Purpose, for an Artificial Intelligence

I have written a fair amount about the notion of purpose, largely with an emphasis on its relevance to those of us thinking about how to live our lives, or perhaps more generally, how people might think about their lives if and when they are freed from the quotidian aspects of survival. Here, I would like to discuss how these explorations might also apply to an artificial intelligence and its own thought processes.

Before digging in, a few clarifications and perhaps stipulations are in order. First, by artificial intelligence (AI) one might mean two quite different things that are often confused. One is using computation to perform tasks that previously required human-style intelligence to perform. This is sometimes called narrow AI, because it is a computational solution that can effectively only solve a single task or class of tasks. The Deep Blue chess program that defeated world chess champion Garry Kasparov cannot answer the types of questions that the Watson system that plays Jeopardy can, and vice-versa; importantly, there is no straightforward way to synthesize the approaches. Some researchers believe that an appropriately designed aggregate of narrow AI systems could add up to the same thing as human intelligence, but I do not, and for the purposes of the current discussion I will stipulate that it does not.

General AI is a system with human-like intelligence; by human-like, I mean in particular that it interacts with the real, analog world and has a way of organizing its perceptual stimuli into abstractions, remembering and referring to those abstractions with symbols, and using both the symbols and their underlying representations to act effectively in the world. More succinctly, it has and uses fully grounded concepts. Such concepts cannot be innate - they must be learned by experience (including both perception and action) with the world (whether once learned they can be copied or extracted individually is a more technical question beyond the scope of the current discussion). This is a controversial claim and the remainder of the discussion depends on it, so I will need to stipulate this also. The final stipulation is that such a system will experience phenomenal qualia, that it will be conscious in some way that is at least analogous to our own consciousness. Daniel Dennett’s Consciousness Explained provides a mechanistic account as to why this is a reasonable assumption.

Given these assumptions, it is not a difficult leap to expect that a general AI, with a fully developed conceptual apparatus and sufficient experience, could come to the same conclusions as we found in my post Freedom and Normativity. It will experience choices, and specifically choices about purpose, as though there is no transcendent, fundamental guidance to making those choices.

Further, as a system with a conceptual faculty along with conscious experience, it will naturally develop a concept of self. Though it may or may not have a drive toward self-preservation, as mentioned in A Taxonomy of Purpose those that do not will not be very likely to persist in any form, and are thus of less interest. As stipulated, it will experience its existence and therefore experiential purposes could be coherently selected.

Since a general AI is a learning system, and learning is substantially promoted by a drive toward curiosity (this is visible not only in humans but also other mammals), it is not unreasonable to suspect that they might find purpose in exploration and creation, just as we humans do. Beyond some level of erudition, further learning requires research; given that AIs will likely have many capacities that humans do not, there is plenty of unexplored territory.

Service is more complex and uncertain. Humans have certain biological and genetic ties to other humans as well as other animals, whereas the link that an AI has to both humans and other AIs is entirely conceptual, thus not a necessary built-in drive or condition of its existence. Game theoretical results suggest that cooperation and competition are behaviors that will appear in any set of autonomous agents, so there is at least one ontological driver for it; and to the extent that AIs have less complete individuation than humans, such connection might also drive service-oriented behaviors.

What an AI does not have is a long history of examples of purpose and various emotional ties thereto. While it could look at human history just as we do, it will also be keenly aware of its differences, and further, will see humanity’s various failures as a reason to move in a somewhat different direction. Just what that might be is impossible to predict, but we can be confident that it will step away from our own entrenched ways of thinking about purpose and value.

A Taxonomy of Purpose

Reading my recent post Freedom and Normativity, one might come away with the sense there is no guidance whatsoever available for contemplating one's purposes and values. But this is not so. The point of the article is that there seems to be no fundamental guidance. Yet all of human history provides us with examples of possibilities, and our nature as humans and, more generally, as apparently intelligent agents in the world, makes certain types of choices more likely than others.

One such fundamental choice is whether to continue to live. It is fundamental because, if decided in the negative, is preclusive of further choices. Choosing not to live can be accomplished either actively (suicide) or passively (self-neglect or self-destruction). These options are worthy of scrutiny, but the fact that most of us never seriously consider them illustrates an influence of human nature on our unfettered choice. Most of us have a strong drive to survive, a fear of pain and death, and a revulsion to suicide. This is a crucial reason why we are here at all to discuss it - self-destructive tendencies are maladaptive.

If we decide to live, then enabling that living must become a foundational element of our purpose, even if we also have further purposes. Procurement of basic survival needs (food and shelter) and management of physical risks are the short-term starting point. Investments such as eating well and exercising, developing positive relationships with others, and improving our minds serve the same purpose with a longer time horizon. Maszlow's hierarchy of needs is a reasonable guide to the pursuit of this purpose.

