Why it's harder for AI to open doors than play chess | Pulkit Agrawal | TEDxMIT
ฝัง
- เผยแพร่เมื่อ 8 เม.ย. 2023
- Today's AI systems can beat the best human chess or Go players. However, they fail to perform tasks involving physical intelligence that humans don't even think about such as walking, climbing stairs or opening doors. Things that are easy for humans, are often very hard for machines and vice-versa, an observation often known as the Moravec's paradox. Human intuition of what is easy and what is hard gets in the way of building truly intelligent AI systems and has been a major lesson of AI research in the past seventy years. Using all the data on the internet, we have made impressive progress in reasoning and language understanding. However, there is no such data repository for learning physical intelligence and evolution spent 50,000x more time perfecting physical intelligence as opposed to language understanding. In this talk, learn why is physical intelligence so hard and how do we get there?
AI, Brain, Decision making, Evolution, Intelligence, Machine Learning, Robots, Technology Dr. Pulkit Agrawal is Steven and Renee Finn Chair Assistant Professor in the Department of Electrical Engineering and Computer Science at MIT. He earned his Ph.D. from UC Berkeley and co-founded SafelyYou Inc. His research interests span robotics, deep learning, computer vision, and reinforcement learning. Pulkit completed his bachelor's from IIT Kanpur and was awarded the Directors Gold Medal. His work received the Best Paper Award at Conference on Robot Learning 2021 and Best Student Paper Award at Conference on Computer Supported Collaborative Learning 2011. He is a recipient of Sony Faculty Research Award, Salesforce Research Award, Amazon Research Award, Signatures Fellow Award, Fulbright Science and Technology Award, Goldman Sachs Global Leadership Award, etc. His work has appeared multiple times in MIT Tech Review, Quanta, New Scientist, NYPost, etc. This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at www.ted.com/tedx - วิทยาศาสตร์และเทคโนโลยี
When terminator machines rise, just close the door. Problem solved.
Lol
Mr. Agrawal really wants terminators to be able to open the doors
😂😂😂😂
Just add water see how they go 😂
they will probably just pull down the entire building instead
Really interesting during this AI hyped period .......
Тема розкрита і стала зрозумілішою. Можливо, прогрес прискориться після зміни моделі обробки інформації в ШІ. Треба щось змінити принципово.
Перші люди, що мріяли літати, будували крила як у птахів. Сучасні літаки з реактивними двигунами і гелікоптери зовсім не схожі на птахів. Але вони літають ефективно.
Ми вже винайшли нейромережі. Думаю, ми недалеко від нових відкриттів, що в сукупності створять прорив конкретно в області ШІ з фізичною компонентою в реальному житті. Роботи майбутнього будуть відчиняти різноманітні замки і двері швидше за людей
Thanks Pulkit for shareing!
Sometimes a door, is a jar... .
Chess; is just Chess.
Fascinating review of the evolution of human intelligence and how AI has developed in the past 35 years. Moravec's Paradox is an interesting concept and will apply to AI applications like substituting robots for human soldiers on the battlefield. Well done, Mr Agarwal.
This is a very good presentation.
Something to think about with the little robot walking dogs, it's an entire machine and ai dedicated solely just to walking, it can't do anything else. It needs its entire being dedicated solely to locomotion just to get around, and it's basically the only one that's actually able to do it.
Just moving, nothing else.
Well, yes, but the beauty of robotics and science in general is the ability to combine things. So once they get it walking, running and jumping around flawlessly, they can just add some more processing power and add the disembodied part of AI too. At that point you can have a walking dog which can do your homework :)
Great talk!
Very nice sharing❤❤❤
“We delight in the beauty of the butterfly, but rarely admit the changes it has gone through to achieve that beauty.” - Maya Angelou.
You might have a tiny, wobbly step to celebrate. If so, bravo! You might have already integrated a change. If so, bravo! Even when you’ve ignored things for a little while, notice that right now, you are here.
Give yourself credit. Delight in the beauty of the butterfly and don’t forget to stop to appreciate the changes it has gone through to get there.
