4 ms·

> Rodney Brooks famously condemned simulation and games positing that AI can only develop in an embodied agents. This is a religious view and preposterous. It'

> Rodney Brooks famously condemned simulation and games positing that AI can only develop in an embodied agents. This is a religious view and preposterous. It's a milder form of the (surprisingly common) belief that only humans can be intelligent. That claim is false even if embodied cognition is true (from Wikipedia): "Embodied cognition is the theory that many features of cognition, whether human or otherwise, are shaped by aspects of the entire body of the organism"

deepnet 9y ago
That is quite some condemnation :) I am certainly interested and hope you will flesh out your positions. Certainly Brooks ran into Moravecs paradox, that walking is the hardest form of thinking and playing chess and planning less so. And so Brooksian embodied robotics never got to the symbolic level. Contrary to your anthropocentric claim, Brooks (and followers such as Tilden) were biomimetically inspired by the intelligence of insect locomotion and 'intelligent' engineering like the underslung chassy of spiders and beetle. I do agree with Brooks that AI will not emerge in simulation due to the lack of real stochasticity and resolution. [/edit] I will temper my stridency and say that it is less likely, my initial absolutist position was merely excitement and really simulation has a big role to play as imagination and plan testing. We live in a world of perception and as [Richard] Gregory says "perception is only a hypothesis, based on our prior knowledge and updated by reality". For instance the Atari benchmarks appears random but the randomness is Mersenne Twisters with few seeds and most expert (human) arcade players learn the patterns. That is not to say that Mnih wt al.s Nature paper wasn't groundbreaking and their Deep-Q RL is certainly insightful, promising and useful but (IMHO) the AI 'stack' and LeCun's Predictive Learning will not come about outside of a real robot acting in the world and using raw sensor data to power actuators directly. My position is not entirely Brooksian. I follow Esther Thelen in the view that a being ( a brain and a body ) needs to learn to be by being and moving and feeling and touching. I do apologise if my brevity and excitement appears rude, AI could well emerge in computers running simulations and in fact I think that simulation can play the role of planning and imagination but it will need to be grounded in a real physical agent - a hybrid approach will bear the most fruit.
- paulsutter 9y ago
  A friend of mine awoke from a coma a month before he could move or speak. He could hear the nurses, they thought he was a vegetable. Are you saying the nurses were right? Without the ability to move, he wasn't actually intelligent? Edit: I see your edit now, sure I like your restatement "simulation has a big role to play as imagination and plan testing"
  - deepnet 9y ago
    You edited your comment after I replied. I am sorry about your friend, I hope they are better. I am not saying that at all. Daniel Dennet is better placed to speculate on matters like that. [/edit] perhaps there is some common ground. Brooks' Behaviour-Based position, though useful at the time, threw out the symbolic AI with the bathwater of computational complexity that robotics was quagmired in then.
Animats 9y ago
That was Brooks' reaction against the logic-based AI of the 1980s. Around 1990 I went to a talk where Brooks was plugging the Cog project he was starting.[1] Brooks had some success with purely reactive insect-level AI, and was trying to make the big jump to human-level AI. I asked him, why not try for mouse-level AI. That might be within reach. He said "Because I don't want to go down in history as the man who developed the world's greatest robot mouse." This is a classic problem with AI researchers. Somebody gets a good result, and then they start thinking strong human-level AI is right around the corner. AI went through this with search, planning, the General Problem Solver, perceptrons, the first generation of neural networks, and expert systems. Then came the "AI winter", late 1980s to early 2000s, when almost all the AI startups went bust. We're seeing some of it again in the machine learning / deep neural net era. This time looks more promising, partly because we can throw more compute power at the problem. Many of the ideas in machine learning and neural nets are old, and are so inefficient that they were hopeless until people could beat on them with racks of GPUs. The big difference this time is that AI is profitable. The field used to be tiny - maybe 20-30 people at MIT, Stanford, and CMU, with a few small groups elsewhere. Now there are hundreds of thousands of researchers, and profitable applications. (I went through Stanford just as the expert system boom was collapsing and the "AI winter" was beginning. I met most of the big names from the logic-based AI era. It was kind of sad.) [1] https://en.wikipedia.org/wiki/Cog_(project) https://en.wikipedia.org/wiki/Cog_(project)
- cbames89 9y ago
  The hundreds of thousands of researchers number, is that just a guess? If it isn't I'd love to see the source.
jbooj 9y ago
This research (https://blog.openai.com/robots-that-learn/ https://blog.openai.com/robots-that-learn/) is extremely relevant to Rodney Brooks' condemnation and an amazing achievement in my opinion.

