4

It is, of course, debatable whether a Strong AI even can exist. After all, philosophical arguments about the Chinese Room and Philosophical Zombies abound. However, it is possible to imagine a world where "Strong AI" is possible and yet people adhere to a philosophical argument that denies the existence of "Strong AI".

It is also possible to imagine a world where humans fall prey to the AI Effect. According to Wikipedia,

The AI effect occurs when onlookers discount the behavior of an artificial intelligence program by arguing that it is not real intelligence.

And while it is true that today's onlookers are correct that 'real intelligence' (which I'm calling "Strong AI" here) does not exist currently, if these onlookers will perpertually "discount the behavior of an artificial intelligence program", that there's a real possibility that one day, these onlookers would be wrong...that "real intelligence"/Strong AI did occur.

This is my fear:

  • Strong AI is possible.
  • Humans build Strong AI for some trivial task (for example: building paperclips).
  • Humans immediately assume that this Strong AI isn't really "strong" intelligence; it's just really complicated algorithm. They invoke a few philosophical arguments and move on with their lives.
  • The Strong AI doesn't care whether humans recognize it as a Strong AI, most likely because it was programmed to build paperclips, and trying to convince humans that it is a 'Strong AI' doesn't actually help it in its goal of building paperclips. So it says nothing.
  • Humans keep on debating over the existence of a Strong AI, long after they invented it.

Could this scenario occur? Or would it is possible, as soon as Strong AI exist, that humans immediately come to a consensus that it does exist?

  • There seems to be no way to accept your scenario without presupposing that "strong" intelligence could be the same as real intelligence. I don't see how such an assumption could be conceivable, so I also don't see how your scenario could be conceivable. – user3017 Jun 23 '16 at 14:20
  • @PédeLeão That assumption was explicitly stated in the question. – Era Jun 23 '16 at 14:40
  • @Era. Exactly. It's necessary to presuppose something inconceivable to address the question. He asserted the following: "However, it is possible to imagine a world where 'Strong AI' is possible..." But since that's not true, it does make sense to proceed as if it were true. – user3017 Jun 23 '16 at 15:10
  • @PédeLeão Conceivability isn't a prerequisite for a hypothetical state of affairs; if it were then proof by contradiction wouldn't work. (It's not possible that you can conceive of a logical contradiction being true.) – Era Jun 23 '16 at 16:02
  • @Era. I agree. But it's still inconceivable, so it doesn't make sense to proceed as if it were conceivable. – user3017 Jun 23 '16 at 16:36
2

First a couple of comments:

  • Humans keep on debating over the existence of a Strong AI, long after they invented it.

Some argue that this already happening: A pocket sized computing device (namely the iPhone equipped with Siri) with which you can have conversations and ask for directions, or a large scale system (Google) which can give us answers to questions that were asked using unstructured natural language queries, would have been considered successful AI (if not strong by John Searle's definition) by most CS researchers in the 1940s and 1950s.

  • Humans build Strong AI for some trivial task (for example: building paperclips) [...] The Strong AI doesn't care whether humans recognize it as a Strong AI, most likely because it was programmed to build paperclips, and trying to convince humans that it is a 'Strong AI' doesn't actually help it in its goal of building paperclips. So it says nothing.

This is by definition NOT Strong AI, it can only be strong AI if it has general problem solving abilities regardless of context. See this discussion for example by Hilary Putnam (it is somewhere toward the last third of the video): What distinguishes the human brain from even the most powerful current computer is that the human brain can solve problems it was not designed to solve. Assuming that human brains were programmed by evolution for specific purposes - hunting, reproducing, avoiding predators, etc ... - they are now capable of solving problems far beyond their original design, such as mathematical models for cosmology or composing and preforming symphonies, etc...you are explicitly defining an AI which doesn't do that, so it can't be Strong AI.

That being said, your question in general still holds, and I will address two separate aspects of it

  • Can we recognize the existence of a Strong AI after it is created?

