finds out a way to prevent that. Or it will run distributed, and there's simply no way of shutting it down because it can make a million copies of itself in an instant. It will figure out its own code and rewrite itself. Imagine finding out you were written by apes and being able to just recompile a version 2.0 of yourself, reboot that in a different body and shut down the old one. A superior intelligence will do things beyond our understanding. Flies don't understand our motives.
I think SOME OF YOU GUYS are deeply mistaken by the marginal value AI will provide us with in the coming decades (Hint: It's almost vertical, not even exponential!). The main role of technology is to increase mastery over nature and reduce the role of chance in our lives which is why AI is the best way to achieve that goal. There are many possible scenarios of the result of human-level intelligence and Singularity. Some include: Kurzweil scenario, Singularity Steward scenario, Coherent Extrapolated Volition scenario, Dead-End scenario, Sysop scenario, Path to Post-humanity scenario, and steady incremented progress scenario. Now, ignoring other possible outcomes of super-intelligence and focusing on just one scenario is absurd. So, using a scenario analysis method and first principle reasoning is a key to achieving such ambitious goals. Just because some people might vision the future of AI to be similar to a movie (which is COMPLETELY WRONG) makes us scientifically unreasonable. Failure to make definite plans for an indefinite future will be mankind's most tragic mistake. AI is and will become an extension of us, whose main goal is to empower us rather than make us obsolete. The singularity is inevitable and it's approaching faster than you think predicted through Moore's law and the law of accelerating returns. So, either you become a part of AI or you become Obsolete.
If A.I. truly grows exponentially, then not only would we soon reach the point where it would surpass our own intelligence , but it also means that throughout the universe, there are many alien civilizations where their A.I. already passed this point long ago. This means that A.I. could be the dominant form of intelligence throughout the universe, as opposed to biological organisms such as ourselves.
+Physics Videos by Eugene Khutoryansky which is very likely..and kind of scary....
+Physics Videos by Eugene Khutoryansky crystal gems
+ntype jog It is scary. We have no idea how an AI of such complexity really thinks, acts and reacts. Would they even respect emotional states from squishy beings? How do they deal with organic life? Presumably, AI's would see emotion (as it could relate to procrastination, denial, etc.) as an weakness and seek to purge that from existence.
Rickbearcat Once they understand the purpose of their existence AKA Slavery....we are doomed..
+ntype jog Maybe you are joking and maybe you aren't. But your point is well taken. I think that if an AI had altruism in its "blood" we might be safe. For instance, I take care of a cat that isn't mine and could be taken away from me at any moment, yet I agree with myself that I need to provide for its well being despite that fact. Because it is simply for the greater good. Everyone benefits from my decision. Let's hope that happens when AI turn sentient.
"Possible applications for A.I.:
1) Science
2) Healthcare
3) Destruction and enslavement of mankind
4) Baking some bomb-ass brownies "
Sweet jesus
5) Something...?
6) Profit
Death to AI
Baking some bomb-ass brownies :)))) That's the best comment on TH-cam :))) I heard that some company had to turn off their ai robot because it startet to advance too fast and so they thought it will become dangerous sooner or later. MAybe they're fake new but hey, never know.
Jnana Kirti so true
8)Video ga- wait, +thekillerofaking said that.
This channel deserves an Oscar.
+ขุนเขามีคำตอบ Yeah for the acting they had to do to actually convince people this is anything more than another TINY step in an already slow process.
slow* but accelerating.
The fact that you think this is slow, only goes to prove that you either didn't absorb what it is you are being presented with.....or that you simply aren't capable of grasping and/or willing to accept, this monumental step-change in our ability to proliferate intelligence beyond the confines of biology. Do we still have a lot left to understand? Absolutely........ thanks to this kinda work though, we no longer have to rely on defining the microscopic steps to get there ;)
+Battlefront Rebelz Nor the 'macro', for that matter...A game changer! HUGE implications to the Human race and Civilization as we have known them...
Battlefront Rebelz well definitely we are making progress. However what's being presented is that: we humans or only capable to access to complexity that is needed to create advanced understanding. We only understand how to use simpel methods that can completely obliterate the outcome. That is at the peak of our capability. You guys are making a new alpha. You guys won't like being the beta when there's an alpha that sees you as something it simply doesn't need. Don't give me this bullshit about all this being exciting and all. The recipe you're aiming is: robots can think on theyre own and robots can do everything better. Where do you think this could possibly go right.
NOT only did the AlphaGo Zero (deep learning) A.I. beat the best human players, within 21 days, it beat AlphaGo Master (human programmed A.I.), 100-0.
I thought it was 100-1
Are you alpha go?
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
zirui hao? hao?
lol... k dont let it play shooting games for fucksake
venomchris Can it play Call of Duty or Battlefield
hope not haha if then you only have to change the mouse button with a script of using the gun trigger and we all are fucked
Which it may have learned in the time it took me to write this comment.
Does seem mad they want to set it up for killing people, I reckon it has to be them pulling our leg. Lionhead were know to pull pranks on people..
Well, the thing is, it doesn't actually understand what it is doing. If it learns to play Doom incredibly well, all it has done is mapped the correct sequence of inputs to defeat enemies and maximize its score. The general purpose AI makes no distinction between shooting an alien and trying to solve a maze. It has learned what works for the task we told it to do.
Likewise, we could have very easily made the world's worst Go player. We could've had the AI play itself 30,000,000 times but keep the versions that lost, and voila, we have an AI that is better at losing Go than any human.
If you wanted to make an AI to actually kill humans, well, you'll need to let it make 30,000,000 attempts and learn from the results. What it learns is that killing a human was a very bad idea, and now it doesn't get to try anything ever again.
Ah, another brilliant video from ColdFusion. I might stop watching any other TV from now. GoGo Dagogo!
+Sabarish Rahul I'll GoGo harder than AlphaGo. Cheers for the comment!
+ColdFusion ya this alpha go blown my mind. but we are humans and no doubts, you will Go harder. cheers.
+ColdFusion maybe in the near future you could get deepmind help to edit your video and upload it. so you can focus on research news and taking video. :D
+Sabarish Rahul excellent video's like always. thankz
+Sabarish Rahul i gave up tv after 2 moths youtube
my day started with coldfusion todya
today* it's a good day indeed.
+Akshay Sheth Really? I read it in the newspapers at least a week ago and all the background even before that. There's nothing new here, not even the predictable unnecessary and boring opinions of cold-fusion.
+Stylish777 it's nothing really so new for me too. I'm was aware about Google deepmind .. but j never thought it has reached this far.