deepnet 9y ago

That is quite some condemnation :) I am certainly interested and hope you will flesh out your positions. Certainly Brooks ran into Moravecs paradox, that walking is the hardest form of thinking and playing chess and planning less so. And so Brooksian embodied robotics never got to the symbolic level. Contrary to your anthropocentric claim, Brooks (and followers such as Tilden) were biomimetically inspired by the intelligence of insect locomotion and 'intelligent' engineering like the underslung chassy of spiders and beetle. I do agree with Brooks that AI will not emerge in simulation due to the lack of real stochasticity and resolution. [/edit] I will temper my stridency and say that it is less likely, my initial absolutist position was merely excitement and really simulation has a big role to play as imagination and plan testing. We live in a world of perception and as [Richard] Gregory says "perception is only a hypothesis, based on our prior knowledge and updated by reality". For instance the Atari benchmarks appears random but the randomness is Mersenne Twisters with few seeds and most expert (human) arcade players learn the patterns. That is not to say that Mnih wt al.s Nature paper wasn't groundbreaking and their Deep-Q RL is certainly insightful, promising and useful but (IMHO) the AI 'stack' and LeCun's Predictive Learning will not come about outside of a real robot acting in the world and using raw sensor data to power actuators directly. My position is not entirely Brooksian. I follow Esther Thelen in the view that a being ( a brain and a body ) needs to learn to be by being and moving and feeling and touching. I do apologise if my brevity and excitement appears rude, AI could well emerge in computers running simulations and in fact I think that simulation can play the role of planning and imagination but it will need to be grounded in a real physical agent - a hybrid approach will bear the most fruit.

paulsutter 9y ago
A friend of mine awoke from a coma a month before he could move or speak. He could hear the nurses, they thought he was a vegetable. Are you saying the nurses were right? Without the ability to move, he wasn't actually intelligent? Edit: I see your edit now, sure I like your restatement "simulation has a big role to play as imagination and plan testing"
- deepnet 9y ago
  You edited your comment after I replied. I am sorry about your friend, I hope they are better. I am not saying that at all. Daniel Dennet is better placed to speculate on matters like that. [/edit] perhaps there is some common ground. Brooks' Behaviour-Based position, though useful at the time, threw out the symbolic AI with the bathwater of computational complexity that robotics was quagmired in then.

Animats 9y ago

That was Brooks' reaction against the logic-based AI of the 1980s. Around 1990 I went to a talk where Brooks was plugging the Cog project he was starting.[1] Brooks had some success with purely reactive insect-level AI, and was trying to make the big jump to human-level AI. I asked him, why not try for mouse-level AI. That might be within reach. He said "Because I don't want to go down in history as the man who developed the world's greatest robot mouse." This is a classic problem with AI researchers. Somebody gets a good result, and then they start thinking strong human-level AI is right around the corner. AI went through this with search, planning, the General Problem Solver, perceptrons, the first generation of neural networks, and expert systems. Then came the "AI winter", late 1980s to early 2000s, when almost all the AI startups went bust. We're seeing some of it again in the machine learning / deep neural net era. This time looks more promising, partly because we can throw more compute power at the problem. Many of the ideas in machine learning and neural nets are old, and are so inefficient that they were hopeless until people could beat on them with racks of GPUs. The big difference this time is that AI is profitable. The field used to be tiny - maybe 20-30 people at MIT, Stanford, and CMU, with a few small groups elsewhere. Now there are hundreds of thousands of researchers, and profitable applications. (I went through Stanford just as the expert system boom was collapsing and the "AI winter" was beginning. I met most of the big names from the logic-based AI era. It was kind of sad.) [1] https://en.wikipedia.org/wiki/Cog_(project) https://en.wikipedia.org/wiki/Cog_(project)

cbames89 9y ago
The hundreds of thousands of researchers number, is that just a guess? If it isn't I'd love to see the source.

jbooj 9y ago

This research (https://blog.openai.com/robots-that-learn/ https://blog.openai.com/robots-that-learn/) is extremely relevant to Rodney Brooks' condemnation and an amazing achievement in my opinion.