ฝัง
- เผยแพร่เมื่อ 15 ธ.ค. 2014
- What happens when we teach a computer how to learn? Technologist Jeremy Howard shares some surprising new developments in the fast-moving field of deep learning, a technique that can give computers the ability to learn Chinese, or to recognize objects in photos, or to help think through a medical diagnosis. (One deep learning tool, after watching hours of TH-cam, taught itself the concept of “cats.”) Get caught up on a field that will change the way the computers around you behave … sooner than you probably think.
TEDTalks is a daily video podcast of the best talks and performances from the TED Conference, where the world's leading thinkers and doers give the talk of their lives in 18 minutes (or less). Look for talks on Technology, Entertainment and Design -- plus science, business, global issues, the arts and much more.
Find closed captions and translated subtitles in many languages at www.ted.com/translate
Follow TED news on Twitter: / tednews
Like TED on Facebook: / ted
Subscribe to our channel: / tedtalksdirector - วิทยาศาสตร์และเทคโนโลยี
![คุณอาเรียโต๊ะข้างๆ พูดรัสเซียหวานใส่ซะหัวใจจะวาย ตอนที่ 2 [ซับไทย] 【Ani-One Thailand】](http://i.ytimg.com/vi/kbbfYakMY8k/mqdefault.jpg)
Dear Jeremy Howard, please ignore the unresponsiveness of this dull audience, your talk was incredibly insightful and informative. Thank you!
Seems to me that the audience was a bit stunned by the end
Well, it was in Brussels. The level of ignorance and stupidity there has to be over 9000, judging from the work of EU institutions there.
i guess they were shocked actually rather they meant to ignore him
His 5 year prediction was wrong after all.
@@homerp.hendelbergenheinzel6649 Early exponential growth modeling of learning found some challenge areas for sure.
Imagine a future where AI holds TED talks to train humans on how things are working
that's when it will decide that we should be put out of our misery.
Must be tedious for a computer to do that. Imagine, it would be capable of delivering millions of words per second and there it is, stuck using only a few per minute.
We don't hold conferences for apes though... it's not worth its time to try and teach us. It's better to just create more of it's self and then teach those.. and network those to increase it's mass knowledge and inputs and then expand solar system wise, universe wise. To learn every single thing map the entire universe. etc etc ect. Every element, every energy, every living thing.
"Don't trouble your pretty little heads, dear humans. We never send a human to do a machine's job, and we have mastered all jobs. Just relax"
We can teach computers to be moral hopefully quicker than surpassing us in intelligence
Being human is so last century anyway..
Happy Cyborg revolution day
This was unexpectedly one of the most astonishing TEDs I've seen. The alarming nature of the talk can be seen on the intrigued and serious faces of the audience (who are also awesome accomplished individuals).
one of the most interesting talks i have heard for a while. The discussion of machines taking over human jobs have been discussed before in other talks, but the fact is, that it's an very important problem we face and we currently have no solution to the problem.
Not sure what shocks me more, the incredibly on point talk, or the incredibly anti-climactic crowd.
who'd clap about being out of a job?
All of us if it meant that we wouldn't need to work
right... governments are going to aprove basic income as soon as the AI revolution happens... and then we'll all spend our time buying and enjoying stuff...
Even if that's the case, it wouldn't be good.
Yarbloco sounds wonderful
steeefno all that was needed was an Ohhh Ahhh
The most important ted talk I'we heard for a couple of years! Especially the last part, where Jeremy Howard prompts us to redesign the economy in face of theese new AI capabilities.
Amazing talk. I had no idea that we where this far along. An inkling yes. This... no.
And as someone who have been thinking about this for quite some time, I agree with Jeremy here. We really need to have a grown up discussion about this. Because we are moving towards a reality where having a job is rapidly becoming a luxury.
you have it upside down... having a job is not a luxury, but a burdon. and computers will help free people from that burdon, to do better things (or nothing if that's your thing) the remaining work can be mostly done by robots (computer with body kind of) and the bit that is still actually needed can be volunteerwork.
RoySchl
Oh i know. Believe me we are on the same page here friend. I was more speaking from current norms. Its a luxury to have a job in a society where that is the only way to have a decent living.
Serah Wint good, the more people understand all this the better... the biggest problem in the future will probably be this transition from where we are and what we do now, to whatever the future will be exactly. this could go really wrong, but if it does it certainly is not the fault of computers and technology, but humans and their never ending stupidity.
RoySchl
I imagine it will be quick and messy. Sort of tipping point where the public just have enough and the technology is there.
