How will we recognize a conscious machine?

Question

How will we recognize a conscious machine (or AI)? Is there any consciousness test? For example, if a machine is aware of its previous experiences, can it be considered conscious?

Answer 1

There are two main subjects you need to look at to understand the problem:

The Turing Test

The Turing test, developed by Alan Turing in 1950, is a test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
_wiki

See also: Turing Test (Stanford Philosophical Dictionary)

There is a linguistic element, which is that intelligence can be interpreted in many legitimate ways. Some maintain that Artificial Intelligence has not yet been achieved; other feel a simple automated switch constitutes the most basic form of AI.

Bear in mind that "intelligence" is distinct from consciousness in the sense of "self awareness", but the generalized Turing Test can also be understood as a gauge of the appearance of consciousness.

This leads to Searle's Chinese Room Argument. I highly recommend reading the Stanford Philosophy link, but the wiki gives a simpler synopsis:

The Chinese room argument holds that a program cannot give a computer a "mind", "understanding" or "consciousness", regardless of how intelligently or human-like the program may make the computer behave.

• The real problem may be, how does one know an algorithm is truly conscious and not merely simulating consciousness?

Philip K. Dick approaches from the opposite direction in Electric Sheep, where one of the conclusions is that "life is life" whether organic or artificial. This might be said to lead to the "Duck Test" for consciousness: "If it looks like a duck, and quacks like a duck, then it's probably a duck." (Dick's philosophy was heavily influenced by Christianity, and his view on artificial life may be though of as radically humanist.)

(A parallel philosophical argument might be "how is the appearance of free will distinct from the actuality of free will?" Free will has been fruitlessly argued about for millennia, but whether the universe is actually deterministic or not, we perceive ourselves to have free will. If the universe turns out not to be strictly deterministic, it wouldn't functionally change anything. To make matters more fuzzy, true randomness is nature is only found at the quantum level and then only within certain models. Quantum uncertainty has been proposed as a basis for free will, but we don't know. Possibly uncertainty is a condition of individuality, which must be subjective in relation to a system or other subjectivities. Nevertheless, how do we know we are not "robots" acting out a pre-determined sequence of actions, which we perceive as decisions?)

I personally very much like the idea of recursion as function of self awareness.

At the phenomenal level, consciousness can be described as a singular, unified field of recursive self-awareness, consistently coherent in a particular way; that of a subject located both spatially and temporally in an egocentrically-extended domain, such that conscious self-awareness is explicitly characterized by I-ness, now-ness and here-ness.
_{SOURCE: Peters, Frederic Consciousness as Recursive, Spatiotemporal Self-Location}

I'd also posit that being able to read and interpret your own code is a form of self-awareness.

This brings me to the idea that understanding (Chinese Room) is a red herring, and what we're really talking about is interpretation.

We are trapped in subjectivity, humans and automata, and perfect certainty is only achievable in a very limited set of circumstances, such as solved games. (There is also the idea of completeness of a system vs. consistency.)

When we talk about understanding in the sense of anything abstract, by which I means semantics as opposed to syntactics, meaning vs. form, one can say that understanding is a function of interpretation, regardless of the accuracy of the interpretation.

--------------------------------------

Fun Speculation:

I've learned never to underestimate the insight of artists, and the best speculative fiction authors are narrative philosophers in the tradition of Plato. Dick related memory to identity, regardless of whether the memories are real or artificial. This might be thought of as the narrative conception of the self--I am a product of my experience. The real "me" is not merely my body, but "the story of me"--the subjective history that led to this moment of me-ness.

I think it is a not unreasonable assumption that artificial consciousness may arise out of understanding of narrative. Words are just symbols, and there are all kinds of semantic issues, but actions have a concrete aspect. Game theory studies actions in the form of choices, not only in the sense of equilibria, but also as a form of communication. (See: iterated dilemma) The choices agents make in iterated dilemmas constitute a narrative history than can be analyzed and understood mathematically, and these analyses are used for decision making.

It seems to me that the idea of consciousness can be related to decision making. If you're just a transistor, that consciousness is quite limited, more akin to a cell than a complex organism. It may come down to whether one considers intelligence and consciousness to be spectrums as opposed to thresholds.

If you believe consciousness to be a spectrum, then limited artificial consciousness has already been achieved. Consciousness analogous to human-level self-awareness is still in the future.

See Also: Definitions of the Self (wiki); Self Knowledge (wiki)

Answer 2

I think general artificial intelligence will only be possible with some form of self awareness included. Many aspects of human communication do not work if one of the communicating partners does not have self awareness. A good example are many of today's chat bots. They seem to not even hear what they say and only rarely seem to have episodic memory.

Advances in machine to human communication and collaboration will eventually create systems with an ever increasing complex inner model that allows the system to interact in a natural way with humans and to fulfill tasks which require human level intelligence and flexibility. However, unless we have developed a very advanced understanding of consciousness it will be hard to judge how similar or different such a machine consciousness is compared to human consciousness.

Answer 3

"Consciousness" does not have a universal definition. However, if you are really into "consciousness", you should probably read about Searle's Chinese Room experiment or Marvin Minsky's society of mind.

In my opinion, there are many more fundamental obstacles in current AI research that we have to tackle first.

Furthermore, a more formal question would be about how an artificial general intelligence (AGI) would emerge. Even for that, there is no clear roadmap, since we are still very new to understanding the true power of neural networks or other successful AI methods.

Franchois Chollet said in a tweet

For all the progress made, it seems like almost all important questions in AI remain unanswered. Many have not even been properly asked yet

Answer 4

As far as I think, the key to this answer lies in the area of unsupervised learning.

Why?

How do we define consciousness? Probably being aware of our existence. The key to understanding our existence starts by asking questions or finding answers which may not have any questions. This all may sound quite philosophical, but in terms of AI, it may just be finding patterns and logic from what we observe around us.

Just a theory but worth giving it a thought.