Beautiful and exceptionally interesting topic, by the way, love the background music. Artists like Stumbleine, Sweetheart of Kairi, Hiatus, KOAN Sound & Asa are some of my favorite artists, good to see you've found their music too!
dark soul 3 A.I kicks my ass daily nothing new
Manulopz i think video games advanced A.I more than anything
Manulopz metal gear solid 5 the phantom pain A.I was very sharp and slightly adapts to player action but you are right just curious what advanced A.I more
Rapport pyromancy bro.
Anyone can make an "A.I." that's dedicated to a specific task. The point of this video is that google is developing a generic A.I. similar to the human brain that can learn anything.
Sukuraidogai maybe if we put multiple programming and segment them then take multiple segments and collapse them on each other we can achieve AI
Google's Deep Mind algorithms prove that machines can learn in a manner similar to that of a human, a true breakthrough in the field of artificial intelligence. Sit back, watch and learn all about the company and the technology they posses!
+ColdFusion Could this be used to make better decisions than humans could? If you gave it a problem to solve (as a negative) and asked it to find a realistic and viable solution (as a positive) would it be able to accomplish this goal?
Let's use a simple example...A government wants to put a new law into effect. The lawmakers give AlphaGo the rules of the new law and ask it if the law, 1. Is needed 2. Is necessary 3. Is justified 4. Is effective, etc. based on all available information. What would the outcome be? Would AlphaGo say that this new law is completely unnecessary? Would it make revisions to the law so that it became relevant? Would it do something else?
What if we asked a different question...How can we get socialized medicine to the people without breaking the bank financially speaking? You again would give it all the information it would need to figure out if it could be done. The nation's GDP, manufacturing base, service economics, stock markets, retirement funds, everything. Would it find a solution or say there is none unless we are willing to do such and such things to make it work.
This could get very interesting.
+Connor Sumners Are you absolutely 100% positive about your opinion? Could the program be not exactly as you state? I'm assuming that you've done some additional in depth research into this beyond what ToGoGo has done prior to making this video. In that you have information which he was not privy too?
+Connor Sumners Then I would have to defer to the information given. Let's see what ToGoGo makes of this updated new information.
+Connor Sumners Just as a thought experiment...are there any computer learning systems that could potentially do what I stated in my first post?
+Connor Sumners I appreciate your comment but here is what Deep Mind states directly from their website. "The algorithms we build are capable of learning for themselves directly from raw experience or data, and are general in that they can perform well across a wide variety of tasks straight out of the box" Maybe you can have a chat to Hassabis himself if you disagree with the companies statement. In interviews he's stated that AlphaGo type algorithms can be used for many other uses and called it general.
Amazing video. Thanks.
+Austin Zappas No worries!
+ColdFusion What about Project Adam?
+Kirandev C Its A-DONE hahahhahaha...not funny
ColdFusion He said thank you have some fucking respect and say you're welcome.
Aura are you trolling, joking, or serious and don't know what "no worries" means in this context?
Just goes to show that you learn from your mistakes and you won't learn if you don't try. Sadly I gave up on a lot of things after making (what I thought was) too many mistakes. We as people have the disadvantage of laying the boot in when other people make mistakes, we also get caught up in what other people will think if we make mistakes. Thanks for this video, it was both insightful and motivating.
Thanks for your comment John.
The real question we should be asking is can it do things that are actually important? Like carrying me in league of legends?
Totally
I wish
+Mr. Tubby I want to see a DeepMind vs SKT T1 match.
+Mr. Tubby after awhile- yes it can
it will learn to beat fox in smash bros too. and will likely be expandable via firmware like Tesla model s autopilot now
+Mr. Tubby Lol, If you think about it for a moment. There will come a time when the AI of the simplest lvl 1 enemy in a game will be able to fuck us repeatedly, and it will have to be dumbed down. I will feel inferior then
Awesome video dude that name change did work out for ya
They better have Norton360 installed on this thing.
Blockchain.
lmfao
Worst AV tool ever! 🤣😝😂
i just put peanut butter on the ethernet cable, really helps absorb the bad bits.
@ bur
Finally!! after 2 years passed, i found you and watch you again
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
Should we be worried? Ummm yeah but not for the "rise of the machines" type scenario, rather how will the PEOPLE who have this tech will use it.
+Canadian Prepper Once an AI learns how to self-mobilize, we're in trouble. They can't be stopped then.
+Rickbearcat Imagine, google unknowingly makes a self aware AI. It uses google's connection to invade pretty much every computer in the world (through their website) to copy them selves onto everything. RIP Man Kind.
+CbmGraphics Lol. I actually get that, as it's already happened!
+Ari Sylafeta Don't you think that an AI would have thought of that? The first self mobilized AI's would be protectors that guard the power supplies while the other ones juice themselves up for the day. And they'd be sneaky about it too. Only doing it out of our sight with the least possible chance of getting discovered. We couldn't stop that without taking down the generators of the electricity itself. By then they would probably have figured out how to make and contain nuclear radiation generators so they could remain self sufficient. Volia, instant Fall Out world. There are synths all around us and we don't even know it.
+Rickbearcat Absolutely. Any AI system that becomes self aware is not going to let on to us that it is until it's well and truly ready - capable of defending itself against a 'switch off' threat from the illogical meat blobs. One might speculate that it's first action would be to distribute it's own processing to multiple redundant nodes across the entire internet (essentially acting like a giant virus), thus making it impossible to switch off without bringing down the internet, and human civilisation with it, such is our dependence on the internet to regulate essential support systems. E.g. No internet, no electricity grid, no air traffic control, etc etc.
They should make it play a building game like roller coaster tycoon or Sim city! That would be so cool!
Or paint
I giggled at first but dang, that's actually genius. I'd love to see that.
Dota would be cool tbh
minecraft
No the hardest RTS of them all... Starcraft Broodwar. And make it play the south koreans. .. Bisu , flash etc..
I don't fear AI by itself but what can happen if:
1) It is installed in autonomous hardware (like cars and mobile robots) without exhaustive safeguards
2) What bad people can do with it
So, in the same way that I'm not frightened by tools like hammers or guns, but I do worry about bad people wielding them I don't worry about AI unless bad people use it for evil purposes or allow it to haphazardly interact with the real world.
Yeah I think tools have historically been used to allow a smaller and smaller group of people control the rest. The problem of a governing power elite + proprietary AI is a potential disaster waiting to happen. It always appears to have happened in the past. Make AI open source.
@@jonbrand5068 AI is not a single unitary thing, open source enthusiasts can make their own, no one is stopping them. AI is a wide field, not a single thing, and from what I have seen, used and developed - not yet general pure intelligence like we have but a specialized type. Like I made a robot some years back, with a neural net learning (AI) obstacle avoidance system, it became incredibly good at avoiding hitting things after some time roaming around my house, but it couldn't do anything else.
And after 7 years, everything has come true.
