Moral Machine

Discussion from FreeWill:

"The nature of reality says nothing about the nature of morality so 'deterministic doctrine' (whatever that might be) cannot destroy human morality."

The statement "one ought to do X" has, as a necessary premise, that one is not foredoomed to do X. That is, a moral imperative can only be addressed to a being with free will. "Moral machine", like "non-deterministic machine", is a contradiction in terms. -- Michael Brazier

It isn't quite that simple. Moral philosophers do not accept any argument of the form "it isn't immoral for me to do X since X is unavoidable" even though one can contrive examples where X is clearly unavoidable. The reason moral philosophers do not accept such arguments is because in reality, you can never know any such thing and in reality, this argument has been used many times to rationalize situations where X was found to be avoidable after the fact. Example: slave masters in the South argued that if every slave were freed, the economy would collapse. It didn't. Incidentally, they also argued that owning slaves was preferable to merely renting them (wage slavery) since you care more about the things you own. In that they were right.

Moral machine may be a contradiction in terms but at the level of abstraction at which morality exists, there is no such thing as a machine. Human beings are seen to obey moral laws, they are not seen to obey quantum mechanical laws. Saying that determinism contradicts 'free will' or morality is like saying that wheels contradict the free motion of a car or even movement; determinism makes these things possible, it couldn't possibly contradict them. Determinism only means that the universe can figure out what is foreordained, it doesn't mean that you know what is foreordained. -- RichardKulisz

Hold on - if reality is totally mechanical at one level of abstraction, it must be equally mechanical at all higher levels of abstraction too. If the human body, for instance, is deterministic when regarded as a set of interacting quantum particles, it must be equally deterministic when regarded as a single thinking and acting being. You can't turn a machine into a not-machine just by ignoring the fine details of its mechanism. -- Michael Brazier

That is exactly what you do by ignoring the fine details of the system. One subvenient property on which the human-level lack of determinism is based is the system's intractability. Since you can never predict the evolution of a large system (like the human brain) in practice, then it doesn't appear deterministic. Subvenient properties don't have to be carried over into the supervenient level. A society of people all concerned with justice and mercy could still be immeasurably cruel because of limitations in individual judgement formation.

And here's the most important fact; Everett's theory shows that a constant piece of the wavefunction is always deterministic but on the human level of abstraction, you are never dealing with a constant piece of the wavefunction but always with an ever-shrinking piece of it. What is deterministic in Many-Worlds is the evolution of the many worlds not "your" path through them and "your" path is shown to be meaningless at that level of abstraction, but that is exactly what you're concerned with at a higher level of abstraction. The entire supervenient level is based on a concept that is proved meaningless at the subvenient level; it's going to break all the rules of the subvenient level. Determinism simply no longer applies. I don't think this explanation is very clear unless you're familiar with exactly how Many-Worlds restores determinism. -- rk

As it happens I am familiar with the way many-worlds "restores" determinism - but that feature of the theory does not get you out of the problem at hand. The pass from the "subvenient" to the "supervenient" is not an abstraction at all, but a selection - a choice function on the set of histories, in fact, of precisely the kind that you argue against on the QuantumPhysics page! The "supervenient" viewpoint itself is meaningless in terms of the "subvenient" quantum formalism, and therefore any concept depending on it (such as moral imperatives) is meaningless too. -- mb

From the viewpoint of quantum theory, yes, absolutely. From the viewpoint of human experience, not at all. If it did then political theory would be meaningless, as would psychology, neurobiology, and even biochemistry. A selection can be a valid abstraction. As I point out in QuantumPhysics, it exacts an extremely heavy price if you include it in the theory, but there is no reason why you can't use it to add a layer of abstraction. Abstractions between different levels are pretty much extra-theoretical; they aren't considered to be included in the theory so you get them for free.

Selections can be valid abstractions only if the things left out are "insignificant" compared to the things included. The "supervenient" viewpoint doesn't pass that test, under the many-worlds interpretation.

Significance is too subjective a concept to base any formal argument on. From a formal viewpoint, selections are always valid abstractions. The only question is whether a particular selection is a useful abstraction. Your criterion is another way to express usefulness and is not related to validity. -- rk

One more example, then: modeling the Earth's motion. You may choose to consider only the gravity from the Sun, in which case you will get a rough approximation to the true orbit of Earth. Or you may choose to consider effects of the Sun, Jupiter, and Saturn, which gives a closer approximation. But if you chose to consider the effects of Jupiter and Saturn, but left the Sun out, you wouldn't get any sort of approximation - the Sun's effects are too significant to be ignored. That selection emphatically does not give a valid abstraction! -- mb

Can you think of a more accessible explanation than the one I gave previously? I had no doubts you would understand but I am concerned about the eventual conversion of this page to DocumentMode. -- rk

By contrast, Boyle's gas law (PV/T is constant) is a real abstraction: the pressure, volume, and temperature of a gas can all be defined in terms of the motions of all the molecules constituting the gas. Only when we consider all the molecules indifferently do we gain the right to think in abstract terms. The equivalent in many-worlds would be considering all the states within a constant section of the wavefunction. Looking at a slice that changes every moment is a sure path to nonsense. -- mb

It would be if we did this for no reason at all. But this is precisely what we perceive. This is how we get the illusion of non-determinism. I think that by "real" abstraction you mean an abstraction whose relationship to the lower levels is fully understood. Morality is an abstraction over human behaviour. Human behaviour is an abstraction over neurobiology. Neurobiology is an abstraction over biochemistry. And the links between any of those are not understood. Of course, this doesn't make them any the less useful.

