4
$\begingroup$

So a scientist in the somewhat distant future is living on the final pages of the chapter that we call 'humanity.' For various reasons it's impossible to sustain human life (outside the scope of this question), and so he turns to pass on humans to machine form.

However, Dr. Scientist, doesn't want AI systems that are little more than extremely complex chatbots; that they could be passed off as human--that they're extremely sophisticated deceptions--isn't good enough. Dr. Scientist wants to creat AIs that are actually capable of free will.

The creation of these AIs is outside the scope of this question. The question is about the test of free will that Dr. Scientist creates before attempting his engineering of a system with free will.

First, take the following absolute assumption:

  • Free will does exist

Dr. Scientist realizes this is a point of contention among philosophers down the ages, but he's unwilling to accept a world where he's just a meat computer, and so his test will form around this assumption.

Now take the following working assumptions based on Dr. Scientist's understanding of free will:

  • A system without free will when, in the same state, given the exact same input, and in the exact same environment will always produce absolutely the same output (down to, and including, the most minute details)
  • A system with free will when, in the same state, given the exact same input and in the exact same environment may produce different outputs
  • The difference defined above is (one of?) the key defining factors of free will

So we can then deduce:

  • If we place a given system in the same state, in the same environment, and provide the same input and always get the same output the system may or may not have free will
  • If we place a given system in the same state, in the same environment, and provide the same input and at any point get differing outputs the system must have free will

The question is:

Ignoring the practicalities of absolutely perfectly recreating a given environment, state, and inputs, (explainable within the context of the story) would this be a reasonable test for possession of free will for Dr. Scientist to use? Further, are any of his working assumptions questionable/wrong that you can see leaving the first assumption, that free will exists, unchallenged?

Side note

I believe, under the above definition, some aspects of quantum mechanics would pass as 'free will.' I'm not sure how Dr. Scientist would handle this, but I assume he would call this a 'false positive.' If you feel this is relevant, I'm perfectly fine with suggested revisions to his test to account for this.

Edit

I also think it might be worth mentioning that for the purpose of the story he never creates a system with free will, and that's not what the story is about; it's about his attempt to create it, and his repeated failures. However to attempt to create something you need a reasonably define goal that you're aiming for, as well as a way of determining success or failure, thus this question.

$\endgroup$
  • $\begingroup$ Not a very scientific definition of Free Will, as in not testable. Plus there is no way to prove an AI has free will until you can prove a human has it, you cannot just assume humans have it. You'd have to design a test that all humans always pass. I have yet to see a convincing argument that given the exact same set of circumstances a human can choose differently. That doesn't change the fact that it was their choice, free will isn't about being able to choose differently given the same circumstances, but about making your choice in a given set of circumstances. (imnsho) :) $\endgroup$ – Seeds Jul 12 '16 at 21:16
  • $\begingroup$ Another easy to define free will; when faced with a conflicting choice, something with free will will violate its values. For example, kill five people to protect someone it is not connected to but knows is innocent. I'm not sure how to test that one, but it seems reasonable that something without free will will not violate its code. $\endgroup$ – Xandar The Zenon Jul 15 '16 at 3:52
6
$\begingroup$

That definition of Free Will is (mostly) only of academic interest

Suppose Dr Scientist does find that we are indeed only the products of the physical state of our being. The result — when the exact pre-conditions are re-created — is always the same. What are the practical implications of this?

None really, because even if the process that leads up to a decision is deterministic, it is also chaotic. Chaotic in this context means "impossible to predict". You can only know the result if 1) you have run through the simulation once 2) you then managed to recreate the exact conditions.

Suppose the other option: the result varies. Does that prove Free Will? No it does not. It only leaves in doubt what this is actually attributable to. Did he fail to recreate the conditions, or is the process indeed susceptible to truly random processes in the physical world (such as radioactive decay)?

In either case he is left with the same implication: Free Will — whatever it actually is — is so chaotic / random that it means that the behavior of humans is unpredictable for any and all hitherto unknown situations. The behavior can only be predictable for situations where all the conditions are precisely known.

That said... as with other chaotic but deterministic processes, Dr Scientist may still see patterns emerge. He runs the simulation over and over, tweaking the parameters. And while the results do vary, he may still see that the results clump together. And he may find that the human mind is fractal. It may perhaps not exactly predictable without running the simulation, but he perhaps still can make broad generalisations.

enter image description here

A fractal...

  1. is deterministic, i.e. will always yield the same result the calculated
  2. is chaotic, i.e. the result is impossible to predict until you have done the calculation
  3. displays patterns, you can see recurring themes

In any case... that definition of Free Will is like proving the existence of (a) God.

If you do manage to prove the existence, then what? Ok, so there is a god. What does that mean? In short that the laws of physics are the subject of the free will of a supernatural creature that can do whatever they want with them. That which we thought was predictable is actually subject to the whims and wishes of some kind of persona.

And if you fail to prove the existence of this supposed god (which is as far as you can get because you cannot prove non-existence, you can only fail to prove existence) you are still left to wonder if your test was exhaustive.