From one point of view this is really a technical question: What tests do we have for recognizing Strong AI? There is of course the Turing test and its variations. The Turing test has its problems. There have been since then other proposals for tests, such as Winograd Schemas, The Lovelace test, and others. Your question in one way reduces to a technical discussion of which one of these tests is the right one, and then how to convince people that this test is, and is one to be resolved by further research in cognitive science more so than a philosophical discussion.

Keep in mind that even once the philosophical and scientific communities arrive at a consensus on what a legitimate test for strong AI is, and then computer engineers proceed to build such a strong AI, there will always remain people that are unconvinced. After all, the scientific consensus is that the Earth is billions of years old, yet many people still believe that it is only 6000 year old or so.

Of course in the case of strong AI, there is a chance that such an AI will, in true Matrix-Skynet fashion, enslave us and become our overlord, and most people would then be convinced the hard way that Strong AI is possible.

  • Or would it is possible, as soon as Strong AI exist, that humans immediately come to a consensus that it does exist?

From another point of view, the question is a metaphysical one. What you need to understand about John Searle's Chinese Room argument and the Philosophical Zombie argument is that they are not technical arguments against Strong AI, but metaphysical ones. What such arguments try to prove is that no matter what a computer or a robot can achieve, it will never have mind, only the appearance of one.

John Searle, with his Chinese Room argument, is trying to how that a digital computer can only simulate human intelligence, because it will always lack an understanding of the information that it is processing, it has no way of knowing the meaning of the symbols it processes. He calls this the syntax is not semantics argument. Searle does concede that if we were to produce a biological artificial intelligence (using cloning or some other biological replication technology), such an entity would indeed posses strong AI.

Similarly Zombie arguments try to show that something can have all of the outward appearance and behavior of an intelligent human, yet it can still be missing the first person subjective experience that characterizes humans. Such a Zombie would talk and walk like a normal person, but there would be nothing inside that it is experiencing the world or experiencing emotions. Presumably, human like AI would fall into this category of being, and therefore would not be considered to have a real mind.

  • Not to argue the point, but I'm curious about how you understand the phrase "regardless of context." From my understanding, it still remains a problem with neural networks that they are poorly suited for tasks for which they aren't trained or programmed. – user3017 Jun 23 '16 at 17:37
  • @PédeLeão to use the OP's example, an AI would be working "regardless of context" if it started using its paper clip building abilities to make origamis, or it if found itself unable to communicate using text or sound, so it started bending the paper clips into letter and words in English to communicate. Right now this is a problem not just for neural networks but for just about any machine learning algorithm - they cannot generalize beyond a specific domain. Deep Leaning networks are supposed to be a step in the right direction, but still very far from achieving anything like this. – Alexander S King Jun 23 '16 at 17:48
3

Yes.

The most obvious and quick answer to your question is: Do you know for sure that there is no Strong AI in existence right now? If your answer is no (which it should be), then you must also believe that it is possible for Strong AI to exist without us being certain that it can exist.

Another, more long-winded one:

Based on the constraints of your scenario, this happens in a world in which Strong AI is possible. In such a world, the entirety of intelligence and consciousness are nothing more than physical processes. IE: The physicalist approach to the mind/body problem is the correct one. I suppose one could argue that Strong AI doesn't require a mind to exist, but I feel like if you redefine Strong AI that way you're not talking about what we normally think of as Strong AI.

If this is the case, then there is no fundamental difference between the thoughts of a person and the actions of a computer, we're just programmed differently. While the effects of the algorithm, or the effects of human thought are difficult or impossible to predict with precision, the fact remains that they are purely the result of physical processes. The system, while perhaps very very complex, ultimately boils down to interactions between molecules.

So the question of whether we will "realize" that Strong AI is possible is unrelated to the truth-value of whether it is possible, our "realizations" are merely a complex interaction of molecules. There is no reason to suspect that the existence of AI will send out molecules/waves of light that will interact with the cells in our brains in such a way that we will "realize" that it is possible to create AI.

0

No, we would not be able to recognize very strong AI. We would think it's "just another person" until it gets into an "accident," and we get to see its internal components.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.