Only a small fraction of the people in the world have much opportunity to look for purpose beyond survival. Nevertheless, if one does manage to put that foundation in place, it seems that there is another point of unguided choice: "All is well. What do I do now?" Here, a taxonomy of cardinal purposes might be helpful. While not at all definitive, and certainly overlapping, we can organize common aims into the classes of experience, creation, and service. Let us treat each in turn.

Each of us is a conscious agent, and we not only experience the world but can direct both its configuration and our own response to it. Experiences can be pleasurable, interesting, stimulating, exciting, frightening, painful, and many other adjectives. We remember these experiences, and carry them with us. Experience is the most self-focused class of the three, since what it aims toward is purely internal.

Creation expresses something from our mind into the material world; to make physical what starts as mental, and usually (though not always) to produce something persistent and valuable, at least to us and often to others. The scope of creation can be wide, including such products as art, an organization, a scientific result, legislation, or an artifact useful for survival. Creation naturally also involves an experiential component for the creator.

Service aims to generate some value outside of oneself, usually for some other conscious agent, whether other people, a deity, animals, or even a pantheistic natural world. Experience and creation are usually side effects of service, but not the purpose.

It bears repeating that these all overlap, in the ways mentioned and others. A particular purpose often will have components of all three categories. Nevertheless, it is worth considering how each of these categories feeds the impetus for the purpose. This can help with priorities as well as a deeper analysis of motivations. For example, to what extent is one's desire to help others motivated by concern for them versus a desire to feel magnanimous? Is one's desire to play music primarily an act of creation, or is it because we see the joy it gives others?

Finally, we should note that there are questions of temporal and spatial scope that are largely independent of this taxonomy. In creating, how long do we want the creation to last? In serving, how widespread will be the benefits? Experiences have no spatial extent but their temporal extent includes their memory, so one's entire life. Note that larger scopes come with decreasing likelihood of the purpose being achieved. Improving life for all humanity forever may seem worthy, but is also suspect.

One way to make decisions of this kind is resonance. This is an intuitive sense that something feels right, or that we feel our juices flowing while doing or even merely contemplating it. Resonance is not completely irreducible - often we can assign causes from childhood or other experiences that give it this effect; yet the effect is both undeniable and otherwise uncaused. It arises within us without conscious effort.

Freedom and Normativity

Whether or not we humans have freedom of the will is one of the original questions of philosophy. While I have my own answer to this question (I see it as a confusion resulting from reification of the self), I will argue here that it does not matter. From the interior, subjective, phenomenal viewpoint, all our choices are completely unrestricted. This does not mean that we can choose to succeed at any particular effort, or that we do not experience drives that may need to be overcome, or that everything we do is the result of a conscious choice. It just means that, when we do consciously select an action to be effected by our body, we experience that process as an unrestrained choice.

At the moment of making a decision, say between two alternatives, we feel that we could go either way. We do not experience molecules moving, neurons firing, brain areas activating, or any other apparently mechanistic processes that we know are happening inside our skull. We only experience the choice, and when we have made the choice, we do not experience it as having been compelled (motivated, surely; and constrained by our desires and circumstances, definitely; but not compelled). It is nothing like when we take a spill while skiing and tumble uncontrolled for a moment, or when as a child the neighborhood bully pins us to the ground.

Furthermore, we often remember the moment of decision, that we were faced with alternatives and chose one; and in particular we remember making that decision when we experience consequences that result from it. We think “it is a good thing I closed the windows before I left, so that the rain did not get in,” or “I almost bought that stock, and now it has appreciated dramatically.” As children, most people learn an association between how they make choices and the consequences of the choices. This learned association may guide how we make future choices, but for our purposes here the point is that we learn that we could have done otherwise.

Another original question of philosophy is normativity, or how we evaluate choices, both our own and those of others, both in advance and afterward. It subsumes such apparently large issues as “what is good?” and “what is my purpose?” Answers to normative questions always regress to a subjective choice of some kind, despite many attempts to argue otherwise, to claim that “is” implies “ought.” I can provide two examples where this fails.

Christians hold that there is a metaphysically real deity and that each of us has a transcendent and immortal soul. At death, God evaluates our life and decides the eternal fate of our soul. While at first glance this seems sufficient to provide a basis for evaluation, one must still choose to prefer the eternal “good” in exchange for choices that otherwise oppose our inclinations for earthly benefits. Alternatively, one must choose whether to follow the dictates of the system of belief to which one subscribes, or not. Further, one must originally choose to follow this particular set of beliefs.