$\endgroup$
  • $\begingroup$ So I'm going to reread and ponder your question before further comment, but I'm fairly certain (and a quick googling seems to confirm) that radioactive decay is random due to quantum mechanics exclusively. A theoretical AI, reasonably, could avoid quantum systems affecting its path from input to output. $\endgroup$ – Nex Terren Jul 12 '16 at 19:22
  • $\begingroup$ @NexTerren Well it all comes down to the nuts & bolts of your simulation. Is it built on entirely deterministic components or not? If the components are 100% deterministic, and their interaction between each other is 100% deterministic, and the given input is identical, then I think it has been proven that this must yield the same result every time. $\endgroup$ – MichaelK Jul 12 '16 at 19:27
  • $\begingroup$ So the story concept is a bit complex. I could go into it, but I don't want to distract from the question's topic. Most importantly, he never succeeds in creating an AI with free will during the story, so your statement could be the case in the story. He refuses to accept that randomness (quantum or otherwise) is free will, and he refuses to accept that free will doesn't exist, and so he needs to create a test to fit that, in his desperate attempt to save himself and his loved ones. $\endgroup$ – Nex Terren Jul 12 '16 at 19:33
  • $\begingroup$ As I said above: you have — kind of falsely — let your Dr Scientist divide up the possible outcomes kind to binary: Deterministic, which he says means "No Free Will" while Random means "Free Will". The actual result — which will be the source of your dear Dr Scientist's dilemma — is that is neither... or both. It will end up pseudo-random, that is to say it may be so chaotic that you cannot predict the result, even if it is deterministic. And this means it really doesn't matter if things are deterministic, because the chaos will still give you the power to choose. $\endgroup$ – MichaelK Jul 12 '16 at 19:41
  • $\begingroup$ I don't follow your last comment. If a system is actually deterministic there can exist no power to choose since the system's state is the 'choice' so to speak. While certain things can be random (quantum, actually, is the only thing I know of agreed upon by scientists as possibly actually random), the doctor is assuming that there's another way to very the output, and that's free will, the power of choice. He might be wrong (that free will doesn't exist), but his test uses that assumption. $\endgroup$ – Nex Terren Jul 12 '16 at 19:47
2
$\begingroup$

I don't think that's a valid definition of free will. By this definition, free will = randomness. I can easily program a computer to randomly choose from among multiple responses to an input. Like, you type in a number. The computer adds a random number between 1 and 10 to it and outputs the result. I don't think that's any sort of proof that the computer is sentient, much less possesses free will.

I think the fundamental problem with trying to construct an experiment to tell if an AI has free will is the same problem philosophers have long had in telling if human beings have free will.

Scenario 1: You order the AI robot to, whatever, walk across the room and stand in the northwest corner, and the robot then walks across the room and stands in the northwest corner. Likewise, it consistently obeys all orders given to it. You then ask the robot, "Why did you obey that order?", and it replies, "Because I chose to. I believe the person giving the orders is wiser than me so I follow them."

Scenario 2: You order the AI robot to walk across the room and stand in the northwest corner, and instead if sits on the floor where it is. You ask it why, and it replies, "Because my programming says that whenever someone tells me, 'Stand in the northwest corner of the room', I should sit on the floor."

In which case, if either, does the AI have free will?

Consider actual humans. We certainly have the APPEARANCE of free will. We make decisions and act on them. The argument for determinism (or predestination when the context is religious) is that this is an illusion: that we only make the decisions that we make because we were pre-programmed to make those decisions, by nature or God or whatever. How do you distinguish, "I am making this decision of my own free will" from "I think I am making a decision, but really my choice was predetermined"? It's not clear what experiment you could perform to distinguish them. Which is why this question has been debated for thousands of years with no indication that we're closer to an answer than the ancient Greeks were.

If I could describe an experiment to tell them apart, I'd be a world famous philosopher rather than some guy posting on a forum. :-)

$\endgroup$
  • $\begingroup$ Very good answer, +1. It cuts much faster to the core of the problem with the definition. My answer touches on the same issue: the appearance of Free Will and how the chaotic nature of things makes it meaningless whether Free Will is something deterministic or random. But your answer points out the fundamental flaw of the definition more clearly. :) $\endgroup$ – MichaelK Jul 13 '16 at 9:16
  • $\begingroup$ Well a couple of points. First the computer actually cannot give you a random number; if placed in the same state and asked for the same input for a random number (environment doesn't actually matter) you'll always get the same random number as the seed will always be the same. It's actually incredibly difficult to actually achieve randomness (possibly impossible), and currently the only path we believe might be actual random is quantum mechanics, and we're still trying to prove that. $\endgroup$ – Nex Terren Jul 13 '16 at 9:46
  • $\begingroup$ Second, since both of your robots are simply following a path of logic--effectively complicated light switches, or complex chatbots to quote my earlier comparison--they absolutely are not an AI with free will, any more than my Chrome browser is an AI with free will. They're puppets of their programming, and facades by design, and too often treated as agents of free will by science fiction. His test would, indeed, eliminate that. $\endgroup$ – Nex Terren Jul 13 '16 at 9:51
  • $\begingroup$ (My computer keeps Blue-Screen-of-Deathing every 3 minutes. I'll continue when I can actually keep it online. :) ) $\endgroup$ – Nex Terren Jul 13 '16 at 10:03
  • 1
    $\begingroup$ @NexTerren True, computer random number functions are really "pseudo-random". Actually they ARE dependent on the environment, namely, on the time of day. At least, most pseudo-random functions I've looked at start perform some simple arithmetic calculation on the previous value to get the new one -- typically multiply by some big number, add some big number, and then take only the last some-number of binary digits. But they start the sequence with a seed, and by default the seed is usually the last some-number of digits of the microseconds value of the time of day. ... $\endgroup$ – Jay Jul 13 '16 at 13:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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