This week-end, I will present at a humanist meeting on “Psychological Hedonism vs. (integrative design-oriented) value pluralism: architecture-based motivation”. Psychological hedonism claims that “only pleasure or pain motivates us.” (Stanford Encyclopedia of Philosophy). Psychological hedonism is a form of value monism (meaning one ultimate or intrinsic good). Value pluralism is the view that we have multiple, top-level, possibly incommensurable sources of motivation. Integrative design-oriented psychology is a way of understanding whole mind with methods drawn from AI (including artificial general intelligence) and philosophy. Architecture-based motivation is the idea that the mind produces many top level (intrinsic) motivators; it has motive generators.
We won’t deal much with ethical hedonism here.
I argue against psychological hedonism and in favour of value pluralism.
For people attending the meeting: this is background reading. I won’t be able to get through it all in the initial presentation. We can return to some of the skipped content during the moderated discussion. Several open-access resources are listed in the relevant readings section.
Man does not strive for pleasure; only the Englishman does
Psychological hedonism is a variegated collection of doctrines about human experience and behaviour. It goes at least as far back as ancient Greek philosophy, but it was much “helped” by English philosophers of the enlightenment. The tremendous (not always beneficial) influence of the English (particularly USA) has caused many psychologists, philosophers and the wider culture in the West to adopt, or at least be adversely influenced by, psychological and ethical hedonism.
In my talk, I will provide several examples of hedonism in cognitive science and psychology. An obvious example is the so called law of effect. To be sure, Skinner tried to rid psychology of mentalistic talk (including of pleasure) in his adaptation of this ‘law’, but he met his fate. There are also much more subtle and surprising cases where hedonism seeps into otherwise enlightening cognitive science and psychology, as if it were gospel. An example of this lies inside the otherwise very interesting theory of mirth by Hurley, Dennett and Ronald, in Inside Jokes. Their idea is that we are unconsciously motivated to seek mirth because it is pleasurable. The architecture-based motivation hypothesis described below is more parsimonious, and plausible.
The behavior of Westerners is (partly) aligned with hedonism not because hedonism is a scientifically true or valid theory. Rather, some of our apparent pleasure-seeking is due to Western culture having important hedonistic undercurrents. This illustrates that academic philosophy does have an impact on culture, and hence on behavior! (This is akin to a self-fulfilling prophecy)
However, culture is never homogenous or coherent; it’s full of contradictions. (See my post on coherence and sense-making.) Non-hedonistic values are also promoted (e.g., puritanical ones). And besides, culture (hedonism, puritanism, etc. ) does not have free range in shaping minds. So the impact of psychological hedonism is not as bad as it could be. But still, academics, and others expatiating about human nature, need to be careful about how they think about and describe human motivation.
I will argue that psychological hedonism is an impoverished, stale, and unfalsifiable (perhaps non-scientific) set of theories that cannot sustain a progressive research programme.
Andrew Moore, without going so far as to state that hedonism is not falsifiable, illustrates some of the moves hedonists make to protect their theory. He writes for The Stanford Encyclopedia of Philosophy:
The capability of hedonists to tell hedonic stories as to our motives does not in itself generate any reason to think such narratives true. To escape refutation by counterexample, motivational hedonists need to tell the tale of every relevant motive in hedonic terms that are not merely imaginative but are also in every case more plausible than the anti-hedonist lessons that our experience seems repeatedly to teach some of us about many of our motives.
But there are other forms of criticism than refutation by counter-example, including some from integrative design-oriented AI. An example of the latter is to assess whether an autonomous agent of human-like capability can be designed according to hedonic principles, and whether other theories can lead to better designs.
There is also the principle of parsimony. I argued in Cognitive Productivity: Using Knowledge to Become Profoundly Effective that hedonism is not a parsimonious characterization of human behavior. Hedonic theory always requires that one add to one’s explanation of behavior , “and s/he did it to seek more pleasure, or avoid pain”. This requires non-trivial computation of hedonic value (viz. utilitarian computation) in cases where it is much more parsimonious to assume that the behavior simply arises from an installed mechanism (e.g., a selfish motivator PDF). Evolution works by favouring the development of tricks (heuristics).
My (original?) claim that Occam’s razor cuts against hedonism should surprise you, because hedonism appeals to some people precisely because it is thought to be more parsimonious. I can understand non-AI people being fooled by the claim that hedonism is overall more parsimonious. I find it disappointing however that some AI researchers (connectionists) do not seem to have realized they’ve been duped. Connectionism and hedonism are not identical, but they are intertwined and I think motivated (biased) by similar cognitive preferences: to summarize and explain everything quantitatively (for some: because they know how to crunch numbers, they like crunching numbers, they have physics envy, and/or they have found had success with the techniques).
There is also the argument I raised in chapter 6 of my Ph.D. thesis, that preferences are non-transitive. The experiment would need to be more complicated than the following to better support my point, but suppose John is hungry and likes all kinds of fruit. He is given the choice between an apple and an orange. He prefers the apple. Given the choice between an orange and a banana, he prefers the orange. But given the choice between a banana and an apple, however, John prefers the banana. How strange. But this does happen. The fact that preferences are non-transitive undermines the claim that humans evaluate outcomes based on expected pleasure.
