Evolution News and friends claim that Artificial Intelligence is necessarily a fake, like listening to a recorded telephone message: pretending to be something that it is not. and only a fool would think it was real in any way.
https://evolutionnews.org/2019/04/artificial-intelligence-as-a-magic-act/
There is some truth in this. We all know about the Mechanical Turk. Even today, most applications of AI have very little depth to them. It doesn't take long talking to a chatbot before you realize you are not going to get any more insight out of it. It does not understand what you are saying, so unless someone programmed with lots of correct answers to frequently-asked questions, you are not going to get anything more.
However, many of us have had this experience talking to humans too, and that is a problem: there is no test which could ever clearly separate humans from computers. People keep setting tests, but computers keep on passing them. There was a time when people said, "Computers will never be able to play chess. It is too subtle". Then they could. There was a time when people said, "Computers will never be able to recognize music. It is too spiritual". Now they can. People said, "Computers will never be able to drive cars. That problem is too complex, with too many unforeseen contingencies." Now they can, and better than most humans.
In the face of steadily losing ground in this way, how can AI deniers respond? They assert that artificial intelligence is nonetheless fake. They tell us to disbelieve the evidence of our eyes. They tell us it is a magic trick. Well, is it?
To be fair, AI is often a gust of bluster. It is a currently-trendy fad (like blockchain), and however impressive it is, it still falls far short of human intelligence. Sometimes machine learning concepts work, but only accidentally, perhaps by latching onto some proxy variable that happens to vary with the variable that a human would see is the real causal connection. But again, this happens with humans too. One example is a cargo cult. It is not a sign of unintelligence, just a sign of being able to make false inferences and be confused. Of course AI is sometimes fake. But sometimes so are humans.
The question is, "Is AI inherently fake?"
Let us get to the bottom of this (do a reductio) with a thought experiment. A mad scientist builds a machine in his lab that perfectly mimics the neuronal activity of the human brain, but using silicon instead of wet organic materials. It is huuuuuge and it is slooooow but nevertheless he connects it to an electronic body with cameras and sensors. He finds it learns to interact in the world just like a human child.
Do you think this is impossible? Why do you think that is impossible? Have you tried it? But it doesn't really matter for our thought experiment because the AI critic has a ready response: that is not real intelligence, that is just a simulation of intelligence.
Ok, but for his next research project, having now realized the technical limits of silicon, he builds a brain out of wet organic materials. Essentially he makes a human brain instead of growing it naturally. This time around he even builds a whole human body, piece by piece. The finished machine learns and it too becomes like a human child. Is this thing truly intelligent? Is it a zombie?
Many people would instinctively and sensibly say, Yes! this is true intelligence. No! it is not a zombie, and then Don't ever do that again! and maybe prosecute the Mad Scientist for playing God. You see, we can no longer pretend that this thing is merely a simulation. It no longer just talks like a duck, and walks like a duck; it dissects like a duck too. It really is a duck (I mean a human, but you understand).
This is what Deep Learning specialists are trying to do. Everyone knows that human intelligence is strongly dependent on the brain, so these engineers thought, naturally and reasonably, that if we want to create human-like intelligence (which we do) then maybe we should try to imitate the amazing human brain. For a long while, they had little success, but in recent years, this field has begun to make progress again. We now see machines that suffer optical illusions (google it), and seem to work by inscrutable intuition (google it) because the patterns they detect are so subtly complex. Progress may well slow down again in a few years; brains are not easy to emulate, but why do we think they will not ultimately succeed and make a machine that rivals a human for apparent intelligence?
Is this fake intelligence? Or is it real but nascent intelligence re-created in silico.
And the irony is, all of these people (EN and friends) believe that God created Adam out of dust just like this. Not out of another living thing, or out of magic materials, but out of ordinary materials, namely dust or clay, and then breathed life into him (and also breathed life into all the animals).
My question to Evolution News and friends is this: What does all the evidence say? All the evidence so far seems to say that intelligence can be made.
But they have already given their final response: The evidence does not matter, you fool! It cannot be done in principle!