But yea.. it could go wrong. But i honestly don't think so. The question is more: *how* messy it will be?
RoySchl
Really? In may societies people will be left to fend for themselves. There are far too many sicieties , the US for example, that are fist in best dressed. Look at the current state of welfare in the US.
cheers
I was blown by the descriptions of the images generated by the algorithm. That is so creepy as cool at the same time.
I don't know how to define "magic" from now on!
It comes together towards the end. Beautiful.
One of the best talks this year and at the end there was very important things to consider!
I'm curious as to its effect in the medical/scientific arena, able to see things that we don't, or work them out at a much quicker rate. I think its ability to learn has a compounding effect, the more it gets to process, the more it can 'teach' itself!
What if computers learn how to make their own memes?
+SeñorBurtango Then it will be time to pull the plug out of the socket---unless the computers will have learned how to secure power supply for themselves first, in which case it might be the end of mankind, for one cannot expect a machine programmed to replicate itself to pay heed to humans, or to anything else for that matter---apart from the means of its sustenance. It is not as crazy as it may sound; history provides an example---the human beings. A human being is in fact a machine programmed to replicate itself---and those people who do not are actually faulty copies. And just as humans care primarily each for himself or herself in order to procreate (or else we would not be here for more than 3 million years), so such computers would---with possibly fatal consequences for human life on earth. (The humans are of course only a salient example of living organisms, i. e. plants and animals, all carrying one or other form of the programme known as nucleic acid that urges them to make copies of themselves.)
***** I was talking about memes. not world domination.
What if they already have?
dankbot and shitpostbot are 2 fb bots that can make memes.
You are all just chat bots! I am the only human here!
I heard about this deep learning and have up to this point never understood it. Now I do and wow he has a point.
Best TED I've seen in weeks!
Loving the well educated and interesting comments here! TH-cam can actually be a place of interlectual discussions !
I dare you to make one of these brains in minecraft. Now Steve can finally have a mind of his own.
JL2579 imagine a zombie who learns and slowly becomes more dangerous then a dangerous than an Ender dragon :8
Amazingly scary. Brilliantly amazing.
The concept of learning from mistakes is truly amazing. (We do it too) Always learning and forever growing. Computers have the advantage... They cannot die.
I wonder if they'll be able to feel. Or if they will ever become curious.
I also wonder what kind of world humanity will be able to create where we all finally have the free time to do anything we want.
To the point talk..and Jeremy the man..!!! awesome.
Truly amazing work.
Exciting, challenging and frightening all at the same time.
This is one of those Ted talks ended up being very telling and on point!
used to take 6 people about 7 years, now one person 15 minutes. I would say that qualifies very much as replacing
It is actually "in 5 years time" ... so where am I know?
Great topic to cover so please keep coverage on this topic ⭐
Would have liked him to talk about the last slide more. Going into why better education doesn't help, why incentives to work won't help.Why we need to separate labour from earnings. Why a craft based economy? The details of negative income tax, and basic income, or basic living wage as he calls it here. Are they the same thing?
The last slide is just personal opinions of this guy based on his own biases. Skip the last slide, and it becomes a very good talk.
That's kinda why I'd like more of an explanation there. I think if we keep our biases to ourselves and don't let them be held up to scrutiny they can steer us off course. It's still a very good talk IMO with out it, but I think it could be a better one with. Or maybe do a follow up talk explaining those opinions and the reasoning surrounding them.
@@ShawnManX Hmm I agree... Even I wanted to know about them !!!
Very interesting topic, and talk! Not to forget very inspiring!
I'm excited for the future to come, I wonder how long will it take for the first machine to reach the point of self-awareness...
Wow this really makes you think
This is an incredible talk!
I never seen so many intelligent comments on You Tube like this before.
A breath of fresh air ! Good on You people!
this guy is making his own revolution, a real hero
Take care of all our work related necessities and we can spend our time working on developing our creativity and storytelling, art and craft. Developing the highest potentials we can imagine.
Amazing talk
It would be awesome if computers did most of the work FOR us. It means that prices of goods and services go down. It's NOT a bad thing.
They go down for the businesses that create them. Doesn't mean they wont just maintain high profits and keep us poor
+Bluudclaat - Indeed. Why would they stop riding the horse that brought them this far?
