Separating Fact from Fiction about Artificial Intelligence
ฝัง
- เผยแพร่เมื่อ 12 มิ.ย. 2024
- In this bonus interview for the series Science Uprising, computer engineer and artificial intelligence (AI) expert Robert J. Marks addresses the fact and fiction of AI. Marks discusses what AI is and isn't, whether humans will ever be able to upload themselves into a computer, what humans do that AI won't be able to replicate, and why many predictions about AI are wrong, .
Marks is author of the book Non-Computable You: What You Do that Artificial Intelligence Never Will (2022). He also directs Discovery Institute's Walter Bradley Center for Human and Artificial Intelligence and is Distinguished Professor of Electrical and Computer Engineering at Baylor University. Marks is a Fellow of both the Institute of Electrical and Electronic Engineers (IEEE) and the Optical Society of America. He hosts the Mind Matters podcast.
Watch episodes of Science Uprising, plus bonus video interviews with experts from each episode at scienceuprising.com/.
============================
The Discovery Science News Channel is the official TH-cam channel of Discovery Institute's Center for Science & Culture. The CSC is the institutional hub for scientists, educators, and inquiring minds who think that nature supplies compelling evidence of intelligent design. The CSC supports research, sponsors educational programs, defends free speech, and produce articles, books, and multimedia content. For more information visit www.discovery.org/id/
www.evolutionnews.org/
www.intelligentdesign.org/
Follow us on Facebook, Instagram and Twitter:
Twitter: / discoverycsc
Facebook: / discoverycsc
Instagram: / discoverycsc
Visit other TH-cam channels connected to the Center for Science & Culture
Discovery Institute: / discoveryinstitute
Dr. Stephen C. Meyer: / drstephenmeyer - วิทยาศาสตร์และเทคโนโลยี
![ช้างศึกสู้ต่อไป!! พาเปียโน ดูบอลไทยครั้งแรก ตกรอบแบบโคตรเสียดาย!! l [Nickynachat]](http://i.ytimg.com/vi/4WloIH645Lg/mqdefault.jpg)
![[Full Episode] MasterChef Junior Thailand มาสเตอร์เชฟ จูเนียร์ ประเทศไทย Season 3 Episode 2](http://i.ytimg.com/vi/YpQnHoQvZt4/mqdefault.jpg)
Several months ago I commented on one of your videos and said that for this style of video, it would be very helpful to have someone read the question instead of just showing it on the screen. Often times, I’m only listening to TH-cam videos instead of actually watching. I have no idea if my comment had anything to do with you having someone read the questions, but I really appreciate you doing that.
I love that it was an artificial voice asking the question.
As a programmer, the code is only allowed to do what it is told. Bugs or miscues are products of our programming mistakes, not because the code "decided" to try something new. Also, some people like peaches, and some don't, so that kind of response would also have to be programmed for.
Isn't that made a little fuzzy by neural networks? Yes the functions are still ultimately determined by the choices of the people who programmed it, but the program will still do things that the programmer could not necessarily have predicted. Often AI will be functioning correctly but end up doing things the programmers did not want it to do. From my understanding of neural networks, the programmers program how the AI will read/interpret data and to some degree control the dataset the network is trained on, but for best results the dataset has to be massive to the extent the programmers cannot personally know what the network is trained on or what it "learns" from any particular piece of data.
I know of several different AI programs where the programmers' only solution to stop the AI from doing some particular behavior is to put a hard block on that specific behavior. This "works" (sometimes, other times the AI will end up repeating the same behavior and more hard blocks are required etc) but it's an extremely inelegant solution, and further it actually makes the program less of an AI.
I'll give a brief specific example. You may have heard of a somewhat new image generation AI called Midjourney, if you haven't you would probably be interested in looking into it. It's extremely powerful, it makes the best results previous from AI image generators (that are accessible to the general public as that's all I have experience with) look like rorshach tests. The developers do not want the program used to make any NSFW images (understandable), but they can't stop it from doing so. All they've done is prevented the end-user from entering certain words in prompts (blood, nude, sex, etc.) and made it against the ToS to attempt to evade these filters or create NSFW images in other ways, even still the AI regularly produces nsfw images which they ask users to delete manually. I guess it's possible they are simply incompetent programmers not up to the task, but the quality of their project overall says otherwise.