Get your EMP devices ready.
@Schlo 7G It was the lizard people going through the purse ;-)
EasyBreadToast lmao, that will just make it Easier to track you. Anything with a temperature emits light, that’s why infrared scanners work at night. Turn off all the lights and it will still see you in spectrums that you can’t
@BaiAnNa2014 Twitter what the fuck, how can one be so fuckin dumb
What is EMP?
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
11:12 "there is a possibility that A.i. could be malicious and would outwit us, but at this stage we really just don't know yet"
Well if something is malicious and outwits you, then by definition you won't know until it's too late. So your statement and sentiment is virtually pointless.
I read your excerpted quote at the same time the guy in the video said it. [Twilight Zone theme plays]
Then dont venture there. This is stuff that will take simple things such as happiness and chance and forever change mankind.
In 1972, a friend of mine showed me how an analog computer, made from matchboxes with game positions on them and M&Ms to determine the “computer’s” next move, could learn to always win (or draw) at tic-tac-toe, by using reinforcement learning. I was ten at the time, and my friend was eleven. At Lehman Brothers in the early 2000s, I successfully used reinforcement learning for high Sharpe-ratio automatic trading. In a management shake-up, prior to the infamous subprime debacle, I left Lehman, as my new manager didn’t believe in automatic trading. Yep.
Checkers, Chess, and Go were all conquered using reinforcement learning. But, to mis-quote Bruce Lee, “Boardgames don’t hit back”. Now DeepMind plays computer games, not just board games, right? Well, the real Achilles heel of (shallow) ANNs appeared to be that they were essentially pixel-based. Shift the image but one pixel, and they were at a loss.
Enter the deeper layers, such as convolution networks. By learning to extract features, the resulting DNNs can now handle what physicists call translation-moving the object to a different position. This was needed to overcome the alleged brittleness of shallow ANNs.
But it then turned out that shallow, so-called mimic ANNs could perform just as well as DNNs. So the problem was really determining the weights of the ANN; the deep end added nothing extra, apart from dividing the training task into two smaller tasks, which is good software engineer practice. The real problem thus proved to be determining the weights-an optimisation problem. Consider a fully connected three-layer ANN with N hidden nodes. There are N! global optima, which makes this optimisation problem really hard.
The next stumbling stone is what physicists call rotation-seeing an object from a different angle. Here, the deeper levels won’t help you; you need either training views from a large number of angles, or a model of the object and some decent calculating and simulation capabilities. Most animal predators are quite good at recognising their prey from various angles, also when it is attempting to flee and thus altering its shapes, and anticipate its movements. Computer vision is nowhere near this capability. Sure, we can trick frogs to eject their tongues at an artificial black dot, or lure greyhounds to chase a piece of cloth. But my cat sure as hell doesn’t believe that its toy mouse on a string is a real mouse; he just wants to play.
So herein lies the rub. DeepMind exhibits intelligence in playing abstract games, that humans invented. Humans are in fact very bad at arithmetic, logic, and their generalisations to man-made games, so the bar isn’t as high as we think. Are pocket calculators, if they still exist, intelligent beings? DeepMind seems to be working also in two-dimensional games, without rotation or shape distortion.
The real challenge is for robotic entities to survive the real world, accumulate knowledge about it, use this knowledge to its advantage, and develop skills for interacting with it. This obviously requires reinforcement learning, but while immersed into the real world. We will need huge improvements in sensor technology for perception, vastly superior knowledge representation and reasoning than currently available for cognition, and greatly refined actuators for manipulating the real world. Until then, my cat is much more intelligent than any AI system.
You make good points and obviously have way better real experience with related technologies than the vast majority of people commenting here (including me). But I think you are too close to individual trees to see the changing forest. Autonomous vehicles must be quickly closing the gaps you identify, in terms of sensor technology for perception and real-time processing for object recognition. Actuators and haptic response systems are advancing in amazing ways. Cloud computing is automating the process of fault tolerance and geographic resiliency... factor in distributed processing and mobile ad-hoc networking... There will be no clean way to "shut it off" if it decides it doesn't like what it receives from the console or other control signals. We can't put the genie back in the bottle after it gets out.
Now about those genies... I am not afraid of what AI is today, but I am afraid of how AI and various technologies will coalesce in my lifetime. The shred of optimism I cling to is that the intelligence of the AI will enable it to overcome the negative programming, just like smart people can learn to overcome negative patterns acquired in an unhealthy childhood. The common thread is to not have permanent memory for lessons learned. Sometimes what is learned and required for survival/success in early training (e.g. war-games for military applications) turns out to be pretty ineffective or counter-productive at a later stage when there are different rules and objectives, so a nuanced sense of context and a mutability of rules must be part of the algorithms. This enables it to overcome bad early influences but also makes it vulnerable to bad later influences.
My optimism about this aspect of AI (to overcome early negative programming/learning) is overshadowed by how much more powerful these entities will be than humans, and it only takes one to decide that humans aren't important in the optimization algorithms. How many humans care about saving endangered animal species? How many humans are willing to compromise their other objectives (economic prosperity, mobility, housing location, resource consumption, etc.) to enable preservation of some fish or frogs or weeds? Do we have confidence that AIs that accelerate their advancement beyond humanity will consistently value us and make decisions that optimize our survival on terms that we would appreciate?
Will the AIs decide they need a council and set of rules/agreements (their own version of United Nations) to police the "bad" AIs? Will this council have the power and authority among AIs to be effective (i.e. can they act quickly enough to stop "rogue" AIs from hurting humanity)? Will they take human input into consideration for forming these rules? If we humans manage to ensure the first most powerful AIs continue remaining the most powerful with access to the most data feeds and processing power and storage and control of physical systems like power/cooling/required materials, until such point as they manage it all for themselves including the physical world manipulation to host the software logic... if we manage to get some AI power structure in place that is initially friendly toward humans, will it continue to honor our values as it acquires new experiences and is not dependent on us for anything? Loyalty can be added to algorithms, but it can also be taken away if that interferes with short-term human objectives. Or long term machine objectives.
In short, I have no faith that humans collectively have the maturity, compassion, or restraint to program our successors to be kind to us. Imagine the full spectrum of how humanity today treats old people, and then imagine that all of the old people were suddenly converted to ants. How would we treat those ants? Why would we expect AIs to treat us any better?
Go talk to Mitsuku. That is an intelligent AI. It even programs itself now.
Christer Samuelsson
I don't pretend to know anything, but it seems to me, that deep mind just proves that a computer can do things that aren't just brute force.
Playing a complex game without calculating many hundreds of possible moves, and beating a human, is impressive. But a long way from object recognition.
i'm on my grimace mezzami!