Of course, one can always salvage hedonism, which is why I suggest it is non-falsifiable. One might argue that hedonic judgments can be skewed by context (surely they can be). However one could also argue, more parsimoniously, that the choice (between some options, such as various fruit) is often arbitrary. That is not to deny that a computation of the relevance of each alternative is performed; it is only to say that the computation need not be a single number representing the value in a common (or exchangeable) currency. (If I correctly recall, in The Intentional Stance, Daniel Dennett has some good examples of arbitrary choices. When you are asked “What are you hungry for?” you basically start making it up. You weren’t hungry for anything in particular. Heuristics kick in.)
Pleasure vs. liking vs. wanting: conceptual and psychobiological considerations
Hedonic theory hinges on the concept of pleasure, or more precisely on a collection of concepts of pleasure. Studying pleasure calls for multiple methods: conceptual analysis, design-oriented research and empirical research (see the next section).
Here I recommend Pleasure (Stanford Encyclopedia of Philosophy).
I also find Gilbert Ryle quite helpful, and accessible, on the topic of pleasure. His Dilemmas deals with many philosophical dilemmas. In the chapter on pleasure, he points out that we know how to use words like “liking” and “pleasing”, but the abstract noun “pleasure” is quite problematic. Languages like English and French create nouns out of other grammatical categories, and in so doing trip lay and professional philosophers alike (and of course, psychologists, who often will not even notice they are on philosophical quicksand) who assume that these nouns must refer to some meaningful entity — “Reification fallacy”. Often, this fallacy terms engenders non-progressive debate, and wastes precious tax dollars that could be spent trying to build autonomous agents. Ryle’s The Concept of Mind is also helpful.
Aside. In one of his discussions of pleasure, Gilbert Ryle briefly puts forth a very interesting theory of emotions that is helpful in understanding the concept of perturbance. (In chapter 6 of my Ph.D. thesis I criticized Ryle’s theory. But I didn’t mean to suggest it was useless. I mainly meant to use it to illustrate a subtlety in the concept of perturbance.)
See also Footnote  about Minsky’s The Emotion Machine‘s treatment of pleasure from an AI perspective. Like others, he shows that pleasure is a polysemous term. But he takes an integrative design-oriented approach, arguing that the different meanings reflect the operation of different information processing mechanisms. (More on that in the next section.
Hedonistic theories appear implausible when one realizes that the concept of pleasure is so heterogeneous.
As an undergraduate, I worked in a “Brain Stimulation Reward” lab. (Thanks to two undergraduate NSERC scholarships.) A hedonistic hypothesis was assumed by many in that field of research: that pleasure (liking) entails desire (wanting). We were studying the medial forebrain bundle (MFB), which was believed to “subserve” pleasure and reward. I worked on a motivational priming effect: stimulating the lateral hypothalamus (LH) tends to trigger wanting the stimulation. These subjects could zap their own LH by pressing a lever. They learned to do this, and would often self-stimulate at a very high frequency. If the electrode is well placed, the zap is definitely pleasant in some important senses of the word. I noticed (as had others before me) that rats quickly cease to seek this pleasurable stimulation, particularly if the interval between available stimulation increases. (The behavior “extinguishes” quickly, to speak in terms of operant conditioning.) However, the experimenter’s zapping of the rats LH can reactivate its wanting — a priming effect. (Compare: having salted nuts can make you want more peanuts.) Obviously, the issues are more complicated than can be dealt with in one paragraph, for instance the stimulation may also produce pain (and the issues call for multiple experiments); but, in sum: the rapid “extinction” of LH self-stimulation argues against hedonism. If pleasure is the driving end of animal life, animals that can generate pure pleasure by pressing a lever should be addicted to it–LH self-stimulation should not extinguish rapidly.
In Dissecting components of reward: ‘liking’, ‘wanting’, and learning, Kent C Berridge, Terry E Robinson and J Wayne Aldridge argue for “three dissociable psychological components of reward: ‘liking’ (hedonic impact), ‘wanting’ (incentive salience), and learning (predictive associations and cognitions)”. This distinction between wanting and liking is an empirical argument against hedonism. (But like I said, Freudians and other hedonists can always reinterpret the data to suit their theories.)
The need for integrative design-oriented theories
None of the BIG and seemingly (though not necessarily) important psychological questions that have dominated Western philosophy, including the ones in this blog post, can be answered in isolation. They require an integrative design-oriented theory of mind.
Consider these questions:
- Do humans have free will? (Better: in what respects, if any, can ordinary humans be said to be free?)
- Is there an immaterial aspect to the human mind?
- Are we driven solely by pleasure/pain? (Better: To what extent are we motivated by pleasure? Even better: how do motive processes compute and weigh pleasure in decision making?)
- Are human beings rational? (Better: to what extent, or in what ways, are we rational?)