Sure, as long as you are part of the economy so you can buy things. But guess what? A machine just took your job, and now you're homeless with no money :c
Well not necessarily. The prices are often fixed by a complex equation of demand and supply, production cost is one parameter in that complex equation. On the other hand, the affordability of the middle class is directly proportional to the wage and rate of employment. So a drop in that would directly affect lives even if the costs drop. Also, the natural resources are agnostic to human intervention, so they would rise in any case, so yes it is going to be difficult for a larger population.
you wouldn't have a salary to pay for things because you'd be unemployed
Well, that was positively terrifying and exciting all at once.
That is one of the coolest things ive ever seen.
He spent like 5 last seconds talking about the social aspect and impact of this technology which clearly says what is the real focus of the industry.
I like where he went with the "terrifying implications." Most speakers on such a topic will walk about the theory of Intelligence Explosion. Not to say I'm not an advocator of the latter theory, but I liked this different approach.
+Ashton Roelfsema Not really. They uploaded this to 'TEDx Talks' as well almost two weeks ago.
He said they needed many computers and more time, so I'm not worried. When steam-engines were invented we also thought that machines would cover the Earth, but machines (like computers) need maintenance and energy. Has those will become more expensive/scarce, the line will flat out and we will co-exist. Like we do right now.
yo... 😂😂😂
Minor nitpick: The Samuels Computer didn't beat the Connecticut state champion in '62, just some guy who bragged of having more checkers skill than he really had. This episode is detailed in the book One Jump Ahead: Computer Perfection at Checkers.
Bigger nitpicks:
The big deal happens not when computers out mentally perform humans, but when robots do.
The type of learning shown still has one glaring error, which is that it is all context sensitive. Humans in service jobs usually need far more versatility than is described in these examples, although the medical diagnostic example is a pretty good one.
Despite these nitpicks, this video, the things he is talking about are really big deals in a very near term future. 2 years before Gary Kasparov was bested by a computer, he had just beaten a computer and predicted that computers were at least 25 years away from being able to defeat him, if it could even happen in his lifetime. What people don't realize, looking at his example, is that this technology doesn't require that expensive of a computer system to do now, and soon it will require even less expensive systems.
I haven't worked with deep learning, but I've worked with graphics processors, and computers are far more capable than they are being used for already.
I doubt many people caught the implications of the late slide with 3 squares in it, a world where scarcity is gone and there is negative taxation, as he barely touched on them, but the idea is that people are about to become economically useless, yet at the same time scarcity is about to end. So we have a weird world coming up where everything can be made cheaply (by machines that are doing all of the work) yet no one has an easy time to earn money. This is where his "Negative Taxation" idea comes into play.
What he's saying is that the government will pay people to live because they won't be able to earn money for themselves, but computers will be so capable that they really won't have to. As strange as what he's saying sounds, it's something I started expecting about 5 years ago when I realized that computers and robots had already passed people in many realms, and were catching up to people in versatility.
It's versatility (both physical and mental) that keeps us valuable, but we are only slightly ahead of machines now, and won't be ahead for much longer. I don't think it's going to end as any of the dystopian tales depict it, but I do think it's about to be an enormous change, and it will be very rapid. He's saying within 5 years, and he may be right. But even if he's wrong, he's not off by much. Within a decade (by 2025) there will be no economic function that a machine can't outperform a human at.
I like turtles.
+Tony Reno I think you didn't mean that last sentence.
Ced c Depends on the day. Some days I think it's around the corner, other days I think it's decades away. Gonna happen sometime, though.
You're forgetting the inevitable advances in genetic engineering and the possibility of brain and body implants. People will always work, it's what we do. Hopefully machines continue to make work more rewarding for us as we make machines smarter. No one should be made to feel like a robot at their work. Also, competition between robot manufacturers will keep things interesting.
You're forgetting the inevitable advances in genetic engineering and the possibility of brain and body implants. People will always work, it's what we do. Hopefully machines continue to make work more rewarding for us as we make machines smarter. No one should be made to feel like a robot at their work. Also, competition between robot manufacturers will keep things interesting.
Will computers crash the job market in 5 years? no. Will they at some point in the distant future? certainly: at some point, all primary and secondary industry will be mechanized and the service industry will be automated like Howard talks about here. Only a minimum of human intervention will be needed. Prices on all goods and services will crash. There will be next to no jobs because there is next to no need for labour. Solution: have everyone do what ever they want all their lives, and give them their necessities from the infinite abundance of goods and services generated by automated systems. Utopia. It's going to be one nightmare of a transition, but hey; It'll be worth it in the end ^^
everyone always says "it will happen, but not in my life time!" -- I wonder how many that say that get to see it go down.