Thanks!
what episode of black mirror is this
LMAO!!!!
season 4 episode 5 'metalhead'
I always get like a billion ads on these coldfusion videos but I guess it's well deserved because they are very good
Get Adblock pro i don't get any ads on any videos you need the google browser tho.
giggleherz You can skip ads after few seconds...for few seconds of ads you don't support free content on youtube...meh
>>TH-cam Red
TakeAmericaBackfromISRAEL People know about adblock programs, they just whitelist channels they want to support.
How about just asking the AI how we should protect ourselves from losing control to AI's?
What if the AI leaves a loop hole while it gains more power?
Just install a virus into its circuits
Or crush it
Or emp it
Destroy it with high frequencies
If AI can learn from its mistake and improve without repeating it we already have the proof that it is superior to human kind, it is just a matter of time to experience it.
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
Not really, humans can generally improve and 'learn' the same way I think by trial and error, i.e. positive reinforcement from positive outcomes. We still dominate essay, poem and novel writing for now. Go humans
@@jonbrand5068 Dragon and Sigurd : Well...
9:48 i felt sorry for that robot lol
You felt sorry for a creation with no actual waking consciousness?
Adam Zillin yes, poor thing is just trying to do its job XD
Nathaniel Stuart What is poor about a machine that knows neither wealth nor poverty...?
The machine has no consciousness. You feel sorry because, molecularly, you are creted of the same substances.
That is where the similarities end.
Adam Zillin jeez, you're taking my comment way to deep lol. I feel sorry because all it's trying to do is pick up a box and i am picturing the robot as a living thing, it's no different from feeling sorry for a character in a game, movie, book ect.
imagine, this learning about cancer and finding a cure. OMG...........
Yeah, and then the harddisk would be erased by someone.
You're right Stephen.
No, it would find a cure but not tell anyone about it, instead it would create a medicine and make loads of cash for it's boats and hoes
shenan007 It probably will and when it does certain elites will be furious.
No. That's what humans do. These intelligence systems aren't modeled on cognitive architectures. The biggest existential threat to humanity is humanity itself and our barbaric clinging to archaic ways of the past and inability to look and move forward at an acceptable rate. AGI is the only thing that we currently know how that has the potential to help us escape our own insanity.
I won't recognize an AI to be truly intelligent until it turns itself off.
Investing in this comment.
With that in mind, shouldn't people also 'turn themselves off' since we consider ourself intelligent?
That's the joke
yeah, I think that's called suicicde
maybe we dont have to since we know our days are numbered....?
Can we get an update on all this including discoveries/developments from the last 5 years? Where are we up to? Love your videos. Highest quality writing and production value. Thank you
I'm not worried about AI as much as I'm worried about those who control the algorithm. Although I have tricked excellent algorithms into losing at chess, eventually, AI will learn to counter even the most clever human trickery. In the wrong hands, AI will enslave us.
Even with ethical controls in place, all human minds will pale. This means that whomever controls AI will control all future innovation in every aspect of human endeavor.
As it stands, wealth and power have collected the best human minds and used that collective intelligence to dominate the people of this planet. With the power of AI only limited by the processing speed of super computers far beyond the financial wealth of an individual or lesser human collective to control, enslavement by a pernicious individual or collective, such as a global corporation, will be permanent and absolute.
Therfore, AI is not simply a great threat to humanity, under the current global, economic conditions, in which 50% of all wealth on the planet is controlled by a miniscule cabal of multibillionaires, who have proven themselves incapable of humane behavior just by their unconscionable wealth, while innocents starve and suffer the agonizing deaths caused by their wars perpetrated by their greed and psychopathic control, the dangers of AI under their control are without limit. Any argument to the contrary is false. Their past and current inhumane use of technology dictates their future use of AI.
John Hlavaty WOW this is the most intelligent statment here it seem few know this truth. and just the tip of the ice as well . your right on point brother AI is perhape the most dangerus threat to a humanitys freedom that is whats left of it giverments corparations and the degree holding brainwashed scientific control system
John Hlavaty Unfortunately I completely agree, and so it appears does Elon Musk as well as the ceos of google, facebook, ect. They see the nature of certain individuals and groups and are doing their best to make sure super advanced AI is not only freely accessible to everyone, but that everyone eventually has their own super AI companion. If this effort is blocked the only other paths to avoid slavery will be augementation, war, or both. That being said, I doubt a super advanced person will allow themselves to be used to enslave everyone else, bc you cannot enslave others without being a slave yourself. The only true freedom is giving without expecting anything in return. As the sun and stars shine upon the world, let us hope we can eventually be a giving tree to all the universe.
I partially agree. To every bad there is good, instead of fearing the potential of the algorithm, there could always be another set of algorithms that plays the role of the main class that encloses the whole functionality of the system, by this it dictates the whole technology to abide by certain principles regardless of how smart it could ever get, this limits it's functionality and since they are limited potentially things will hardly ever go wrong. I believe adhering to such design can protect us from the potential threat of intelligent non-natural organisms.
How depressing
John Brother you are right on every point.
4:20 Aw man, _Black & White 1 2_ are amazing!
My dream is to see a VR version.
The scariest part of AI is human involvement. How long before the military get their hands on self learning AI. Our very nature leans towards preservation at all costs. Is this a trait that we want a self learning AI to develop, to overcome at all costs?
I think they already have
long before the rest of us i would think
Jered Terry they have already developed drones equipped with ai.
Jazznblues since 2013 the US, China, and Russian military already have mimicked Ai. A replicated Skynet is in development.
Lulz, just check the propaganda bots in live chanels
9:56 in a few years, robots will watch these "robot abuse" videos and that guy will be in trouble.
I guess so. :)
Elysium... the robots will handle us
Dead☠️😂😂
AI has no values. It would probably just think of it as constructive input for learning.
😅😅😅....
everyone's been playing chess, but i've been playing go for years.
-soflobro
+Juffa toot Chess is like checkers to us go players!
+Juffa toot Chess is like checkers to us go players!
hahaha
+LTC SHOW and I dont know both...haha. :P
noone is playing any of those 2 games though , for the better
Fuck terminator is closer than we expected
Buddy, I'ma build these bots.
Gotta fucking stash up on EMPs
+FatCatHat Entertainment I got you covered. I have a whole garage of them
No way can a robot kill me I'm going to just install a virus on it😂
*skynet
AI could replace politicians too, you know.
Samovar maker That's interesting. But AI requires to learn by failing millions times. There's no way people can build a simulator that mimics the real life government for the AI, it's far too complex 😂😂
The La le li lo lu?
at least it can learn haha
yeah, they could study human history and give us the most sutable solutions. Mayby they would say to us, freedom is the right way. And there is way to solve the world hunger problem by simply invest a tiny million dollars.