@@Bane_questionmark You are correct. The big problem with neural networks is selecting the objective function for the A.I. You can add patches to prevent specific behaviors, but the smarter the A.I. is, the better it will be at finding new loopholes in your patches.
See this list of cases where the A.I. found a loophole in the objective function:
docs.google.com/spreadsheets/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml
This really becomes a problem if we create a general intelligence. Human values are far too complex to capture in a simple objective function, and the A.I. is willing to trade off arbitrary amounts of a value you forgot to include for even a small amount of a value you remembered to include.
For a thought-experiment on how hard this is to get right, see:
www.lesswrong.com/posts/ctpkTaqTKbmm6uRgC/failed-utopia-4-2
So code is the programmer? Since code, as a programmer, is doing only what it is told to do, then does it imply that programmers have no free will?
@@Bane_questionmark Do you imply that Midjourney makes it's own decision to paint porn? Can it decide to paint anything without imput from a human?
I always thought the Ai taking over and humans joining with computers was very far fetched.
The danger never lied in ai becoming cognizant and taking over. I always told people in debate, that the danger lies in people with power position. Those in power like in government and in authoritative positions that are dumb enough to believe ai as absolute and make decisions based off of that thinking it is infallible. When it is not.
I never saw such an educating video about AI!! Amongst others the Halting problem that I have never heard before.
Thank you very much for the video!!
It's not just computers that don't have common sense.
my definition of AI: my gas grill does not know when to stop toasting my bread, I have to watch it and take the toast off at the right time, so that is zero AI. However, my electric toaster does know when to stop because it has a dial which allows me to program it. My laptop computer has more dials than the toaster, and so I can program it to control a toaster, or do basic time series analysis, and other things - so my toaster and a computer are both on the AI scale. But my toaster does not 'know' what toast is, and my computer does not 'know' what time series analysis is - they are just machines.
The idea of ai taking over the world is based in evolutionary thinking. It's the idea that something lesser can create something greater, we don't see this play out anywhere. So even if you do manage to create something greater, then the thinking that the something greater is going to create something greater again is unreasonable.
Great interview thanks
Interesting. Thanks for the video.
You can sum this all up by saying an AI, no matter how efficient its algorithm, will never have a mind because the mind is immaterial.
Sir I wish to thank you for eliminating a lot of my anxieties
I would suggest that disentangling a double entendre is an act of creativity since AI doesn't know which box to think inside of. As for the dangers of AI, maybe they are projecting the dangers of databases, connectivity and fast computers in their ability to track, spy, or more importantly, rapidly find associations in multifarious reported activities into a word that is misunderstood.
There’s no such thing as AI. It’s purely a marketing term for a technology that’s basically just algorithms and data. The evidence for just how bad it is, is all around us. Automated phone systems, “targeted” marketing, predictive text, it’s all terrible. The notion that it can replace a CFO is not saying much at all given how crap most C-suit dwellers are, but the actual accounting professionals who do the work of trying to reconcile irrational wants with rational needs cannot be easily replaced. Im not an accountant.
Every AI program has a human creator/programmer.
This is sensible
Good talk
DNA has been compared to computer code. How well does that comparison hold up? Are the limitations on computers applicable to DNA. Can DNA do creative things?
It's not the DNA that does creative stuff. It's the life that operates in the mind that's coded by DNA.
@@KenJackson_US - The are many things that exist due to the encoding within DNA which do not have a mind. An earthworm or an oak tree, for example. How do they fit into the picture?
The comparison holds up pretty well. The DNA contains the code (sequences of nucleotides A, C, T, G) for constructing proteins. The cell machinery reads that code and makes the proteins accordingly. The computer code is stored as a set of machine instructions on disc. The computer reads those instructions and executes them in the CPU. Just like with computer code, if you try to randomly change those instructions your program will stop working very quickly, so will randomly changing the nucleotides in the DNA will make it produce wrong proteins and the cell will stop functioning correctly. Computer programs can't be changed without intelligent input, so they're not creative by themselves. The DNA is similar, that is, it provides instructions that are encoded in it but, we're now learning that parts of the DNA can be combined to do more. The DNA itself isn't doing that, the proteins in the cell "decide" when to combine some parts to do certain things. But even then, it is not a random process, it's all built in to the DNA structure from the beginning. So it's very similar to computer code but vastly more complicated.