Beautiful.
Thanks for sharing.
The only thing missing from the comment is my name.
Isn't it ironic that "physical intelligence," which was once regarded as low-paid menial work for less educated individuals, will maybe become arguably more crucial than the academic abilities of the "mentally intelligent," who were once highly respected and well-paid? (Thanks to AI)
its already is more crucial but since people bring you food from automated mega farms youve got nothing else to do but sit around and cook up law job for yourself.
A non woke, interesting and high quality content Ted Talk? yeeee! 🤗🤗
Epic !!
Mind - blown 🤯
Proud of you an IITian from India
I thought that he was going to say, “When the human told the robot ‘make me dinner’ the robot killed, butchered, and cooked the human.
A healthy mind lives in a healthy body. They need a body, not just a simulation of a body
That's why robots will just bust thru or blast away a door and not worry about opening it to get at the last stragglers of humanity left on Earth.
Oh, and the first people to die will be the smartest people on the planet. They possess the most danger to an AI because they can defeat the AI on that intellectual level.
This was a funny comment ty
Please, next times: time from left to right.
newb here. isnt a door part of a set of standardized physical interfaces? why dont roboticists just take the easy way out and match the door knob to a database of standard interfaces then just use the proper end effector for the job? why are we working so hard to have A.I. "figure it out". Isnt that what the robot does when it plays chess? they build an end effector that can grasp the peice and they use basic inverse kinematics to pick and place the object to the different location? The reasoning of why to move the peice is a different story. when i play chess i just try a bunch of outcomes in my head for the horse and then try outcomes of those outcomes. its hard for me because ive got other things on my mind but robot is easy its software module running just does that in isolation. I dont get why its so hard to open a door. Its like the A.I. people want the robot to figure it all out from scratch instead of kinda sorta hardcoding things as i suggested. plenty of robots do hard things thats how we have all these computers and phones and earphones and microcomponents. robots build those. thats harder than opening a door.
Because opening a door knob is arguably one of the simplest manipulator tasks you can give a mobile robot. You have to walk before you can run, and you have to be able to grip a door knob before you can flip an egg with a frying pan, or any number of non-arbitrary examples. Also, in the chess example, the chess piece manipulation is more for "showmanship" rather than practicality.
Nice 👍 we should mimic biology more
The problem is that mimicing biology tends to significantly increase the amount of compute required (as well as the mechanical design complexity). For example, the human hand is most adequately modeled using a soft end effector with millions of touch sensors spread across the mechanism as well as many independent joints to model each degree of freedom. Good luck solving any nontrivial motion planning problem under this design, and that assumes none of these joint motors have failed (which they almost certainly will when the hand comes into contact with nontrivial forces, and it would only take one failure to severely limit the range of motion of the hand).
Worldpeaceplease 🙏🕊️✌️
Make AI that can Ask question why it can't solve some physical intelligence
First we need an AI to remove accent from speech to make it more comprehensible... and teach people to speak with brevity, not repeating themself all the time.... all the time
I would argue if we even have the cherry, ai can play chess but it has zero understanding of what chess is.
We can build conscious mind but not subconscious mind.
Ai will help in space exploaration
My Heavenly Father made me, I just that's why I can walk and open doors so good!
Cos they don't have a damn arm lol
Pathetic
I don't see the point of this. OK, difficulty for humans and machines are not the same. SO WHAT?!
They improve faster than anything, so it does not matter.
if only you would lose that accent within 3.7 billion years
कौन कौन मानता है माता पिता के चरणों में स्वर्ग है
Evolution is theory. Is funny how folks are so casual with something that's debatable😅.
God loves you and takes care of you for the arrival of this message to you. God is the one who created this universe and He is the one who controls it. The biggest loss that a person loses in this life is to live while he does not know God and the Messenger of Muhammad, the last of the prophets and the Islamic religion, the last of the heavenly religions. From the great intelligence of any person before he He believes in something or does not believe in it, that he read it, study it, and understand it well, and then he has the choice and judgment on it. I advise you to do so before you do not have time for that.