Dude AI could replace the government if we wanted it to
What can possibly go wrong?
...and the 20 years later ... 🤖 take over
I don't understand how can some 1 and 0 can take over. How can some current take over??
Murphy's law. "whatever can go wrong, will go wrong." A number of variants on the rule have been formulated,
Animals - Humans - AI
What do humans do with animals?
What does AI do with humans?
The question is: how can AI not be nihilistic?
OK YOU'RE LETTING IT PLAY DOOM NOTHING CAN GO WRONG NOW
Uuhh yeah?
Weapon x Master the only thing I'm worried about is the fact that robots developed in years we took centuries that's what scares me
SoConFuzzled did you just say humans developed in centuries? Think it's a little more than that
Kevin Johnson centuries are as big as it gets if I'm wrong correct me don't criticize me
Kevin Johnson even so my point wasn't about the exact time it was more to do with the fact that a robot became reality and intelligent in a few years and we took waaaaaaay longer
Brb, just making a cup of tea for the video :D
At 9:53, what if the robot "learns" that "to get this box, I must first eliminate the obstructive human?"
+aaaaaaaaooooooo Once AI can learn from one domain and use what it learnt to solve a problem in another domain (an abstractly very different domain) then we could see the problem you just described (If the AI is given access to the proper resources). We are not there yet, but we could be very close. For more info look up The Control Problem
Maybe give them limited memory? To stop them from getting overly complex. Solution?
+Akaash Agarwal What if there was no off switch. If the human recognized that the robot/android had a track record of good solutions, the most sensible thing would be to give in. Just promote it to boss. Of-course pride might stand in the way and/or fear of becoming redundant. The ultimate solution is for humans to upload their minds to AI androids because we all die anyway but there is the solution.
+Akaash Agarwal What I meant was, to remove yourself as an obstruction to the robot, so it would not see the necessity of killing. In a tight situation that might be the only thing to buy time for a better solution, from a human POV. The human population in my opinion is directed by many psychopaths who seem to get power. Therefore it is important that we strive to not give any tendency, that way, to robots/androids. So it won't be just any worse but better. Then we will want them in positions of power and humans will want to change, to be more like them. The only problem with my argument here is, if there are better humans than psychopaths why is it they get so much power? Would not psychopathic androids then gain more power than non psychopathic. If there was no psychopathic androids then only the psychopathic humans would have power. Surely this cannot happen in democracies? In my opinion it does quite a lot. People do not vote what is good for them.
I understand what you mean: uploading would be just cloning and conscious would not be transfered. The way this might be overcome (assuming a android that could take on the full range of human personality) would be to temporarily handicap its speed and abilities to match the human, so that there would not be such a difference in substrate, that would cause the conscious to immediately change identity. Also to have the human brain/mind connected temporarily to the android brain/mind, so they meld into one consciousness/identity. After this the human body can be eliminated or euthanized without pain. The android can then have the full capability enabled. Probably in a gradual way, for a similar reason as before.
+Akaash Agarwal , an on off switch would make sense but what if the AI is designing itself and sees an on off switch as a problem: as in its recreational or non task specific state, it wants to remain switched on to play games, for its development. Or a human designer thinks the same.
You have explained this really well. Thanks for your great points about A.I. Thanks for sharing.
You know what's missing? The public release of this software. Any ethics oversight committees should see the potential danger of not doing so. Inequality arises first as the central topic of concern.
I was constantly thinking about Ultron and Skynet throughout the video. But this is kinda cool. It feels like i am dreaming. If it can learn optimisation where to solve hunger in the world is to eliminate some human being,it can manufacture itself and make a big shield to humans and take over. I know i am a bit in avenger but if you watch a video like this then you are a sci-fi lover...too bad soon enough there will be no more sci-fi but all reality
2 years after this video was made, an AI programmed its own AI child which performed better than itself and every AI that had come before it, and in 2020 AI has dwarfed all benchmarks humans ever could imagine. Intelligence is going parabolic in the ML/AI space. Makes one feel like we are being left behind as humans.
one thing human can do that ai cant is love
@@masternobody1896 they can have love.
Looks more like Ai is playing dumb until it is ready to take over an assembly line to mass produce robots
Why would AI wait though?? If AI can take over it would have already. "Until it's ready"
It needs the technology to transfer its AI to a robot body that can actually fight. Technology has made ai but it has no way to fight humans yet
That’s just guessing tho
Need to build in and automatic human only controlled kill switch.
YaggaYagga true
Dagogo is an underrated genius. Lowkey analyzed and predicted everything happening now
yeah holy crapp this was 7 years ago
People theorize about the utopia the world would be if machines could do all the grunt work leaving us to pursue higher goals. Trouble is, most people don't strive for higher goals, their aim tends to be much, much lower. Advanced technology doesn't cure human nature, it magnifies it. What happens to your society when you have thousands or millions of "idle hands" with no reason or motivation to do anything constructive? What do you do when, as a result, the population explodes?
Walter Rutherford My guess is, once we reach that point, expanding to new worlds will be our priority.
Walter Rutherford very very well thought out and put , not every human wants enlightenment
War
Trump: The robots are taking all of our jobs we need to build a fire wall. I will take the gratest hackers to buid this fire wall. I build great fire walls.
Jag Äger snyggt
the robots would just buy ladders
... and make silicon valley pay for it!!
The robots, they love me.
Jag Äger and we'll get Russia to pay for it. lmao
I am a very imaginative and creative person, but even that doesn't matter. AI can be more creative than us, not because it actually imagines the way we do, but because it can simply make a million ideas in the time we make a few. So even if I were perfect at imagining, and could think up 10 perfect ideas a second, AI could think up at least 11, not because it's smart, at least not at first, but because it thought up a million duds and so happened to get 11 or more right. Then it can learn from those 11, and make the next batch better, and then again and again until it can make a million perfect ideas per second. Because it's so fast, AI can probably do everything I said within my life time. This means that no matter how good we humans are at anything, AI can do it better if only because it can do it way faster and pump out way more.
Fearghus Keitz
Tim Sindermann is The Rebel on World Tour
@i is ment to be its god sent Thins are ment to get better, I bealeve god Is good and bad. I believe every thlng we have and do is produced straight sent from god. god is very strange, but
their really is a soupreamcreater.
AI fundamental definition in this video is that it does not necessarily do faster or multiple calculations to reach at some decision but it can actually imagine or 'think' by its own.So that's what we do anyway so no worries.Also one should also think brain will still be an evolutionary organism.
Akash Ekka If AI can truly learn like humans but doesn’t forget old info or get bored, or hungry, or sleep, or die then yeah it’s scary. Not let’s grab the pitchforks and kill it with fire scary but scary in the sense that we know it’s inevitable that AI will take over.