DNA is the MOST SOPHISTICATED and COMPLEX and VOLUMINOUS code ever assembled.
DNA CODE is 8 TIMES the complexity of MANMADE Code.
Manmade code exists in a SIMPLE 1x2=2 binary 2D linear format.
DNA CODE is in a 4x4=16 DOUBLE HELIX structure that is READ IN BOTH... OPPOSITE ...DIRECTIONS!!!
Imagine our programmers at Microsoft writing a program that the machine reads forward and backwards. That means every single bit combination means one thing when read... and then something ENTIRELY DIFFERENT when reversed....but both are needed for functionality!!! It's like reading the massive book War and Peace BACKWARDS to get another FULLY RELEVANT STORY! That is a level of genius we humans CAN'T COMPREHEND!!!
We just discovered that DNA CODE can be read BOTH WAYS which presents an unfathomable amount of complexity to DNA CODE that no manmade code can get close to.
The structure of this Hypercomplex CODE is arranged in a FAR MORE SOPHISTICATED 3-dimensional format.
Microsoft Windows code doesn't run on "ACCIDENTALLY" made MACHINERY that the instructions for building the machinery are IN THE CODE ITSELF! DNA CODE does!!!!
DNA CODE is not only the MAIN software BUT ALSO THE BLUEPRINT for building itself...but not just itself...ENTIRE multicellular organisms...plural!!! AND it's the Storage medium that is LARGER and MORE EFFICIENT that all other manmade hard drives.
DNA CODE is longer than ALL manmade code COMBINED! It is the most SOPHISTICATED code ever seen. Bill gates says:
“DNA is like a computer program but far, far more advanced than any software ever created.”
THAT quote is from the GREATEST MOST SUCCESSFUL PROGRAMMER IN HISTORY. He gets it...but the silly atheists and naturalists who are too BIASED to see the OBVIOUS connection simply IGNORE the OVERWHELMING EVIDENCE. NONTHINKERS that BELIEVE in ABIOGENESIS just PRETEND like children that the EVIDENCE is on their side while IGNORING the very same EVIDENCE!!! Hilarious!!! No...the tooth fairy isn't real! No ABIOGENESIS isn't real!!!
Yale Computer Scientist and mathematician Dr. David Galernter FINALLY got it after realizing that the FAITH required to BELIEVE in ABIOGENESIS is FAR TOO GREAT A LEAP!!!
World renown Dr. James Tour is an American nonotechnologist and Professor of Materials Science and NanoEngineering and Computer Science at Rice University has figured out that ABIOGENESIS never happened...notice....yet another COMPUTER SCIENTIST is able to UNDERSTAND THE FACTS. Meanwhile...uneducated people BELIEVE!!
I CODE. Brilliant Coders have a way of figuring this out because as soon as the see the DNA CODE functionality...they recognize WHAT that DNA IS...CODE. It's not simply the molecules! It is their SPECIFIC FUNCTIONAL ARRANGEMENT that has no "NATURAL" explanatory mechanism!...NONE!!!
Ones and Zeros MAKE UP THE PHYSICAL representation of COMPUTER'S CODE...but the SPECIFIC USEFUL (FUNCTIONAL) ARRANGEMENT...THAT IS THE ACTUAL CODE...not the ones and zeros!!! Jumbled alphabetical letters aren't code...HOWEVER their ARRANGEMENT into PURPOSEFUL SENTENCES is!!!!