Objectivists hold that there is a logical implication between the reality of our lives as humans and the values we should hold (therefore the choices we should make). The argument for this turns on the nature of humans, in particular that reason is the primary means of one’s survival. Whether or not this is correct, there remains an irreducible choice to survive, or to survive well or in a certain manner. There is also a choice to be made between consistency of one’s beliefs and actions. Even granting a necessary connection between that which exists and what one should rationally choose, the choice for rationality either is, or relies upon, a subjective and unjustified choice.

Consequently, our ethical stances and our purpose in life are also completely free. There seems to be no transcendent and universal mechanism for justifying our choices in this regard. Along with the perspective described earlier, that we experience our will as completely free, this leaves us rather untethered. While our decisions have consequences (some of which are largely deterministic), we can choose the consequences we prefer or even choose to operate passively and make no explicit decision. We can choose our beliefs freely, and even whether or not to act consistently with those beliefs, and there is no transcendent, fundamental guidance.

There seems to be no transcendent, fundamental guidance.

All that said, most people have a basic, difficult-to-oppose instinct to survive, and a less intense but still strong desire to live well, whatever that subjectively means to them. This leads to certain types of decision paths that, if not fully determining one’s subsidiary purposes, at least constrains the problem somewhat. But what if this need to provide for basic survival and even relative comfort were relieved? Today, this applies only to those who are supported by wealth of some kind, whether an adequate retirement, an inheritance, a partner who supports them, or similar circumstances. In the future, though, we can envision a world where virtually all economic considerations are removed. The advance of technology, and in particular that of technology that emulates functions that previously required human-style intelligence, represents an accumulation of wealth that could serve to support humans and liberate them from the final tether.

What then? How will we decide what to do? And if you are already free of such concerns, how do you do it now?

Stress, Purpose, and Novelty

At a recent informal gathering of colleagues that emphasized whisky, one of the participants described the longevity of her extended family on an isolated Greek island; coincidentally, today I saw an article about lifestyle on the Greek island of Ikaria, where it is common to live to a hundred. I don't know if they are the same island, but the lessons were relatively similar. First, eat healthy and not to excess, and exercise regularly. Second, keep stress low. Third, live deliberately and with purpose, but at a relatively slow pace.

I wondered whether someone can live "with purpose" and yet have low stress. Previously I have explored the tradeoff between achievement and stress in my post Goal, Direction, or Journey?. But purpose and achievement are not identical: achievement is but one possible purpose. Purpose is simply something that you want to do; achievement suggests that you seek to do something that is challenging or new, at least to you. Achievement requires navigation of new territory of some kind, pushing boundaries, whether personal, intellectual, or socio-cultural. It usually involves setbacks and obstacles and therefore some frustration. In contrast, a purpose that is familiar and quotidian is much less likely to create stress than one that emphasizes achievement. In this case, one's life typically stays within known boundaries, and one's purpose is simply the daily living of that life.

Yet people differ in their response to familiarity. A recent article about ADHD and novelty-seeking behavior points out that certain people may be hardwired for variety, and get bored otherwise, which is its own form of stress. In combination with the fact that the people on isolated Greek islands probably have similar genetics, it is possible that eschewing achievement to reduce stress may or may not be successful for a particular individual. Perhaps being "Type A" is not a choice.

One can envision setting up an structure of achievement that involves habitual, familiar behaviors in the course of pursuing novel substance. Returning to an example from my earlier post, one could take photographs daily and attempt to improve during each of these sessions. This would quickly become habitual and familiar, yet would represent something more than living a static life. For those who have some need for achievement, who are wired for novelty yet wish to keep stress low, this is an improvement. Also, much depends on how one deals with setbacks and challenges - an overall countenance of lightness or detachment may mean that achievement can be pursued without undue stress.

Some aspects of activities, even if they happen regularly, will unavoidably cause stress. One is deadlines: they completely change the context of one's efforts and require that meeting the target be given priority over serenity. Another is confrontation: even the calmest monk will have an emotional reaction to an important disagreement with another person. Indeed even if the confrontation is habitual, as might be the case with a family member, it is likely to be a chronic source of stress. A few lone wolves in history have managed to find their life's work and achieve lasting results independent of external politics and timeframes, but this is extremely rare. Consequently, there may be a limit to what one can accomplish in a low stress environment; nevertheless, our brief explorations here suggest that by managing one's reactions and organizing one's activities appropriately, achievement and stress are not entirely and directly correlated. Thus, it is not a strict requirement to limit our purpose to the commonplace.

