Malcolm Pollack on AI and its Threat: Determinism, Predictability, Free Will

Our friend Malcolm Pollock in Brake Failure expresses a reasoned pessimism about our future under AI. I share his concerns. Will we humans have a future? Or are we facing what I have elsewhere called the Ultimate Replacement?

In Stephen Wolfram on AI and Irreducible Complexity, Malcolm explains

. . . a distinction between two [disjoint] subsets of deterministic systems: those whose behavior are describable by simplifying formulas that can be used, by taking their initial conditions as inputs, to predict their future state, and those for which no such reduction is possible.

An example of the former is the movement of two bodies under mutual gravitational attraction, such as a planet and its moon, or the earth and a ballistic projectile. Given the masses of the two, and their initial positions and velocities, it is possible to calculate their positions for any future time.

A good example of the latter is what Wolfram examined at length in his book A New Kind Of Science (which I labored through when it cam[e] out in 2002): the behavior of “cellular automata“, simple systems whose behavior is defined by a small set of rules, but for which, given the system’s state at time t, the only way of determining its precise configuration at time t+n is actually to iterate over every step between t and t+ n. Chaotic systems, such as weather and turbulent flow, are of this kind. So is biological evolution.

Fascinating. What struck me is that the first type of deterministic system allows for prediction, whereas the second type decouples determinism from predictability. I would add that if time is a continuum, then there are continuum-many iterations between t and t + n, which implies then there will have to be continuum-many iterations total.  That would be the ultimate nail in the coffin of predictability, a nail that not even the ultimate claw hammer could remove.

A couple of further questions occur to me.

In the second type of determinism, what becomes of the distinction between determinism and indeterminism? There would presumably still be the distinction, but how could one tell  if a type-2 system was deterministic or indeterministic? Malcolm, glossing Wolfram and Greene, writes, "There is no quicker way, no shortcut, for predicting the future state of such systems than simply letting them run, and seeing what they do." That boils down to saying  that in the second type of deterministic systems there would be no way at all of predicting future states of such systems.  How then could one 'determine' (come to know) whether such a system was deterministic or indeterministic?

If the deterministic systems that really interest us are of the second type, then Laplace's Demon is, if not out of a job, then bound to be underemployed.

Second question.  If we humans are deterministic systems of the second type, might this permit a deterministic reduction of the much-vaunted free will that we feel ourselves to possess? I don't think so, but knowing Malcolm, he may want to take this ball and run with it.

Comments

[john doran]

1) "unpredictable" and "indeterminate" are not synonyms.

This hearkens back to your #6 in the Fuchs post:

Truth does not entail knowability by us

Just because we can't know (and so far forth, predict) an outcome based on some putative state of original affairs, doesn't in any way entail that the outcome of that state of affairs isn't strictly determined.

There is absolutely nothing on Wolfram's second scenario that is not just an example of this basic point, writ large.

This is, to me, a strange focal point for what has been, through my philosophical career, the typical principal point-of-entry for this debate, namely the conceit that quantum mechanics shows that reality is, fundamentally and inexplicably, random (i.e. acausal).

The Wolfram essay basically concedes that our inability to predict outcomes for cellular development is an epistemological failure, rather than a function of a fabric of reality that is fundamentally random.

2) if we are Wolfram Machines, then everything we do is determined, though sometimes unpredictable.

If we are Standard Model Quantum Machines, then every "choice" we make that we think of as "free" is random, and so far forth neither "free" nor "chosen".

I think of it this way:

(i) human beings can make free, undetermined, choices

(ii) all matter behaves in a way determined by strict covering laws

therefore:

(iii) human beings are not solely comprised of matter

Or so it seems to me.

[Malcolm Pollack]

Hi Bill,

Thanks as always for the mention.

You write:

If we humans are deterministic systems of the second type, might this permit a deterministic reduction of the much-vaunted free will that we feel ourselves to possess? I don't think so, but knowing Malcolm, he may want to take this ball and run with it.

I did, in fact, run some distance with this ball (or a closely related ball) in a linked series of posts beginning sixteen years ago (the latest post in the series was written in 2014).

At that time, I was still pretty firmly committed to scientific materialism, and the series was written from a "compatibilist" POV, strongly influenced by Daniel Dennett's books Elbow Room and Freedom Evolve.

A central point is the seeming incoherence of what "free will" can even really mean. Do we really imagine that our decisions should be wholly independent of our prior state? If so, what are they caused by?

It all hinges on the idea that we "could have done otherwise". One interesting discussion of this, which I mentioned in the seventh post in the series, was by J.L. Austin, in the example now referred to as "Austin's putt".

I'll be honest: many years later, with a far more inclusive view of possible metaphysics, I still find this topic as perplexing as ever.

As for determinism and predictability: yes, it's important to pry them apart (just as I believe we should pry apart consciousness and intentionality). But, given that systems can be computably irreducible (i.e., unpredictable) while remaining wholly deterministic, I rather doubt that the distinction between determinism and predictability tells us anything important about the thing we call "free will".