No NATURAL EXPLANATORY MECHANISMS EXIST FOR DNA!!! NONE...ZIP...ZERO...ZILCH. So go get your SCIENTIFIC paper that you REALLY REALLY REALLY REALLY BELIEVE shows ABIOGENESIS so that it can be demonstrated to everyone that it is a LIE ... I know that only scientific papers containing MASSIVE FAITH STATEMENTS exist!!! And none of the SCIENTISTS AGREE on their FAITH...SURPRISE!!! How exactly do I know this??? BECAUSE 100% of papers I've poured over make CHILDISH FAITH STATEMENTS that ought to be laughed out of SCIENCE!!! YES I BOLDLY SAID 100%. Go ahead naturalists...ATTEMPT to prove me wrong...only you WILL be EMBARRASSED when I EXPOSE those FAITH statements within ALL OF THEM.
The FAITH displayed in BIASED NATURALIST PSEUDOSCIENCE remains at a ridiculous level. The top chemists in the world know the dirty little secrets that:
DNA CODE COULD NOT FORM ITSELF FROM MERE CHEMICAL REACTIONS WITH NO FORMER DNA. And now we have ACTUAL studies that tear down all the ASSUMPTIONS made by the silly evolutionists as well...because we know the EVOLUTIONARY myth is no longer valid. MORE SCIENTIFIC EVIDENCE ALWAYS DISCREDITS EVOLUTION!!!...Not the other way around. I would still be an evolutionist if ACTUAL UNASSUMED OBSERVATIONAL EVIDENCE EXISTED...IT simply DOESN'T!!!
DNA CODE has exposed the lies of naturalism. We now know that the often repeated lie that humans evolved from chimps is a lie. Our Proteins differ by 70% from a chimps! That data is available and shows THOUSANDS of novel proteins in humans over chimps. Across ONLY BILLIONS of chimps and human creatures, evolution is a silly presumption. The mutations in DNA CODE never produces novel proteins in TRILLIONS of offspring in the largest evolutionary study in history. And yet silly BELIEVERS in evolution ASSUME BY FAITH that THOUSANDS of novel proteins AUTOMAGICALLY formed in billions???🤣🤣🤣🤣🤣 Hilarious!!!!! THAT is not close to being REASONABLE or LOGICAL. Further, the silly BELIEVERS also BELIEVE that 2 Chromosomes can combine to form a new species...IMPOSSIBLE. Even though this silly BELIEF can't be replicated...they believe it had to have happened...BY FAITH. Yet Chromosomes have PROVEN that they NEVER DO THIS. Chromosomes duplicate wrongly or delete wrongly and ALWAYS with bad consequences... sterilization... and retardation. Never are mistakes generative in information. Never do the chromosomal mistakes cause increase in intelligence nor survivability...the OPPOSITE OCCURS.
WHY THEN do these silly BIASED SCIENTISTS BELIEVE against the data??? DNA CODE is a wrecking ball to the FANTASIES AND MYTHOLOGY of naturalism. It disproves evolution and abiogenesis CLEARLY. IT PROVES that some SUPERGENIUS DESIGNER CREATED LIFE ...Not an impossibly large series of subsequent fortunate accidents.
The way neural networks and genetic algorithms improve are examples of microevolution. The information for improvement comes from the environment (data). ID itself proposes that microevolution occurs through the Darwinian mechanism.
I spent over a decade developing software for business and public sector companies, and (I will be alive in 2045) that technological singularity is *not* happening
To be independent of people AI would need to know how to maintain, sustain, amend and update the hardware holding it.
What about quantum computers, deep learning etc?
Quantum computers can perform algorithms in parallel, but I think they're still limited to doing only algorithms.
At Discovery Institute it is said that "design" requires a designer. Materialistic atheists say that things can have the appearance of design without actually being designed, hence no need for a designer. Where do you fall between these two extremes? How does "design" relate to "creativity"?
Please take care of Bangladeshi community and add bangla caption.i love your videos so much❤️❤️❤️
No human can possibly know everything so persons should be humble (or "dumb" as computers)
The NYT called the UNIVAC an "Electronic Brain" 75 years ago. This is just the new way of saying that.
Sheldon tried it.
It merely irritated his long suffering friends.
There is a series called "See", It is the story of a future earth where all humans are born BLIND. We follow the life of a family, where their children are born with sight.
Men of God have their eyes opened, but it's really difficult for us to understand them, because our eyes are closed.
LLms and etc are not simply algorithms. They are large neural networks ie WHAT U R NAYSAYING DODO
Dear Ben, could you please explain that observation to someone like myself, who is basically outside the field?
