Monday 18 November 2019

Why computers can't become Artificial Intelligences (AIs)

I am saying that computers can't ever under any circumstances become AIs assuming that an AI is a meant to be a conscious agent; with 'free will', able to generate thought that is original with itself (and not random nor merely an output directly determined by inputs).

You will see immediately that the discussion goes to a deep, metaphysical level concerning the nature of life, consciousness and agency.

For me, believing in an 'animated universe' and given that I regard everything as 'alive and conscious' to some degree and in some way; the question is one of agency. Computers are alive and conscious in the way that - maybe - rocks and rivers are alive and conscious; but is not an agent in the way that a Man is.

This is because agency is possible only to Beings, and Beings have existed from eternity. And Agency is not possible to all Beings, but only to Men.

I don't know why this is, but it seems to be a basic fact of reality. God created creation, and the main purpose is that Men may develop (if they so choose) to become gods on a level with the creator and participating in creation.

But this does not seem to apply to everything that exists or is created - not everything is apparently capable of evolving to become divine. It may not be appropriate to ask 'why?' - because things must be set-up one way or the other: either some things (including Men) can become gods, or all things can become gods - and my intuition is that it is some things, not all.

A Man might 'become a computer' with agency, but a computer could not become an agent. 

Discussion is welcome - but in the end it will probably merely be a matter of different assumptions. So, it is best to start from there.


Interdimensional Spiritualwarrior said...

I agree about AI. There is a clear demarcation, it will never be like Us, conscious, with Free Agency and more as well.

I do however see it as possible , and it may well be going on now and already for decades or perhaps centuries amongst certain ‘elites’, and the agenda may be accelerating now - that the use of AI and ability to make cloned meat suits that have some biological aspect to them but aren’t Human, I’m talking about hybrid biorobotoid clones that look human but aren’t but are in society not just in high places such as Hilary Clinton, but everyday members of the public too. AI might be running and directing much of this infrastructure but then negative demonic entities ‘walk in’ and inhabit the AI hybrid.
As well, demons could infect computer systems and AI itself and be a foothold physical interface for them in this place. This may be happening.

But I agree AI in and of itself is no match for organic human ensouled connected to Spirit and Creator Source with free agency.

Robert Brockman said...

Less intelligent organisms (dogs, cats, apes, etc.) do seem to be capable of some agency -- perhaps first we need to pin down what we mean by agency?

Where things can get nasty with AI is that rather soon we will have computers that have a *very convincing* appearance of agency backed by super-human learning and pattern matching ability. If these systems lack either subjective experience or free will / agency we will be in a world of trouble, because most people (possibly including their creators) will begin to treat them as if they are people (or worse, gods).

What's much more important than whether or not AIs have true consciousness / agency is whether or not we have correctly judged if a particular AI is a true "being" or not. A mistake either way will have bad consequences. We cannot afford to "guess", "suspect", or "hypothesize" on this matter -- we need to *know*.

-- Robert Brockman

Bruce Charlton said...

@RB "What's much more important than whether or not AIs have true consciousness / agency is whether or not we have correctly judged if a particular AI is a true "being" or not. "

I don't see how anyone could do this until after they had understood what a Being was. What They will want is to have a reduced and systematic definition of Being, and a reduced and stereotypical criteria of 'evidence' for deciding - like defining the ultra-simplified and impoverished 'model' of a Turing Test as the 'proper' way to detect and test intelligence...

Robert Brockman said...

Right, and the reduced and systematic definition of Being will certainly be "functional": "Does the candidate have a similar "input-output relationship" to a known Being?"

Since by definition the behavior of a Being with free cannot be completely described by an input-output function, the (quite predictable) efforts to impose this sort of systematic definition will end poorly.

Clearly one of the issues here is the big push to redefine *humans* in terms of their "input-output function". Some of this seems merely misguided, though there is likely significant evil at work. "Functionalizing" humanity will make the human Being / AI distinction much harder to sort out.

I think we need to push really hard against this notion. I'm coming at this from the physics end, where nondeterminism seems to be a fundamental aspect of matter. Obviously any authentic spiritual tradition will resist defining human Beings in terms of their functional relationship.

-- Robert Brockman

Wm Jas Tychonievich said...

But “Men,” as biological organisms, have not existed from eternity. The eternally existing Being is something that comes to be associated with the organism, and the latter is produced through physical and chemical processes which could, in principle, be mimicked artificially.

It seems to me that “artificial intelligence” is neither more nor less possible than “biological intelligence.” Both terms are strictly misnomers, since neither a machine nor an organism can be, or produce, a thinking Being, but there could in principle be a machine that could be called an AI with as much justice as Man is called a rational animal.

Bruce Charlton said...

@Wm - There is a way of thinking that has Being as fundamental - so Being cannot 'become associated with' an organism - Being was always there, and transformed through time. I'm pretty sure you are using the mainstream metaphysical assumption than evolutionary development starts with matter and adds consciousness, whereas I'm saying that we start with consciousness and matter 'condenses' from it.

Wm Jas Tychonievich said...

I’m not talking about evolution. I mean that when a new baby is conceived (or born, or whenever it happens), an already-existing consciousness attaches itself to it (incarnates). The physical organism is new, but the associated consciousness is not.

Dividualist said...

Nobody cares if an AI is *really* a conscious agent, at the very best, they want one that *simulates* being a conscious agent so well that neither you nor anyone could tell it is not one. Turing test and all that. This is what I don't understand about metaphysicians, like Ed Feser and you. There is a difference between really being vs. seeming, but in these AI stuff we are talking about doing seeming so well that it is empirically indistuingishable about really being. No one can tell the difference. So why bother about something that is empirically not testable? I mean, if you are talking about the metaphysics of God, at least it is useful for the purpose of saying it is at least possible that Christ was not only Man but also God. Useful for purposes of faith. But in this case, the hypothetical case of an AI simulating a conscious agent really well, there is no information at all in saying it is not really a conscious agent.

But the reality is that no one even cares about the Turing test. They simply want smarter software tools that work more autonomously. And the interesting part is that it is harder than it seems. It is always vaporware, it is always 5 years into the future, and projects like Amazons autonomous delivery drones quietly fail. So here is an interesting question: beyond a certain level, complex and autonomous tools are impossible without this metaphysical stuff? If yes, what would be that level? That would be quite useful to predict.

Bruce Charlton said...

@D "Nobody cares if an AI is *really* a conscious agent"

You are adopting the assumptions of philosophiocal pragmatism while denying that these are assumptions.

Metaphysicians are people who admit the reality and neccesity of their own (and other people's) assumptions - Not to be a metaphysician is merely to be ignorant (or dishonest) about the fact that there are always assumptions.

Whether you personally care about this question, or whether the masses, or the bureaucrats care about it is just a matter of contingent psychology. Clearly some people care a lot about it, and have done for 60-ish years - since it is a constant theme of popular fiction (especially scifi). This is a general recognition that real/ simulated consciousness does make a difference.

Bruce Charlton said...

@Wm - I amswer your question here:

Robert Brockman said...

Dividualist: “No one can tell the difference.”

You could tell the difference if it was you. More importantly, there *will* be differences that affect others, but they by definition will not be predictable or empirically verifiable. There are hard limits to empiricism — one of the triumphs of modern physics that it has found some of these limits. (I can discuss this at some length if you are interested.)

BC: “We start with consciousness and matter ‘condenses’ it”

Essentially, yes. (The Surangama Sutra discusses this in some detail.) Similarly, consciousness comes before cognition.

The real fun begins when you realize that perception comes *after* the event that is being perceived temporally but *prior* to the event causally.

— Robert Brockman