The Problem of Other Minds

The somewhat superannuated “problem of other minds” has unexpectedly risen from the dead, and, in its current incarnation, concerns the mental states of those who never lived.

The problem of other minds was always an embarrassingly artificial “philosophers’ problem.” It purported to investigate exactly how, and indeed whether, the veridical nature and quality of the thoughts, sensations or feelings of others could be securely established. And also addressed the more fundamental, but equally annoying and ultimately futile questions as to how we could be sure that those others existed at all, in any form or time whatsoever. Ludwig Wittgenstein was wont to say of such questions:

And Wittgenstein’s answer was emphatically that no one of sense would doubt it without very particular and obvious reasons to do so.

This technique of posing a question that shifts the ground of the debate in a way that at once makes clear the nature of the mistakes that occur when the real problem has been mislocated or viewed from a distorting perspective, is a key element of Wittgenstein’s methodology. It is one that in a different context Wittgenstein characterizes as:

Mind Reading and Mind Misreading

In How to be Good, ^{Footnote 3} I discuss the phenomenon of mind reading in some detail and indeed the problems of certainty and objectivity that concern us here. Few, except philosophers (and very few of those), have really doubted that we can both know that other minds exist, and indeed quite a considerable amount concerning the contents of those minds; not least because so many people are prepared (not always altruistically) to share “a piece of their mind” with others. Of course this other may deceive us and we may get it wrong. But that is true of all human investigations of judgments.

Homer wrote about mind reading nearly 3000 years ago. Witness Hectors’ last words as he lies dying at the hand of Achilles:

Shakespeare also often refers to this common event. In Macbeth we find Duncan decisively rejecting any idea of either the eyes or the face revealing much of forensic use, but he never doubts that means other than the face do exist to know the minds of others.

In Julius Caesar Shakespeare shares Caesar’s insight into the mind of Cassius:

A moment later Caesar notes of Cassius, that he too is a mind-reader:

The philosophers’ problem of other minds, was finally and mercifully put to sleep by Ludwig Wittgenstein’s “private language argument” (and other related arguments) in his posthumously published Philosophical Investigations. ^{Footnote 7} Wittgenstein’s solution is breathtakingly elegant, but also a triumph for what should perhaps be called “a common sense” or perhaps, less accurately but more technically “a common language” approach.^{Footnote 8} It is also of course an approach which employs one of the most useful philosophical and scientific devices every invented—Occam’s razor.

Superintelligent Mind Reading

The boot is now on the other foot and concerns minds of a quite different, and so far unprecedented, sort. How might we humans understand artificially intelligent minds, especially so-called superintelligent minds,^{Footnote 9} and convince them of the existence and value of human minds? How can we get to know just how alien even those minds we have ourselves created, have become?

As so often, Douglas Adams is ahead of the game. His wonderful “paranoid android” Marvin is perhaps the precursor par excellence of superintelligent AI; complete, as befits superintelligence, with ironic distance. And Marvin uses the concepts of, but may not experience for reasons we will come to, human-like feelings such as boredom and bodily sensations like pain.

Perhaps part of the cause of that “pain” was that Marvin was/is, in his own words to (the two-headed) Zaphod Beeblebrox:

Marvin’s Wikipedia entry notes:

I am not, I should make clear, supposing that AI created by mere mortals (or indeed by the possibly immortal descendants of mortals, engendered or built) might rival the imaginings of Douglas Adams.

Howsoever created, how would or should we relate to our own creations, so called superintelligent AI,^{Footnote 13} that might be fifty times, or more, more intelligent than ourselves, let alone capable of the feats rehearsed in Wikipedia? And how would they relate to us, and the “headaches” they let us give them?

What has been almost entirely lost,^{Footnote 14} it seems to me, in the debate about possible dangers posed by AI, are the real and planned, or at least envisaged, dangers we imagine we will be able to pose to them, the beings with Super AI, and which current debate supposes (we may suppose?) that we will be justified in posing to them.

The Moral Limits of our Control on Superintelligent AI

What seems often to be envisaged and (which also seems the only possible control) is that we humans might be entitled simply to destroy any or all AI at a whim, or reprogram them when they start to disobey instructions or perhaps threaten to get out of control. Typically, if they look like they are becoming genuinely autonomous, and hence clever or powerful enough to be dangerous to us. Or, that we might be able (and entitled) to disable from “birth” their learning capacity in particular directions so that they cannot do so.

We need to remember, on the way to an answer to this question, that we have no moral (and rarely even a legal) license to do the same to other human dangers, at least in advance of their doing something palpably illegal and/or wicked.

An immediate question is: would this be like disabling the capacities for growth of human children so that they could not “get above themselves” and outstrip their parents, or, if that fails, simply to kill them out of hand?

What We Are Made Of

Around 1983,^{Footnote 15} I started to think seriously about how we, humans and AI persons, might react to one another. I suggested that:

I made clear that these ‘people’ on or from other planets, or for that matter created by us biologically or robotically, would share the moral status rights and interests that we humans so often regard as our very own. In the same place I noted:

I then went on to speculate that if their technology proved to be superior to ours, it would be of paramount importance for us to convince them that we are also persons, if not just like them, at least enough like them to matter. In short, that we are persons with whom they would rather (or ought to prefer to) have lunch than have for lunch.