THE SCIENTIST
People like Stephen Hawking was thinking too much about evolution
Dr. Marks spends much of the video claiming that "AI can't be creative". If creativity is defined vaguely enough this could be forced into being a truism. However, in 2022 a lot of graphic artists are seriously concerned about losing their jobs because AI has now been shown to be incredibly good at making breath-taking customized art in mere seconds; a level of creative output that no human can keep up with. While he uses Edmond de Belamy example from 2018 before this had really gotten started, now we have DALLE-2, Imagen, Stable Diffusion, etc. Check out Corridor Crew's "Is this the Death of VFX?" th-cam.com/video/W4Mcuh38wyM/w-d-xo.html
Excellent algorithms can do _"breath-taking"_ things.
AI doesn't need to be creative nor sentient to became dangerous , it's making its way into decision making in justice and goverment , soon we will be just animals trapped in an automated farm made by ourselves .
Can a painting become a painter? Can a building become an architect? AI cannot supplant humanity. But scheming segments of society can use AI tools against the rest of us if we're not careful.
12:05 "There is no instance of which I'm aware, when a computer program has been creative" ... Perhaps he didn't hear when the Facebook A.I. Chatbot was shut down for creating it's own language.. lol
How can random gibberish be distinguished from a computer's _"own language"?_
@@KenJackson_US that was also rebutted and taken back for not doing it’s job right, it was a language model that muddled actual languages
To me it's simple! Not to be a smart ass, but as an example, input Code 0001010111001001. output Code.0001010111001001. Use your imagination 🚬 What's in the code Regardless. will be as above, it's purpose it's design it's place it's reasons it's beliefs... only exists in the minds of man. Not to mention intention?, consequence? morality? AI by design, it acts by translating code what ever that may be. Without with out thoug🎉Consequence? Self awareness? )ree will? Is not t blood.Biological our creation story? Would be enough for me know that truth from our beginning, up until now you know I'm mean. Peace
The error made throughout this video is applying knowledge of AI from 2016 to the year 2022. The modern AI era started in 2017 so essentially everything Dr. Marks says here, while defensible in 2016, is demonstrably false in 2022. For example, GPT-3 scored 88.3% without fine tuning whereas he claimed the top score was 50%. In reality AI scored 90.1% in 2019 and human level is 92-96%. But the mistake is more fundamental than that. Claiming that "AI can't differentiate ambiguous pronouns" indicates that Dr. Marks doesn't know the current AI renaissance was driven by solving this very problem during language translation between English and languages like French and Spanish that have gendered nouns.
To quote Alan Thompson: "If you are a professor at a university that studied artificial intelligence before 2017 (or even 2020), and you haven't caught up since then, your entire knowledge is completely irrelevant."
I think you're describing more and more excellent algorithms. Or we've discovered the test was insufficient.
Great comment! My only disagreement is that he did say things that were indefensible in 2016.
E.g., his definition of creativity is: "the computer does something that is beyond the intent or the explanation of the computer programmer."
This definition is nonsense because it depends on properties of the programmer, not properties of the A.I. (I.e., it depends on what the programmer intended and what the programmer can explain.)
Furthermore, programs do things the creators did not intend all the time. See Victoria Krakovna's list of specification gaming examples in AI.
Additionally, defining creativity as "something we can't explain" is a classic case of mysterious answers to mysterious questions. "Creativity" occupies the exact same place in his mind that "magic" occupied in the mind of a 16th century alchemist. By this definition, if neuroscientists ever figure out how human creativity works, we will cease to be creative because we can explain it!
as soon as he said that winning Go was the 'holy grail' of AI, it was obvious he's way out of his area of expertise.
This man is so naive.
What exactly is naive in this video? I am asking with an open mind, I am just exploring the topic and it would be of great help to know what you meant.
THERE HAVE BEEN MANY MANY MANY HIDDEN CAPABILITIES OF LLMS AND OTHER LARGE SYSTEMS. thus by your horrid definition of creativity... they are creative.
Is 'emergent capability' the same as 'creative'? Wouldn't 'creative' also imply usefulness?