11
$\begingroup$

It is often implicitly assumed in computer science that the human mind, or at least some mechanical calculations that humans perform (see the Church-Turing thesis), can be replicated with a Turing machine, therefore Artificial General Intelligence (AGI), defined as a human-like AI, may be possible.

I do not know of any other argument that AGI is possible, and the foregoing argument is extremely weak.

Is there a rigorous proof that AGI is possible, at least, in theory? How do we know that everything the human mind can do can be encoded as a program?

$\endgroup$
1
  • $\begingroup$ It seems you're asking if it is possible to develop AI that operates at a similar level to humans (terms like "Artificial general intelligence" or "Strong AI" are often used to describe this). It's a fabulous question, but it is poorly defined. It begs another question: how do you define human-level intelligence? Is it the ability to convince other human agents that you're human (this is somewhat circular logic)? Is it the ability to write music or create a painting? Depending on your definition, the answer varies wildly, and it must be clarified before proceeding. $\endgroup$
    – Jor
    Jul 4 '17 at 3:30
6
$\begingroup$

A strong reason why people think the mind can be implemented on a Turing Machine stems from the Computational Theory of Mind (CTOM), which is the leading theory of mind for now.

There are lots of reasons for supporting the CTOM, one of which being that the language of belief/desire psychology (propositional attitudes over mental representations) seems to fit nicely to a computational framework.

But most simply is that the computation analogy is very helpful in fields such as psychology and neuroscience. When we know of an input/output pair, but don't know how it is implemented, we could say "its performing the relevant computation".

And since Turing showed that any computation can be performed on an appropriate Turing Machine, the natural extension is that the mind can be implemented on a computer.

However, the CTOM is more of a useful idea than a complete theory. We still don't know how to analyze thought in a logical syntax, which can be implemented in a computer. And we also don't know how/why "computation" (whatever that means in this sentence) is performed in the brain.

$\endgroup$
2
  • $\begingroup$ Comments are not for extended discussion; this conversation has been moved to chat. $\endgroup$
    – nbro
    Jan 18 at 17:02
  • $\begingroup$ 4 years later: Turing did /not/ show that "any computation can be performed on a Turing machine", but that "Turing machines can compute a very large class of functions, beyond which we do not know whether exists functions that are not computable by a Turing machine." this is known as the Church-Turing thesis (that TMs accurately describe what we know as "computation"). check out the Busy Beaver function for a negative example (which fails (i assume) due to being defined as an implicit function) $\endgroup$ Jun 3 at 21:05
4
$\begingroup$

Consciousness is not well-understood

As an AI practitioner and philosopher, I don't think that humans will be able to create a truly conscious silicon-based AGI.

  • Humans are incapable of creating some "thing" from fiat (a decree). It's never happened in human history. The innovation cycle must begin with some "thing" (some "stuff" of some kind), and consciousness is not a thing.

  • The essence of consciousness is imperceptible (it is unseen), like gravity, and attraction. Humans are incapable of creating things that they are unable to observe. Even if they are able to observe it, the human perceptive ability is unable to actually perceive the true essences of things seen, much less those unseen.

  • Humans do not adequately understand the "essence" and "nature" of consciousness - which is a fundamental prerequisite to creating "anything" at all.

  • The "easy" problems, those physical by nature, although not yet solved by empirical domains of psychology, cognitive science, and neuroscience, are expected to be solved in time. Regardless, they are "not" yet solved today.

  • The "hard" problems, those determining why or how consciousness occurs given the right arrangement of brain matter, might not ever be solved, since it must explain why certain physical mechanism gives rise to consciousness instead of "something else" or "nothing at all". This is significant and is the most damning of all arguments against the idea of humans creating true existential consciousness in silicon creatures as a whole.

Dualism vs physicalism

The greatest philosophical debate on consciousness has focused on the distinction between dualism and physicalism.

  • Dualism is the theory that consciousness somehow falls outside the domain of the physical (these are the hard problems)

  • Physicalism holds that consciousness is entirely physical. (significant arguments below view it as false).

Problems with dualist views

  • Why would one be motivated to hold a dualist view?

  • How can something that is not part of the physical world interact with the physical world? That seems impossible!

  • The physical world is a closed system, how can you have a consciousness that is not part of a closed system?

Consciousness is a lot like mass or charges, it's a philosophically "fundamental" thing, you either "have it or you don't", you can simulate them, but you cannot existentially "be" them unless you have those specific "properties", and behavior "simulating" human consciousness is not a fundamental thing.

So, despite the sensationalist tendencies of rogue journalists "parroting" wildly spectacular concepts from the fringe camps of the transhumanists (aka science fiction) - a quick perusal of the more rigorous communities of the grounded and thoughtful philosophers camp strongly and convincingly argues otherwise.

More musings on physicalism

Actually, consciousness has never been properly explained by the biomechanical, which is more or less the key issue of all philosophical studies of the mind - which is essentially the study of consciousness.

Physicalists have trouble explaining several aspects of consciousness in a way that is consistent with our "observations" of how physical properties interact. Let me list a few more problems, with a reference to the titans of philosophy.