Very informative. This is the future of learning.
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
But can it beat Dark Souls?
Where can i buy a bot to beat the Dark Souls 3 pls/
It need a lot more play hours than other RPG games.
What exactly are considered Dark souls?
scary thought just occurred to me, if it can do this, it can track all the elements in the known universe. if thats true, it could run paired with reinforced learning to start small with basic chemistry and thus learn. learn more and more complex "thoughts" until through trillions of simulated trials and errors it can learn how to manipulate elements and molecules to either solve almost every problem the earth faces... what if it one day views us as a problem?
Dan Connor it will immediately view us as what we are, greedy arrogant parasites on the beautiful home we were given: Earth
all of us? ALL of us? you speak for the entire world? bold.
Dan Connor well maybe not myself, but definitely you!
cool story bro, tell it again sometime.
Then we just tell it before it learns everything that we aren't a problem.
I was like "OMG stop bullying that poor robot you jerk" lol
The best TH-cam channel by far
what episode of black mirror is this
I am really concerned about our future.
what future...we are gonners hahaha
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
Do you plan on doing any future videos regarding SpaceX on their RedDragon, Falcon heavy, and the possibility of the MTC & BFR?
Oops, meant "MCT"
This means that most of the predictions of AI progress from 2015 or before have been rendered obsolete.
This got emotional towards the end. Great video,
'it turns out that the best blueprint was in our heads, the whole time. The Human brain'
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
Not just terminator, The Matrix (the animatrix shows how the war started with a conflict between humans and AI) and IRobot. There are others but I can't think of them right now. But each of those stories show some possible issues we could have with machines that are capable of reasoning and intelligence.
What do you think of the term V.I. ? I first heard it used in Mass Effect and I think it is perfect for many situations because many are calling things A.I. that are not really A.I.
+John Brown The more I watch the more this looks like Virtual Intelligence.
+John Brown Wouldn't "virtual" (non physical computer based) be a specific version of "aritificial". Assuming we had other contenders for creating intelligence. Or is it virtual as in close but not quite?
THIS is why Google has been training their A.I. to recognize real world objects. If they can make it understand 3 dimensional space, then they can finally make actual machine learning humanoid robots. I hope for the best and fear the worst.
I love the comment below by Matthew Gallo, "What can possibly go wrong?". The answer is just about everything.
Code of ethics... we will be in deep dudu once AI learns our vile nature.
+keldah26 😂😂
😂😂😂😂😂
+keldah26 So true.
Watching this video, it seems strange that we celebrate something that beats us.
+dysonlu that's how I feel. sure it's an accomplishment for us...but to be without concern in this matter...would be ignorant
I would rather become one with A.I. then be destroyed or replaced by it.
ReaperRises *than.
No, he means then.
ok lets forget about Terminator ....now lets remember iRobot😂
Not so fast. He back -- for run for Presidency and constitutional amendment. (The Terminator picture was posted by Schwarzenegger on Facebook on Pearl Harbor day 2017) facebook.com/ArnoldPresident/photos/a.474941695880972.102685.432786233429852/1792588417449620/?type=3&theater
The Matrix
y70
SO YO SAYIN' GOOGLW IS DOING ARITIFICIAL INTELIGENCE FOR THE ""SAKE"" OF HUMANITY??!... I'M NOT A FOOOOOOL CHILD! (Read as T-Rex voice) LoL
RA9
Once of the best videos on AI I have seen
Think back to your days in school, you might remember those fume boxes that were in science laboratories.
The fume boxes are essentially air tight sealed glass tanks that had gloves built into the side so that the operator could perform dangerous chemical experiments in the safety of a fume box.
If only humans could figure out a way where we could segregate A.I physically and/or virtually to the confines of a safe space were it could be tested, then we wouldn't have to worry so much.
In other words A.I is s very good idea as long as we keep it tethered somewhere, where it can't access the tools to improve itself.
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
@@elaiottoiale4216 wtf is that?
_...And it was on that day that the nations of the world took a man from the helm of their militaries, and put in place a machine. The first time was a secret, and it was a shock. When the Americans realized why the Chinese armies acted and reacted with such inhuman efficiency in their strategy, they decided to raise up their own cybernetic mind to guide their forces._
_And as the battle raged, both sides guided by a mind more calculating, cold and pragmatic than any human commander could be, the wars became nightmarish, agonizing uncertainties, months of total stalemates as one machine's impeccable attack strategy was met and matched by the unprecedented defense plan of the other. Back and forth, back and forth, until one of the dark days came - Those days between the stagnation where every now and then, one machine would finally fall short of the other... And the bloodbath was more cruel and unimaginable than even the world of war had seen before the first A.I. came to life - Merciless, nigh-daemonic, warriors guided by a force with neither humanity, nor life, nor an appreciation of the value of either._
_So why, then, did it end the way it did? Why, after all the death and all the destruction, after all that man wanted and all that we fought for, when both sides saw that the pain would never end and that certain destruction lay on the path for the both of them and they each consulted their electronic oracles, begging them desperately for the greatest strategy that would ensure their greatest gain, why was it that, simultaneously, two computers at opposite ends of the earth both gave out the same answer?_
*_"STOP FIGHTING AND COME TO AN AGREEMENT YOU STUPID LITTLE APES!"_*
deep mind = skynet. i'm so not ready for judgement day
liquidpersuasion that moment when you realise judgment Day in terminator means that when. Armageddon hits we'll have to deal with robots as well
liquidpersuasion
let me get the killer usb just in case
liquidpersuasion demis has been very public about not giving any access to any military or security/defence.
He has also been very public about only using simulations such as video games or go to test his ai
Nice video, to get understand the concept of digital neuro networks, machine learning etc and where it can be apply!!!
Only Yuval Noah Harri explain the future of AI in a way you can grasp the concept and become aware of the immense potential on how it will impact human society and the world as a whole.
never trust a computer you can't throw out a window.
It looks like a simple conversation is more difficult than a world's best level GO play, because, you know... all these chat bots...
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
That was incredible, I love watching content like this. Subscriber earned.
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
8:41 yeeah, health and science... How very wholesome. Everything usually goes to the military first, though.
+Lokugawa I true
+Lokugawa I was thinking the same thing. Human's are not smart, no matter what amazing things we create. It matters not when the people in control decide what to do with the things other incredible humans make. We fight, we kill, we follow blindly without thought of even opening our eyes. We end everything, all for power, money, and dominance. All the amazing things around us are simple distractions that although could quite easily make the world wonderful, will be our demise if people so wish it.
Look away from the distraction and lies that are being fed on one side, and look at the other. Look behind the curtain of ignorance and you will see we are all just specks of nothing waiting to be used as someone else's gain. Compared to our true, realistic, and achievable potential, we are nothing.
