By what criteria would we recognise a theory of AI?

Question

By what criteria do we judge as to whether we have understood consciousness on a theoretical basis?

One criteria might be the construction of AI (but this can't be the whole story as everyday new conscious's are born - ie birth). Still, one cannot experiment (ethically) on consciousness; and one notes how our understanding of how consciousness is constituted has been enlargened by pathologies of the mind or brain; but one can experiment (presumably ethically) on AIs; which should bring some additional insight.

The obvious criteria is that of the Turing Test; however, important as this test is, to a large extent it misses the main issue; what the Turing Test discovers is how well can one simulate an intelligence. Simulation isn't the thing itself, and though there is a level of disagreement/controversy about the notion of 'inner life' or qualia - to my mind its of the essence for this question.

The Turing Test puts in a sharp form the test for a construction of an AI.

The sharper question is how does one know one has consciousness-in-itself or a simulation?

What other criteria are possible, necessary or sufficient?

Or is the horizon of this criteria infinite (ie impossible).

To go back to human beings, we denote ourselves by consciousness, by the cogito, and to others, by some principle of continuity or resemblence (I am human, you are human; I am conscious, you are conscious); which can be put in syllogistic form; but the essence, surely of our recognition of consciousness in others is continuity, or even more basically, resemblence.

Answer 1

I think this is a misleading formulation. What do you mean by 'theory'? It seems to me that we have an overabundance of theory, and strong pseudo-paradigmatic camps that resist closure because they are attached to 'pet notions' and resistant to any alteration of vocabulary that might increase the ability of any proposition to be falsified.

What we need is faith that a falsifiable complex of definitions of things like consciousness, intention, choice, etc, will be taken seriously by anyone but its originators and a small camp around them. Until we have that, advancing actual theories is pointless.

Three camps in particular are resisting progress:

• Subjectivists are worried that defining qualia in terms of memory and perception (with which they would have been just fine two generations ago, since it was a big part of their original approach to the problem) reduces tons of closely-held biases to something falsifiable by physical testing.

• Physicalists are worried that creating a real and stable distinction between thought and computation reopens the question of whether life is something more than a mechanical process (which only they consider closed to begin with.)

• Traditional Idealists are worried that allowing such definitions to converge will remove the already established uses borrowed from religious traditions that avoid specificity and make cover for magical thinking.

All told, I do not think that the issue here is one of difficulty, it is one of will. We as humans do not want to know the answer to this question, unless it is the one we already know and love. And there is not just one of those answers.

Answer 2

It is hard to say for sure, since we generally don't set goals for what an understanding looks like, but instead say in retrospect whether we've got a good enough one to satisfy us. (This may involve both advancing our knowledge and changing our minds about what is satisfactory.)

However, I would say that most people feel we have a satisfactory theory of life--that is, what the essential difference is between the living and non-living, and an extensive though far from complete understanding of how "living" works.

Using this as an analogy, but with the difficulty that we have very few examples of different kinds of consciousness, I am rather doubtful that we'll gain this understanding until we can create a synthetic consciousness (assuming that this is possible and that we ever can) and probe the boundaries of it by altering it. When our theories of what happens to synthetic and our own consciousnesses under various conditions seem to produce reasonably reliable results in many scenarios, we'll probably start to say we understand it (as we do in many other fields involving the study of something really complicated).

Alternatively, we might find that there are structural parallels between humans and other animals and that what we call "consciousness" seems inseparable from those (presumably neural) structures; and eventually through work on animal models come to understand the boundaries and properties of consciousness (relying heavily on the architectural similarity between their brains and ours, as we do on the aliveness of our cells vs. yeast).

Not a very satisfying answer, but I do not see that we could say much more presently.

Answer 3

The Turing Test is a thing of the past and obsolete.
Any modern simulated artificial intelligence can/could pass it.

The answer is in your question, we need to look at the definition of artificial intelligence. An artificial intelligence is an intelligence created by humans - as opposed to natural. I won't elaborate what intelligence is, but from a computer program point of view it would mainly mean problem solving abilities on its own, given a set of data and rules.
I don't know if true AI can ever be achieved. Maybe with a better understanding of the brain model. Future will tell.

Consciousness is meaningless for AI, because it doesn't need it in order to successfully perform the tasks assigned (so far at least... hello true AI...).
It's all about data and its processing/transformation in order to achieve what they've been built for.