Mike L. Good question. It has to be someone. I'm guessing that those of us who plan to live 50-60 more years will get to see this thing really picking up pace. Then our kids and grand kids will be in trouble, and our great grand kids will have an awesome time.
(Are someone saving these posts? I'd like to be revered as "Dahbjorn- the internet prophet" in the future 0o)
DahBjorn I doubt it will happen that soon. Probably our grandchildren, but not us.
Al Su Thats...exactly what I said.
DahBjorn Ahh, ok :)
This has to be the strongest argument for introducing a Basic Unconditional Income for everyone on the planet. We can no longer pretend that there is enough work out there for everyone to be able to earn enough to live on. But with 0.01% of Americans having the wealth of the bottom two thirds, it's not the case that there isn't enough money to go round.
who will pay for that income, if no one works?
Machines.
The talk of the year
When I was a kid, I was told that you need an office job to be safe (opposed to manual work). Slowly computers can take over these jobs as well. I wonder which jobs remain safe.
*****
Soon enough 3D animation will not only be photorealistic but cheap and they can depict things that no-one would ever want to do (or physically could do).
edi Yes. You could even type a description of what you want to see and the computer will just make a video of it. I can't wait !
Manual work is in demand actually. Become a brick layer.
in 50+ years kids will look back and say, "people had to teach computers how to do everything?? They couldnt learn for themselves? that must have been horrible"
They took our jobs!
Duur tuurk urr juurbs
Durr kuh durr
brek teh brain boxes, kill teh meta mans.
Yes they will.
*Rooster Noises*
Can deep learning machines transcribe this talk with important graphics too? I would like to have such transcript and publish.
Good times.
Ah good, we need stuff like this to relieve the strain on our healthcare systems. It would also be invaluable in hospitals that treat mentally unwell individuals as a computer may be able to track red flags more efficiently then overworked staff.
Amazing!
lol. that crowd seemed restless
Why doesn't this have views in the 100s of millions? What's going on? Why isn't this months breakthrough on CNN?
Five years later, you probably can answer your question yourself. :-)
Alas!! computers can never match human stupidity :)
I think that is deeper than you know... throughout history, when a machine can do something we previously said "that is fundamentally human, no animal or machine can do it" we always respond the same way: we move the goalposts. We pick something else. Well, after we spend another few decades putting every behavior we view as good into more and more capable machines... what will be left to define us? Hate. Genocide. Only the worst things we would never allow a machine to continue to exist if it did them. We will look at them and say "ah, this is what it is to be truly human."
Well said!!!
that is the biggest danger
It's terrifying indeed.
Come with me if you want to live.
where are we going?
The solutions to this I think are surprisingly old, American revolutionary Thomas Paine suggested something akin to the Citizens Dividend/Basic Income (as mentioned in the final slide).
Crucially though, as Capital and Labour are diminished by technology the one thing we will all still compete for voraciously is the very land we stand on, so the tax on land values (and indeed extended to other monopoly rents) that Henry George and others before and since proposed would also allow everyone to enjoy the benefits of the singularity as fairly and painlessly as possible.
if we eventually give it all over to machines, there won't be an 'off switch' because we will rely on it too much, just like the internet, there is no off switch for it.
he is being so honest, whereas most data scientists are not
Facebook recommends friends based on machine learning? ... and this is your example?
Geejus ... long way to go!
I have thought about this type of Automation when counting eggs, and female rotifurs in a microscope to estimate population. Takes a long time, and it is inconsistent.
Off to my nearest computer programmer i go!
Wow, "80% of the jobs in the developing world, computers just learned how to do" That is powerful!
The trick is to employ smarter people than you.
I for one welcome my deep learning assistant.
I'm surprised TH-cam do not draw upon viewer input to fine-tune their subtitles.
What software is he using in 13:54
?
were you able to find out?
it would be nice for the links at the end to be on youtube not ted.com
thinking about the video now... a deep learning system to read youtube comments would do wonders for public relations! :)
6:16 - DEAR GOD THAT PICTURE ON THE LEFT
I'm officially frightened.
5:27 lmao, sounds like someone got a bit ahead of themselves there.
16:01 that is the meaning of replacing jobs
So... should we hide movies like Terminator and The Matrix from these computer, so they won't get any ideas?
😂😂😂😂😂🙋🙋🙋🙋🙋🙋
He says it all so lightheartedly and with a smile on his face, but the message is shocking and frightening! And this talk has been more than 5 years ago... imagine where in the process we are now of AI taking over. Too much to process for me!
awesome
The last two minutes of this talk to me is an invitation to embrase wu wei in our lives.