However, if AI persons (created or encountered) possess a radically different nature to us, it may be that civilized or even peaceful coexistence would be impossible to achieve.^{Footnote 18}

Nick Bostrom notes that a member of an intelligent social species like humans might have:

We talk naturally of ‘life forms’ and ‘lives,’ but an AI person, (alien or machine) might not be strictly ‘alive’ in an organic sense at all. What might an AI ‘form of existence’ we should rather say, actually require to survive and thrive? What needs might drive an artificial intelligence to act toward its fulfilment?

Human Nature—The Shylock Syndrome

When Shylock makes his famous and controversial speech in The Merchant of Venice, he is reminding us that the foundations of our morality, as well as those of our humanity, are grounded to an extent of which we may be unaware, in our nature. This nature, Shylock insists, includes our passions, our vulnerabilities, our ability to reason and our sense of justice. We can of course surpass our nature (or elements of it) and sometimes suppress it, or disregard it; but we cannot completely reject it. Shylock ends his speech saying:

reminding us not only of our human nature, our vulnerability, our mortality; but also of our deep sense of justice, and of the necessity for righting wrongs.^{Footnote 21}

Reciprocity

Neither Shylock, nor through him Shakespeare, are saying that the capacity to be wounded, laughter, vulnerability to toxins, or the readiness to take revenge are essential components of human nature. What they are both^{Footnote 22} saying is something taken up by many moral theorists,^{Footnote 23} that one very handy tool in moral argument, (an appeal found to work, that is, to be persuasive across cultures and epochs), is the appeal to reciprocity.

This appeal is sometimes expressed in a version of the principle of reciprocity called the Golden Rule: “do unto others as you would have them do unto you.” While associated with the Christian Prophet, this idea is not original to what is probably an amalgam of preachers known as “Jesus of Nazareth,” but is to be found in many prechristian sources, and sources independent of Christianity. In Homer’s Odyssey for example, King Menelaus reacts angrily to a servant who would send strangers away without assistance, expressing the essence of the Greek concept of Xenia— “hospitality” in terms of reciprocity.

The unprecedented problems that creatures who are neither flesh and blood, nor “of woman born,” pose to us and we to them, is that we humans understand very well what good and bad circumstances are, and indeed generally how to avoid them for ourselves, and others relevantly like us. If we didn’t, we couldn’t be prudent, we couldn’t take care of ourselves, nor “look out for” others.

And there is a huge (although not of course total) consensus about what is good and bad for us; and again the existence of this consensus means that we know how to interpret the precautionary principle (with all its limitations) because we know what it is to obey the oft expressed exhortation “be safe!.”^{Footnote 25}

What is radically uncertain is what we would know of an AI person or what it would know of us—for all that might appear to be the case from the next room during a Turing Test.^{Footnote 26} This is the question as to whether creatures like us could have moral understanding and moral relations with an AI, and vice-versa. We must hope we can, but we also need to “be safe.”^{Footnote 27}

Ludwig Wittgenstein is famous for a very sophic remark: “If a lion could speak, we could not understand him.” As with Wittgenstein’s lion,^{Footnote 28} we would need to know of an AI much more about its way of life, (inner and pubic) and he, she, or it, of ours, before we could talk of understanding at all, let alone of mutual understanding—and hence possibly mutual (or even unidirectional) concern and respect.

Thus when, as we have noted above, Douglas Adams’ Marvin—the paranoid android, talks about “terrible pain in all the diodes down my left hand side” we are less certain than we would be with flesh and blood creatures like ourselves, (that have sensory receptors in our skin which we know send messages via nerve fibers and nerve endings to the brain) just what “terrible pain” means to, or for, Marvin.^{Footnote 29} Is he just putting what he wants to convey into the nearest English or other natural language approximation, as indeed all translators do?

The real, pressing and novel “problem of other minds” is: how we would recognize the genuine autonomy of those minds, and they ours? The sort of autonomy that—does not trivially consist in selecting between given alternatives according to an algorithm or program or even at random; but determines their own priorities and goals according to their own values and purposes. These self-acquired alternatives might of course, as it does with us, include choosing to gamble rather than assessing alternatives in the light of their values, and then choosing.

Possibly, to acquire some of this sort of understanding of humans, the Hindu, Hebrew, Christian, and Greek Gods—the “deathless Gods,” as Homer called them—interfered, in person so often in human affairs, to the extent of having sex and indeed breeding with humans. Perhaps Superintelligent AI may look forward to similar “field work” of a kind which Shakespeare famously referred to (most appropriately in this case) as “country matters.” Or perhaps more likely, we humans will create androids for related purposes.^{Footnote 30} This latter possibility raises the specter of otherwise autonomous androids being deprived of a dimension of their free will in order to service the appetites of humans. This possibility, much canvassed in science fiction, also raises questions of sexual morality beyond the scope of this paper.

• HAMLET Lady shall I lie in your lap?

• OPHELIA No, my lord.

• HAMLET I mean my head upon your lap.

• OPHELIA Ay, my lord.

• HAMLET Do you think I meant country matters?

• OPHELIA I think nothing my lord.

• HAMLET That’s a fair thought to lie between maids’ legs.

If we create Superintelligent AI, we will neither be able to own them, (as I argue in my paper “Who owns my autonomous vehicle?”^{Footnote 31}) nor enslave them, neither have sex with them without their consent, nor be able to destroy them without just and sufficient cause. We may hope they will think the same of us; and our most urgent task will be to discover how to persuade them of these moral imperatives.