Arguments against physicalism

  1. It is impossible to imagine how mere neuronal tissue could produce conscious experience (Huxly)

  2. Failures of supervenience, such as zombies and inverted spectra, are conceivable (David Chalmers, John Locke, etc.).

  3. Mary learns something (Frank Jackson).

  4. Brains have mass, volume, and other physical properties, but experiences do not.

  5. Paranormal phenomena (near-death experiences NDErs, ESP, etc) are real, and involve consciousness implemented in a nonphysical substrate.

  6. If shrunken so I can stroll around your brain and look about, I will observe neuronal processes, not experiences (G. W. Leibniz).

  7. The soul is the seat of consciousness, and the soul is not physical. (Theological constraints recognized BTW...).

  8. Conscious experiences have intrinsic qualities, but science can only tell us about relational qualities (Russell, Rosenberg).

  9. Consciousness cannot be observed; there will never be a consciousness detector that can tell you if a given creature is conscious.

  10. Conscious experiences are not simply the movement of molecules, consciousness is more than mass in motion (Mill, Ward).

$\endgroup$
1
  • 1
    $\begingroup$ I edited this interesting answer to structure/format it properly and to include links to some relevant references that you mention. However, I have not included links to all references, so you may want to finish the work ;) $\endgroup$
    – nbro
    Jan 18 at 22:29
3
$\begingroup$

I'm going to go out on a limb and suggest that this is a matter of evolution, that humans are in no way exceptional in the grand scheme, and that AGI will manifest so long as technology advances, because human consciousness is simply a matter of complexity of the system.

The idea comes out of emergent complexity in Conway's Game of Life. In Conway's words:

"There are Life patterns which behave like self-replicating animals… It’s probable, given a large enough Life space, initially in a random state, that after a long time, intelligent self-replicating animals will emerge and populate some parts of the space.”
Source: Winning Ways for Your Mathematical Plays

I came across a paper Computation in Cellular Automata: A Selected Review, which I am still working my way through, and which you may find interesting.


For those who use philosophical arguments to make the case algorithmic consciousness is not possible, I'd posit the question "how do we know we're conscious?", not because I'm interested in the answer, but merely to throw a wrench into that line of inquiry.

Because ultimately it doesn't matter.

Consciousness in the sense of human awareness is not a requirement of life, and the most basic definition of consciousness is awareness of any kind, no matter how trivial.

I find the idea that there is something "magical" about human consciousness, that ideas are not things because they do not have material form, to be problematic.

Intangibility I don't have a problem with, as intangibles clearly interact with the physical world.

(As an analogy, I studied for many years with a famous Tai Chi teacher who never talked about "chi". I suspect this disinclination derived from the way in which the concept of "chi" leads to magical thinking, which is illusory as opposed to practical. The practice and application of Tai Chi techniques is purely a matter of physics and physiology, even when such applications seem to defy natural laws. Possibly there is something going on that we don't understand, but if that were the case, such phenomena are natural in origin.)

We know there is randomness in nature at the quantum level, and if this proves to be a component of human consciousness, we can use quantum computing to provide a medium for artificial consciousness.

$\endgroup$
1
  • $\begingroup$ Comments are not for extended discussion; this conversation has been moved to chat. $\endgroup$
    – nbro
    Jan 18 at 17:19
1
$\begingroup$

Rather than prove that Artificial General Intelligence is possible, I would consider an argument for why it is impossible.

We start by defining what we mean by AGI. You state that the human mind can be replicated by a Turing Machine, and therefore AGI should be possible. This seems to imply that humans have `General' (capital G) intelligence. By this I mean that you are implying that with enough time, humans can learn any task or problem. However, if you are asserting that humans minds are machines replicable by Turing machines, you must also concede that they have some finite representational power. Finite representational power implies that there will always be problems or tasks where our intelligence will fail (a consequence of the No Free Lunch Theorem).

Fortunately (maybe unfortunately), finite representational power is what allows us to learn at all: VC Dimension (a measure of the complexity or representational power of a class of functions that a learning algorithm can learn [also here and here]) implies that a learning algorithm that can learn any problem is actually useless, as the ability to explain any set of data yields the requirement that the algorithm see an infinite amount of examples in order to generalize. While this result comes from the relatively constrained class of binary classification problems in the statistical learning setting, the intuition seems to apply more broadly.

To summarize, I would refer to this quote from Shalev-Shwartz and Ben-David (2014):

If someone can explain every phenomenon, his explanations are worthless.

It truly is the case that our decision to systematically ignore some possible outcomes is the only thing that allows use to learn useful representations of real-world problems.

$\endgroup$
0
1
$\begingroup$

Although it is not a rigorous proof, Marvin Minsky's book, The Society of Mind gives us a blueprint for creating a "mind" (general intelligence). In his book, he posits that by combining mindless components ("agents") together in various competing and cooperative structures, we can create actual minds.

IMHO, the recent popularity of Boosting, Bagging, Stacking, and other ensemble techniques will eventually evolve (through research) into Marvin Minsky's "agent" metaphor. Subsequently, as we learn to make these agents compete and cooperate (looks like this has recently begun with Generative Adversarial Networks), we will be able to write "programs" that mimic (or surpass) the human mind.

$\endgroup$

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