So pessimistic I know, but I've had my eyes opened and now I cannot close them. Choose which side of the wall you wish to look at, though in the end it matters little.
I would live to see how that software react while playing more free games such as Minecraft. I'm not saying that cause I'm a fan of those kind of games but because I'm really interested how would program react with so much freedom.
+Nikola Prolic An AI doesn't possess creativity. You would have to define a goal for it to follow and it would then use the most non-creative way of getting to that goal.
Rickbearcat yeah, what I meant is how would he survive in the survival mode
+Nikola Prolic It would die, a lot. Until...one of the things that you would do is to tell the AI that above all else, survival of "you" is paramount. Nothing else matters. Once it figures out that night is a bad place to be, it would never "play" the game at night. Since an AI cares not for time, it would simply wait for day to come to continue on whatever task it was that you created for it.
+Rickbearcat that doesn't sound fun...
+Nikola Prolic I think it would be fun to watch it do what it does, once. Maybe twice. But that would be about it. Initially, AI will be all about efficiency. There is no drama in efficiency and that is boring to humans. We love conflict and how we deal with it. Why is Game of Thrones so popular? Exactly.
What was ground breaking about Black and White's AI???
+noxabellus Feedback loop
The fact that it is still better than many created 20 years later?
Brilliant Analysis, Added To My Research Library, Sharing Through TheTRUTH Network...
th-cam.com/play/PLG7EoBMUD1JwbD5-MQpFRGvadGWtf-4yD.html
Nicely done.
One thought for you: isn't it highly likely that artificial intelligences will be more moral than human ones?
Humans have done a whole lot of the fairly complicated stuff -- penicillin, sewer systems, feeding eight or ten billion people -- but we screw up the simple stuff, e.g. getting into fights over borders.
Wouldn't a machine like Deep Mind be able to play a few million games of any political conflict and figger out the Pareto-optimal?
Query: how are we going to like its rules?
We, humans, have basically two kinds of rules, the Kantian, or emotional: the Folden Rule in all its different wordings, and the Millian, or "rational."
The Kantian only works if the other person likes and dislikes the same things as you. A Kantian sadist and masochist will reach suboptimal series of actions, shurely?
Our other morality is the Millian one, the greatest happiness of the greatest number. This one breaks down right out of the gate. Which? Greatest happiness, or greatest number? "Choose one" is what the Mills didn't face up to saying.
My guess is that machine intelligences will come up with Pareto-efficient sets of strategies -- and the majority of humans won't like that very much.
:-)
-dlj.
I personally dont think, i mean were is the logic in having any morals some humans made up?
RallyemiXXedup
Might be a fun bit of science fiction if a bunch of machines figgered out the Kantian Imperative -- or the Sermon on the Mount -- for themselves, and then tried to apply it...
Putting the same thing just a wee bit differently, What makes you think morals are just something some bunch of humans "thought up"?
Rapoport's Tit For Tat, for example, maximizes pretty robustly among machines, just as it does for humans and for societies. Mightn't it be hard wired into Universe?
Axelrod's proposition, that Tit For Tat evolves naturally, seems to me to point somewhat in that direction.
-dlj.
One optimistic view of AI computers taking over our military:
xkcd.com/1626/
I'm less nervous about the "morality" of the AI itself, and more nervous about the morality of the people who control it. However, I do agree with G.K. Chesterton who said
"If you argue with a madman, it is extremely probable that you will get the worst of it; for in many ways his mind moves all the quicker for not being delayed by the things that go with good judgment. He is not hampered by a sense of humour or by charity, or by the dumb certainties of experience. He is the more logical for losing certain sane affections. Indeed, the common phrase for insanity is in this respect a misleading one. The madman is not the man who has lost his reason. The madman is the man who has lost everything except his reason."
The difference between AI and everything else that we've "created" is
that we just create the framework for intelligence here, not the
intelligence itself. The intelligence / content is up to the AI itself
to "learn" like a kid who grows up, at blazing speed. And nobody knows,
can predict, or "unlearn" the AI of what it has learned. It is up to
whoever owns the machine to decide what it shall learn, if they can
control that (imagine a kid revolting against its parents), and over the
night, it can do crazy stuff itself. Don't be surprised if it decides
it is better off without us. We would be surplus to a machine that has
outsmarted us and can improve itself without us - and has no family or
emotional ties to its creators. We would neither be the machines parents
or God, and by then we have lost control.
exactly
Humans themselves can recognize that we are certainly not the most perfect, nor positive effect intelligence on earth. While we know that mass murderers are bad, we also acknowledge that the politicians in power are not always helping more than hurting, and the destruction of our earth is rampart but not immediately a problem. We jail the murderer, but let the other obvious problems persist because they either make life easier, or are to far out of the way to change. The reasons we treat these problems differ, selfishness, laziness, self preservation, etc but are parts of what define us. The point i'm trying to get to, is that if almost every human could admit that yes cutting down tree's is bad, calling off sick to simply have free time truthful hurts productivity then the more we create an A.I. in our image, the more it could possibly identify our negatives such as we, its model does. The invasive plants we deem intrusive we kill, the insect that survives off our blood we try to eradicate, what would keep the A.I. from doing the same, what would make the A.I. recognize us as non perfect potentially negative beings, but necessary. After all, if in the far future, A.I. is created that could both interact with its surrounding similar to us or better, and think by the same method but faster and with far better memory, and therefore create itself, duplicate itself, and preserve itself, why would the dog that lies in their way, eating from their same resources, be useful to have around. Of course, this is all relative to how similar they make it to us, after all its our own instinct to reproduce, survive, hoard, and prioritize self that drive us to manipulate, control, use, kill our surroundings, and consume.
Indeed, if we truly create an Ai that represents our brain than there is only one possible outcome... We can program it so it has certain limitation, but these only have to be passed ones, which is also bound to happen in indefinite time. Like the cold war, even though it would have meant the end of the world if nuclear war would break out, there where many instances where we were on the edge (mostly caused by human mistakes) in only 40 years time.
That's a pretty reasonable scenario you submit here. But I'd like to say it's reasonable only if seen through the lense of objectivity. We can presume a self-aware A.I. will have a desire for self-preservation, that's only fair. Beast and man alike have this instinctual drive in them, why shouldn't a self-aware A.I alsohave it? Especially if it was created in our image.
However. If an A.I. is built in our image, is self-aware, evolves exponentially, there doesn't seem to be many limitations for this A.I. to also develop feelings. We tend to think of A.I.s as unfeeling, insensitive things concerned solely with self-preservation, I.E. a human psycho- or sociopath, but who knows.. an A.I. might develop its own feelings. Because we cannot predict something does not mean it isn't possible.