By a "real" abstraction I mean a high-level concept that can be defined in terms of the lower levels. Whether we are aware of the definition yet is not material. Another example of a real abstraction is the notion of a "chemical bond": the science of chemistry was built around that concept, and it led to any number of useful results, long before it was defined in terms of a lower level (electron orbitals within atoms, in this case.)

Now, once the low-level phenomena that give rise to chemical bonds were understood, it became clear that the concept of "chemical bond" was not quite correct - but it wasn't mistaken, either. The central problem with your "supervenient level" is that according to the lower level of the quantum formalism, under the assumptions of many-worlds, taking that viewpoint is simply a mistake. Not a simplifying approximation to the complex truth, but flat out wrong. It does no good to point out that our perception works on that level; that only means our perception deceives us.

If you want human perceptions, moral imperatives, and other concepts depending on the "supervenient level" to have any validity, you must grant the reality of state reduction as a fundamental phenomenon, and explain "supervenience" in terms of it. Many-worlds does not have room enough for it. -- mb

I beg to differ. Your criterion is only about usefulness, not validity. You talk as if it's a big deal that reality is deceiving and that our perception of it is an illusion; it's not. There is no reason to accept state reduction unless you cherish the illusion and insist on a naive view of reality. Talking about our perceptions, even if they're just illusions, is still useful. -- rk

If our perceptions are illusions, talk about them cannot be useful; an invalid concept has no utility whatever. (Validity is not sufficient for utility, but it's certainly necessary.) -- mb

Yes, but I dispute your claim that selection functions are invalid abstractions. I think they are valid, therefore our perceptions are entirely valid. Deceiving, illusory, and sometimes downright not useful but certainly valid.

Did you know that meta-theories of quantum mechanics have implications for moral theory? Rolling a die to see who you kill in a LifeboatScenario? is not a legitimate moral process in a single-history universe, but it is a legitimate process for a many-history universe like Everett's. It has nothing to do with non-determinism but with the fact that the outcome (who dies) is necessarily asymmetric in a single-history universe, which makes the decision process arbitrary and thus morally illegitimate.

I've always thought that the allegation that it was immoral to kill one person to save the many was a defect of Kantian moral theory and I found a way around it. I think that is the coolest thing. I have been told that in most actual LifeboatScenario?s, everyone perished as a result of refraining from choosing someone to murder. If this is correct then people's behaviour is clearly based on their illusory perception of a single-history universe. Hmmm ... it seems like I'm giving an argument for your position, but no matter it was just aimless rambling to begin with. :-)

I've been told that in at least one historical case, the people in the lifeboat were rescued only a few hours [days actually] after they killed one person. When they returned to land, the survivors were all convicted of murder. I seem to remember they were all pardoned by the Queen [In fact, their sentences were commuted to 6 months' imprisonment.] ... The date was 1884, and the 17-year-old boy killed was Richard Parker. See The Custom of the Sea: A Shocking True Tale of Shipwreck, Murder, and the Last Taboo (ISBN 0471383899 (bn.com, isbn.nu)) and http://wings.buffalo.edu/law/bclc/web/dudley.htm.

LifeboatScenario?'s are so rare that most moral philosophers simply refuse to discuss them, saying that human morality is about likely circumstances and not aberrant cases (not a good argument in my opinion) or that human morality is about the rational choice in a situation and there may be situations to which there is no rational resolution (a very thought-provoking argument). I guess the point is that when it comes to morality, even if it's based on our extremely deceptive perceptions, we still hit the right answer upwards of 99% of the time. Most people simply don't care about LifeboatScenario?'s; I do because I'm an aberration. :-)

Anyways, the fact that many-worlds has such neat consequences in moral theory (and also time-travel logic) indicates to me that it's likely to be right. Good theories don't just solve the problem at hand, they solve problems you couldn't possibly foresee them solving. -- rk

Which brings up another point: human beings, apparently, perceive a perpetually changing slice of the wavefunction, and the thoughts of human beings are, apparently, inferences from their perceptions. Under many-worlds, therefore, the thoughts of human beings tend to be nonsense - unless appearances are deceiving. -- mb

'':-) I think nonsense is too strong a word. Deceiving and illusory are better." -- rk