3 ms·

Body Language and Machine Learning

38 points by luu 6y ago

YeGoblynQueenne 6y ago
>> Anyway, the other day I was thinking about how this is an example of machine learning. I can see what the author means by this, of course- that the way that they have learned to predict the behaviour of other cars on the road etc. is _reminiscent of_ the way that statistical machine learning systems work. The author clearly doesn't mean that learning by a human really _is_ machine learning, because obviously a human is not a machine. There is however a danger, in anything to do with artificial intelligence (the broader discipline to which machine learning belongs) when extracting conclusions from superficial similarities between the behaviours of AI systems and humans. For example, it is easy to assume that, because a human mind can make predictions about the behaviour of other road users and a machine learning system (driving a car) can make predictions about the behaviour of other road users, the human and the system are "doing the same thing", or, even more egregiously, "doing it the same way". In particular when comparing statistical machine learning with human learning it is easy to see that, beyond the superficial similarities, there are significant differences that suggest each system operates in a very different manner than the other. For one example, a human who learns to predict the behaviour of other road users can also give an account of this behaviour to a third party at a later point in time without additional "training" on the task, while a machine learning system that can drive a car cannot also generate text to explain its decisions, etc. Whatever "models" humans build of the world, they are very different than statistical machine learning models. >> P.S. The above are my conjectures and loose thoughts. There could well be a literature on all of this. If so, feel free to inform me about it in the comments. Indeed, there is. For an early introduction to the field of machine learning research I would recommend to the author Tom Mitchell's textbook "Machine Learning". Or, going even further back in time, Ryszard Michalski's "Machine Learning: an Artificial Intelligence Approach". For a shorter introductory read Tom Mitchell's "Generalisation as Search" paper is a good introduction to some early ideas on the theoretical underpinnings of machine "learning". These are old papers and textbooks, but I believe this is exactly the kind of literature that should be brought to the attention of experts from other disciplines who wish to understand the field. Machine learning looks very different with the perspective of a body of literature going back to the early '70s, than it does with the perspective of a few articles in the lay press waxing lyrically about the progress achieved in the last few years.

38 points by luu 6y ago

>> Anyway, the other day I was thinking about how this is an example of machine learning. I can see what the author means by this, of course- that the way that they have learned to predict the behaviour of other cars on the road etc. is _reminiscent of_ the way that statistical machine learning systems work. The author clearly doesn't mean that learning by a human really _is_ machine learning, because obviously a human is not a machine. There is however a danger, in anything to do with artificial intelligence (the broader discipline to which machine learning belongs) when extracting conclusions from superficial similarities between the behaviours of AI systems and humans. For example, it is easy to assume that, because a human mind can make predictions about the behaviour of other road users and a machine learning system (driving a car) can make predictions about the behaviour of other road users, the human and the system are "doing the same thing", or, even more egregiously, "doing it the same way". In particular when comparing statistical machine learning with human learning it is easy to see that, beyond the superficial similarities, there are significant differences that suggest each system operates in a very different manner than the other. For one example, a human who learns to predict the behaviour of other road users can also give an account of this behaviour to a third party at a later point in time without additional "training" on the task, while a machine learning system that can drive a car cannot also generate text to explain its decisions, etc. Whatever "models" humans build of the world, they are very different than statistical machine learning models. >> P.S. The above are my conjectures and loose thoughts. There could well be a literature on all of this. If so, feel free to inform me about it in the comments. Indeed, there is. For an early introduction to the field of machine learning research I would recommend to the author Tom Mitchell's textbook "Machine Learning". Or, going even further back in time, Ryszard Michalski's "Machine Learning: an Artificial Intelligence Approach". For a shorter introductory read Tom Mitchell's "Generalisation as Search" paper is a good introduction to some early ideas on the theoretical underpinnings of machine "learning". These are old papers and textbooks, but I believe this is exactly the kind of literature that should be brought to the attention of experts from other disciplines who wish to understand the field. Machine learning looks very different with the perspective of a body of literature going back to the early '70s, than it does with the perspective of a few articles in the lay press waxing lyrically about the progress achieved in the last few years.