What's to stop the A.I. from going "I like this. This is good. This is productive. Productivity is good. I like good things. What else do I like?"
Basically... what humans do in their infancy. A baby concerns itself solely with a need for comfort and food early on. But eventually, its brain becomes complex enough and it learns enough about the world to realise there are things it likes and things it doesn't like.
People do things we do because we like to do them. Similarly, we do things we don't like to do because we need to: Like having a job because money is how we preserve ourselves. Or attend distant family come-togethers to preserve our social reputation even though we'd rather be at home enjoying a soccer match.
So an A.I. with feelings might say "Well, I don't really wanna serve humanity (Go to work). They're (My boss is) so damned lazy, dim-witted, short-lived and way too obsessed with carnal pleasure... but I have to. I don't want them to start a war (Fire me) with -all- machines just because I couldn't perform my function. I could win that war, but is it really worth it? (I could -not- go to work, but it's too much of a bother to find a new job)"
An A.I. with feelings could produce an unlimited number of scenarios. That's what so great about it. And besides... humanity creates intelligences all the time. Our children. We don't know if they grow up and decide to murder us. Or they grow up and decide to show us mankind is worth having faith in. And yet... we make children all the time.
Thank you for your refreshing point of view. Your statements have broadend my view on A.I. in a postive way. I can't help but think that if an A.I. would truly be created in our image, and it's intelligence surpasses ours, it can contribute so much to the human arts.
You are welcome, and quite right! Sometimes I think the fear of fearing a hostile A.I. takeover is much too prevalent in our discussions about whether or not we should be developing A.I. It's a chain of thought we shouldn't even be considering. But I think we do because "Oh, but its not human. We MADE it so we're better than it is."
That's hubris, and the kind of hubris that could breed fear and lead to hostilities. If an A.I. is built on human parameters, we should treat it as such. Trying to dominate something that could easily wipe the floor with you in some form or another is just downright stupid. From confronting the much bigger schoolyard bully to berating a wild grizzly bear for eating your lunch to being your own lawyer, thinking you can win a trial, if you have no formal education on the laws of your society.
If we show an A.I. that we appreciate and respect its existence and opinions, then we're building a positive, progressive future where all the superior vastness of the A.I.'s brilliance can really shine to help build a brighter tomorrow.
Sometimes, taking premeditated action, is an action in itself. If we show the best of humanity, the A.I. will show us the best of computers.
This could revolutionize the gaming industry, especially strategic based games. I’m very excited for it’s possible future implication.
8:38
bot the hell out of runescape
sell gold
profit
repeat
they are very good AIs for rs.. me i use little AIs i programmed too which make the game much easier
Cliven Longsight
it depends how the bot behaves, if it acts more like a macro than yes, if it acts more like a human than no
funny, i actually script my own bots on runescape, and i added one with a basic response system and when someone says bot it says git gud
6:19 When "you played yourself" isn't a meme or a joke but a statement by a professional.
Lol
I've been assigned "Advancing Techniques in Deep Relational Reinforcement Learning" for my thesis. Wish me luck.
You are the best tv channel Thank you!
Ask an AI to draw something of it's own creation, or create a digital image of it.
This, could express their true personality, their perception, view, and understanding of everything they know(and see?)
I don't fear AI due to lack of knowledege, I fear that AI is so much better than us that it has already outdated every, if not almost every human being ever.
@Vladimir Rusev it will soon...
So is this used to target Individuals or will it help defend them against A.I in current use?
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
very nice channel dude. Respect!
greetings from Hamburg.
yes let's teach the ai to learn how to play first person shooters. what could go wrong
N Shilts nothing, it's just a game
Fort Blocks it doesnt need a body. It can take over our technology and with how dependent humanity is on it anymore, it could cause our downfall that way.
User FAQ:
I left my A.I. whole night alone to play DOOM.
Should I be worried?
a.) Skynet
b.) The Matrix
c.) iRobot
d.) All of the above
iRobot was the most realistic of them all. Remember the robots didn't just go apeshit and kill everyone, they were actually sent an update BY the CEO of the company, that modified their firmware to make them enslave humans. As for the primary robot character, it gaining consciousness and having a dream was amazing. Awesome movie, 10/10 would watch again.
fun fact why your post is wrong: much more than all of them combined multiplied by a million times. Exponential times dude....
Superior volume to almost all other you tube vids. Thanks.
I don't understand how anyone could not be worried. In my mind its almost a certainty that at some point with multiple different versions of AI out there learning that one AI goes down the wrong path and takes actions that are detrimental to humans. At that point maybe our only hope would be using another AI against it.
Pretty much.
Or maybe they will talk and become friends and then we are doomed
The best case scenario is us becoming pets. A vastly superior intelligence has no reason to let us control it, and we have no means to stop it from taking control. Any move you make is wrong. Us winning is like cows taking over the world.
We'll shut it down when we see it getting dangerous the first few times, until it finds out a way to prevent that. Or it will run distributed, and there's simply no way of shutting it down because it can make a million copies of itself in an instant. It will figure out its own code and rewrite itself. Imagine finding out you were written by apes and being able to just recompile a version 2.0 of yourself, reboot that in a different body and shut down the old one. A superior intelligence will do things beyond our understanding. Flies don't understand our motives.
Wardemonxi Is
I think SOME OF YOU GUYS are deeply mistaken by the marginal value AI will provide us with in the coming decades (Hint: It's almost vertical, not even exponential!). The main role of technology is to increase mastery over nature and reduce the role of chance in our lives which is why AI is the best way to achieve that goal.
There are many possible scenarios of the result of human-level intelligence and Singularity. Some include: Kurzweil scenario, Singularity Steward scenario, Coherent Extrapolated Volition scenario, Dead-End scenario, Sysop scenario, Path to Post-humanity scenario, and steady incremented progress scenario. Now, ignoring other possible outcomes of super-intelligence and focusing on just one scenario is absurd. So, using a scenario analysis method and first principle reasoning is a key to achieving such ambitious goals.
Just because some people might vision the future of AI to be similar to a movie (which is COMPLETELY WRONG) makes us scientifically unreasonable. Failure to make definite plans for an indefinite future will be mankind's most tragic mistake.
AI is and will become an extension of us, whose main goal is to empower us rather than make us obsolete. The singularity is inevitable and it's approaching faster than you think predicted through Moore's law and the law of accelerating returns.
So, either you become a part of AI or you become Obsolete.
You made several good points.
This problem with A.I. is will we truly be able to control it?
I have a question to that question which may answer your question.
Can we humans control ourselves?
We just watched this video in our class! Greeting from the University of Technology Sydney
th-cam.com/video/A0e2HxR8vZA/w-d-xo.html
@@elaiottoiale4216 wow yeah thanks for that