This is a kind of fundamentalism if ever I saw one.
https://evolutionnews.org/2019/04/artificial-intelligence-as-a-magic-act/
There is some truth in this. We all know about the Mechanical Turk. Even today, most applications of AI have very little depth to them. It doesn't take long talking to a chatbot before you realize you are not going to get any more insight out of it. It does not understand what you are saying, so unless someone programmed with lots of correct answers to frequently-asked questions, you are not going to get anything more.
However, many of us have had this experience talking to humans too, and that is a problem: there is no test which could ever clearly separate humans from computers. People keep setting tests, but computers keep on passing them. There was a time when people said, "Computers will never be able to play chess. It is too subtle". Then they could. There was a time when people said, "Computers will never be able to recognize music. It is too spiritual". Now they can. People said, "Computers will never be able to drive cars. That problem is too complex, with too many unforeseen contingencies." Now they can, and better than most humans.
In the face of steadily losing ground in this way, how can AI deniers respond? They assert that artificial intelligence is nonetheless fake. They tell us to disbelieve the evidence of our eyes. They tell us it is a magic trick. Well, is it?
To be fair, AI is often a gust of bluster. It is a currently-trendy fad (like blockchain), and however impressive it is, it still falls far short of human intelligence. Sometimes machine learning concepts work, but only accidentally, perhaps by latching onto some proxy variable that happens to vary with the variable that a human would see is the real causal connection. But again, this happens with humans too. One example is a cargo cult. It is not a sign of unintelligence, just a sign of being able to make false inferences and be confused. Of course AI is sometimes fake. But sometimes so are humans.
The question is, "Is AI inherently fake?"
Let us get to the bottom of this (do a reductio) with a thought experiment. A mad scientist builds a machine in his lab that perfectly mimics the neuronal activity of the human brain, but using silicon instead of wet organic materials. It is huuuuuge and it is slooooow but nevertheless he connects it to an electronic body with cameras and sensors. He finds it learns to interact in the world just like a human child.
Do you think this is impossible? Why do you think that is impossible? Have you tried it? But it doesn't really matter for our thought experiment because the AI critic has a ready response: that is not real intelligence, that is just a simulation of intelligence.
Ok, but for his next research project, having now realized the technical limits of silicon, he builds a brain out of wet organic materials. Essentially he makes a human brain instead of growing it naturally. This time around he even builds a whole human body, piece by piece. The finished machine learns and it too becomes like a human child. Is this thing truly intelligent? Is it a zombie?
Many people would instinctively and sensibly say, Yes! this is true intelligence. No! it is not a zombie, and then Don't ever do that again! and maybe prosecute the Mad Scientist for playing God. You see, we can no longer pretend that this thing is merely a simulation. It no longer just talks like a duck, and walks like a duck; it dissects like a duck too. It really is a duck (I mean a human, but you understand).
This is what Deep Learning specialists are trying to do. Everyone knows that human intelligence is strongly dependent on the brain, so these engineers thought, naturally and reasonably, that if we want to create human-like intelligence (which we do) then maybe we should try to imitate the amazing human brain. For a long while, they had little success, but in recent years, this field has begun to make progress again. We now see machines that suffer optical illusions (google it), and seem to work by inscrutable intuition (google it) because the patterns they detect are so subtly complex. Progress may well slow down again in a few years; brains are not easy to emulate, but why do we think they will not ultimately succeed and make a machine that rivals a human for apparent intelligence?
Is this fake intelligence? Or is it real but nascent intelligence re-created in silico.
And the irony is, all of these people (EN and friends) believe that God created Adam out of dust just like this. Not out of another living thing, or out of magic materials, but out of ordinary materials, namely dust or clay, and then breathed life into him (and also breathed life into all the animals).
My question to Evolution News and friends is this: What does all the evidence say? All the evidence so far seems to say that intelligence can be made.
But they have already given their final response: The evidence does not matter, you fool! It cannot be done in principle!
This is a kind of fundamentalism if ever I saw one.
Comments
Post a Comment