[Mehra]

Typically what we want to predict is some gross feature of the system. We don't need to predict all the 6N degrees of freedom of a N-particle system. We may only want to know pressure or temperature at a point. These gross features may be predictable even if the exact location and velocity of all N particles are not.

Also, chaotic systems typically have attractors. Their dynamics is confined to the attractor and hence remains predictable in certain degrees of freedom.

In systems such as weather or solar system, we are not interested in knowing the precise configuration.

[Tom T.]

FA Hayek addressed these issues in a paper in 1961, A Theory of Complex Phenomena https://www.semanticscholar.org/paper/The-Theory-of-Complex-Phenomena-%3A-A-Precocious-Play-Hayek/b76f00cfcd2155439adc73d5c90122f045ae5c20. This was two years before Edward Lorenz developed chaos theory in connection with weather systems. Hayek noted that the very successful deterministic, causal systems like that in physics generally involve two or maybe three variables. In contrast, complex systems like biological organisms, the evolution of species, human consciousness, and social orders involve multiple variables with various feedback loops that make precise predictions on a determinate, causal law basis all but impossible. Hayek said:

"It would then appear that the search for the discovery of laws is … merely a characteristic of the theories of simple phenomena as we have defined these earlier; and that in the field of complex phenomena the term 'law' as well as the concepts of cause and effect are not applicable without such modification as to deprive them of their ordinary meaning."

In sum, simple systems are both causally explainable and predictable in terms of the future states of a system and fit well with a notion of strict causal determinism. Complex systems, however, are only causally explainable in abstract, recurring patterns or models that cannot yield precise predictions, but only a range of possible future states based on hypothetical ceteris paribus assumptions, as Hayek put it. The assumption of "all things being equal" radically changes the causal explanation of complex systems from a strict deterministic one to a hypothetical with an often wide range of future possibilities.

Concerning your second question, even if human beings are "deterministic systems of the second type," Hayek concludes that "The chief fact would continue to be … that we should not be able to state the full set of particular facts which brought it about that the individual did a particular thing at a particular time. The individual personality would remain for us as much a unique and unaccountable phenomenon … whose specific actions we could generally not predict or control, because we could not obtain the information on all the particular facts which determined it."

I will note that Hayek was writing in a philosophy of science context amidst numerous conversations with Karl Popper and others of his time, and framed his argument ultimately in terms of the practical impossibility of prediction in complex systems. But on a deeper level, I believe he was subscribing to a Humean notion of cause and effect, and to the extent I understand it, I find philosophical support for Hayek in this old paper of yours, https://web.archive.org/web/20060112051635/http:/www.independentphilosopher.org/could_the_universe_cause_itself_to_exist_vallicella.htm, which tracks somewhat close with what I believe Kierkegaard was getting at in that metaphysical riff on cause and effect, necessity, and freedom in Philosophical Fragments.

But that would involve an extended argument and this comment is already too long, so I will have to leave it there.

[Tom T.]

The link to the Hayek paper appears broken. Here is a better link:

https://www.semanticscholar.org/paper/The-Theory-of-Complex-Phenomena-%3A-A-Precocious-Play-Hayek/b76f00cfcd2155439adc73d5c90122f045ae5c20

[Tom T.]

The link to your (BV's) paper also failed; apparently, an ending comma or period is interpreted by your blog as part of the address. Here is a clean link:

https://web.archive.org/web/20060112051635/http:/www.independentphilosopher.org/could_the_universe_cause_itself_to_exist_vallicella.htm

[EG]

@john doran,

you write:

"I think of it this way:

(i) human beings can make free, undetermined, choices

(ii) all matter behaves in a way determined by strict covering laws

therefore:

(iii) human beings are not solely comprised of matter

Or so it seems to me."

This seems simple, straightforward and unobjectionable to me. Do you have a guess about what the non-matter is?

[Tom T.]

Following up w/ some more specific points regarding the Wolfram book, as quoted by Malcolm. I haven't read it, but I would be interested to know if Wolfram cites Hayek's works on complexity and the proper scientific methodologies. Hayek is still quite relevant in economic and social science circles, as well as neuroscience; the 1961 paper I noted above alone has been cited almost 400 times since 2020. And Wolfram's "new kind of science" appears to be somewhere in the neighborhood of what Hayek advocated for most of his 60+ year career, first in economics and then more generally in the biological and social sciences.

The Wikipedia article Malcolm cites does not give (for me) a clear explanation of the mathematical algorithm underlying the behavior of the cellular automata, nor the actual behavior exhibited. But following Hayek, I would suggest that if the algorithm does not precisely predict the what and when of any future behavior, then it is not a deterministic covering law like that found in physics or mechanics. It is something; but for the algorithm, the cellular automata would exhibit no behavior at all. So in that sense, the mathematics does determine the outcomes. But not in any sense like the strict determinism found in say, spreadsheet algorithms used by a standard desktop computer.

This puts the question of determinism v freedom in a different light. If in complex systems like human consciousness, there is no strict determinism to be found, then human actions would be relatively free. With no covering law determining any choice made, the choice would be free but nevertheless constrained by the finite number of possibilities under consideration.