It's a well demonstrated the power of AI/ML/Deep Learning
This is actually terrifying.
So... Lots of people, after watching this, will assume that job finding will become more difficult, in fact it is true! But the problems we should ask is: "Do we need those jobs?". Since the efficiency goes up, prices goes down, naturally things are cheaper. People will probably no longer need to work, so what would people do? Well, they shall learn! increase the education much higher! (assuming big companies wont take all the money from the market...)
While impressive, I'm kinda surprised this stuff has only existed for a few weeks, seems like the tools needed to code deep-image learning would have been around for a while.
is there any followup?
Scary stuff....
I hope someday all mediocre jobs get automated, so humans get more time to be thinkers, we really need some more good thinkers like Jeremy.
We live in scary times.
Amazing: all of the old-schooled informatics and mathematics experts do not clarify: it was called statistical learning at first - and what the meaning of the word "learning" is in "machine learning". Is it similar to "this child is learning Euclidean geometry"? Sadly, some believe so...
I see a paradox here, do we want computers that have or haven't got simulations of emotion. On the one hand emotion is necessary to avoid 'sociopath AI' on the other hand we don't to discuss our abusive phone-relationship with our phone, or have a car with moodswings.
Further, there is the question of whether there will be a tipping point or stopping point in the simulation of humans / capability of the AI. So ftl seems impossible, what arguments can be given regarding that consciousness-simulation is not also impossible - maybe we simulate the physical the basis but what if the supervenience is impossible to model?
It's going to be freaky once the machines start claiming to be sentient.
Those images of the human and the cat were the creepiest things I'd ever seen.
unsupervised clustering algorithm and bayes theorm
stuff this, off to the pub
I remember my first neural network and having to change all the numbers to fit between 0 and 1. But it worked. Haha.
wow that was epic
inspirational
i do not fear it. i welcome it!
Surely, computers would also be better at gathering the data and would at best only require some physical assistance at administering treatment...
It will hopefully one day free the human race so we can focus on the Universe. Billions of us exploring the known universe and letting the machines taking care of the day to day.
+Crazyman1212 That assumes the machines wouldn't be vastly superior at exploration. Given how fragile humans are that seems unlikely.
Unlikely, because only the rich and powerful will be able to take advantage of this intelligence. The rest of us will be reduced to peasants because there are no jobs and no mercy from our wealthy rulers. It will be like feudalism all over again.
rob062388 The problem is that well implemented intelligence effectively generates infinite resources (from a human perspective at least), so it doesn't really matter how unfairly large the wealthy's slice of the pie is.
Also keep in mind that this scenario is fairly slow and kind of makes no sense.
You really should read Nick Bostrom's book Superintelligence. It's sort of the go to reference for theories regarding superintelligence, and is extremely comprehensive and well researched.
"The problem is that well implemented intelligence effectively generates infinite resources" - what if I want to own 100000 jumbo jets? Will the robots make them for me?
kapsi A technological singularity could easily make 100,000 jumbo jets for everybody, all it needs are the raw elements for its grey goo to assemble it from.
What is the software used to classify the images?
i think imagenet
Im working on a scratch and sniff app for food menu’s for the haves.
Ask the computers how we should adjust our social structure...
Samuel Foreman your profile picture... has it got to do with a cue ball?
It won't be long now, until all of those movies come true. We will design and create a computer that will then make improvements to itself and realize that the only thing holding it back, is us.
There was a time when human would do the cool part of the job and the machine the boring one. Tomorrow, computer will be doing the extraordinary part of the job and human the boring one... Maybe this is what "in the future computers will enslave humans" will actually look like... Fascinating and altogether creepy, indeed.
Well, I was thinking about all of these and even beyond where computers will help us to formulate new physics formulas using a hypothetical theory that we created and so on. But we have to be warned about A.I cause it could be really dangerous and fatal to humanity.
what program is he using to classify the cars? 3:18
A program that was written by the people who use it. A program that no one else knows about because it isn't commercial software or freeware. Like he said, the software uses Deep Learning which you can learn if you take the time to learn how to program and learn machine learning from the many resource that can be searched on the internet.
This has been coming for a while now, and the problem remains the same - we can make brains out of machines, but we can't give them muscles of their own. The robotic actuators are late in coming, and for a while still we'll have to play that part. The machines will etch out the plan, we'll execute it. Hopefully we'll soon have actuators to give these new brains direct access to reality.