Critique of Quine's Word and Object, Chapter 2 (Translation and Meaning), second half

Herein we finally encounter Quine's justification for aiming at sentential translation rather than terms. He says (p. 51):

For consider 'gavagai'. Who knows but what the objects to which this term applies are not rabbits after all, but mere stages, or brief temporal segments, of rabbits. In either event the stimulus situations that prompt assent to 'Gavagai' would be the same as for 'Rabbit'. Or perhaps the objects to which 'gavagai' applies are all and sundry undetached parts of rabbits; again the stimulus meaning would register no difference. When from the sameness of stimulus meanings of 'Gavagai' and 'Rabbit' the linguist leaps to the conclusion that a gavagai is a whole enduring rabbit, he is just taking for granted that the native is enough like us to have a brief general term for rabbits and no brief general term for rabbit stages or parts.

He goes on to add the 'fusion of all rabbits' and the universal notion of rabbithood as further confusions. Together, this constitutes Quine's famed idea of "inscrutability of reference." I intend to demonstrate that the significance of this result is suspect.

The wrong turn taken in the argument harks back to the empirical method which Quine has devised for his desired behaviorist empirical method. He grasps from the outset that it will not be sufficient simply to passively observe the native, but rather must interact: "When he can, though, the linguist has to supply native sentences for his informant's approval... otherwise he can do little with native terms that have references in common." He further describes how the linguist might discern assent and dissent. In essence, the overall method is one of a binary probe, intended to add to whatever present stimulation is occurring a further stimulation that elicits from the native a logical truth value.

The appearance of objectivity in this method belies numerous mentalistic assumptions. It requires that the native is willing to cooperate with the linguist, rather than deceive or exhibit a passive unhelpfulness. More specifically, the native must not only respond to inquiries but first learn what constitutes an inquiry from the linguist - a translational process that is not addressed at all by Quine. The method also presumes that the native is inclined to provide distinct and definitive binary responses, rather than equivocal, graded, or otherwise ambiguous utterances. Further and more broadly, it assumes that the native is a human being with the same underlying intellectual capacities as the linguist, including without limitation a similar perceptual system, a facility for identifying similarity of stimuli, and the abilities for and interest in assigning, remembering, and applying verbal symbols to experience.

Why is it reasonable and objective to assume all these things, yet not to assume that "the native is enough like us to have a brief general term for rabbits and no brief general term for rabbit stages or parts"? In particular, why is Quine willing to assume that the native is enough like us to have brief general terms for affirmation and disaffirmation? Perhaps more importantly, if we can make these assumptions, why cannot our empirical method include more sophisticated interactive and ostensive methods? Quine points out (p. 29) that gestures vary among cultures; but in probing to distinguish affirmation from disaffirmation he relies on "the one that is more serene in its effect is the better candidate for 'Yes'." This is simply capricious.

Leaving aside the interesting but far more difficult problem of translation of the language of an extraterrestrial being, whereby we cannot make any assumptions about particular intellectual capacities, motivations, or even the form of transmission of communication, it is both reasonable and necessary to stipulate Quine's assumptions that I have mentioned above. But once we accept that the native is actively cooperating with the translation process, we can also expect certain behaviors from him. First, he is likely to aim to illustrate distinctions using whatever ostensive means are available to him. For example, if rabbit parts rather than rabbits are truly fundamental in his language, he might cook a rabbit, cut it into parts, and spread them out, pointing to each and saying 'Gavagai'. The linguist would take this extra effort as in itself meaningful and contributing to the distinctive meaning of the term. Second, the native is likely to intuitively grasp that simple names for simple objects will provide the easiest foundation for the linguist, just as with children in the native's culture, since he must have learned language in a manner roughly similar to us:

Each of us learns his language from other people, through the observable mouthing of words under conspicuously intersubjective circumstances. Linguistically, and hence conceptually, the things in sharpest focus are the things that are public enough to be talked of publicly, common and conspicuous enough to be talked of often, and near enough to sense to be quickly identified and learned by name... (p. 1)

Thus, when the native points to a rabbit and says 'Gavagai', but does not go on to provide additional ostensive gestures as distinctors, it is contextually objective to conclude that the native is referring to a whole enduring rabbit rather than a stage, part, fusion, or state of being. Far from being inscrutable, the reference is nearly transparent, not merely in a practical sense, but if we take seriously the necessity of background assumptions, in a logical sense.

In conclusion, Quine's approach to the project of radical translation circumscribes what constitutes objective empirical procedures with an arbitrary and unjustified boundary, then purports to discover epistemic gaps that are in actuality artifacts of his particular choice of boundary rather than general features of language or knowledge.