- What does it mean to be creative? (Better: in what respects are ordinary adults creative?)
In discussions on these questions, one quickly realizes that the key term needs to be defined: free-will, pleasure, rationality, or creativity. Often there is so much disagreement on the key definition that the discussion goes nowhere.
The problem is that defining a single term is never enough. One needs a theory according to which the various terms can be used coherently together. The concepts have to be linked together in a coherent theory of mind.
Take for example Newton’s theory of motion. One cannot simply have a discussion of mechanical force without understanding its relation to mass and acceleration. The various concepts involved in Newton’s laws must be understood in relation to each other and in relation to many data.
When psychological questions arise, such as the ones bullet-listed above, one does not need just any theory. One needs an integrative design-oriented theory. By integrative, I mean (partly) that it integrates multiple functions of the mind: conative, cognitive, affective and ancillary. I also mean that the theory must be interdisciplinary and multi-level (true to cognitive science). By design-oriented I roughly mean using methods drawn from AI (see this postscript file, which macOS will automatically convert to PDF for you).
The poverty of much empirical psychology of affect
In empirical psychology of affect, pleasure looms quite large. I would recommend in this context the excellent book, The Measurement of Affect, Mood, and Emotion by P. Ekkekakis. Most affect theorists agree that pleasure/displeasure is one of two or three dimensions that structure our experience. The other two are arousal (or activity) and potency.
Affective psychologists trace their empirical roots to the work of the number-crunching research psychologist, Charles Osgood, particulary his 1957 book The Measurement of Meaning. Strangely, neither Osgood, nor the major authors that I have read in the context of his work, mentioned Immanuel Kant. Yet Kant had provided the most important and detailed arguments regarding the structuring of experience, particularly with respect to space and time. One would think that a researcher who is talking about the fundamental structure of experience would at least acknowledge Kant and explain how his work relates to Kant’s. No such luck.
What strikes me about Osgood’s work on pleasure, and the work of affective psychologists that I have read, is the little a priori specification of pleasure and other structuring dimensions of experience. Osgood has barely a paragraph on one of his three dimensions, potency. Osgood basically delegates the definition to one of his RA’s, as the ‘”football playing factor’. No conceptual analysis to speak of. Nor does his later book, Cross Cultural Universals of Affective Meaning offer conceptual analysis. And yet in affective literature, many authors refer to that book as if one could learn something about the basic dimensions they assume! They leave it up to factor analysis to supply that meaning.
(Demonstrating the problems with affective science’s over-reliance on factor analysis would require a paper in itself. So I will be brief and merely suggestive in this paragraph. Admittedly, factor analysis can certainly augment conceptual analysis. But to replace conceptual analysis with factor analysis, as Fontaine & Scherer do in their 2013 book, is in my opinion a recipe for scientific disaster. Their book starts with an out of context quotation of J. L. Austin, as if Austin was opposed to conceptual analysis, which was obviously not the case. Austin’s highly influential book was an extensive exercise in conceptual analysis! Here’s a suggestive question: could Newton or Einstein have constructed their famous theories by administering questionnaires to ordinary folk asking them questions about weight, mass, velocity and energy? Obviously not. And in fact Einstein used conceptual analysis and thought experiments (philosophical techniques) to produce his theory of relativity. Similarly, and in contrast with much affective science, I think effective affective autonomous agents can be designed and implemented without relying on factor analysis. But they cannot do without conceptual analysis. Compare Footnote  regarding Marvin Minsky’s Emotion Machine book.)
For the specialist I will note that the criticism in this section applies not only to dimensional theories of affect (e.g., that of James Russell [PDF]), it also applies to the Componential Process Model of emotion. Scherer and Fontaine very clearly endorse Osgood’s 3 dimensions of affective meaning and experience, e.g., in their 2013 book, Components of Emotional Meaning. Their theory differs from simply dimensional theories of affect by adding many components to the underlying dimensions of affect.
My point is that, low and behold, psychology lacks integrative design-oriented theories in which pleasure plays a dominant role. That’s not a knock down argument against hedonism — very little integrative design-oriented work is being done in psychology. But I would argue that it is not a coincidence either.
Having said that, I think the Componential Process Model is in many respects very useful. And we can learn from the strengths and weaknesses of dimensional theories of affect too.
Pleasure vs. Pain
I won’t have time to discuss this in detail. Let it be said that Gilbert Ryle attempted succinctly to dispel the notion that the concepts of pain and pleasure work in the same way, let alone are opposites. He accomplished through conceptual grounds what neuroscience later made clear to some of us.
And yet we have in psychology of affect several theories (per above) that treat pain and pleasure as if they formed a continuum. This problem is not just a failure of psychometrics.
An alternative: Value pluralism
This section develops the concept of value pluralism first from an AI perspective, using Sloman’s concept of architecture-based motivation. (Admittedly, this requires more text).
An alternative to hedonism is that the human mind generates multiple top-level motivators. That means it has motivator generators. Such motivators are an inherent source of value. This is not to say that all motives are intrinsically valuable (some are derivative), but that some are. Gordon Allport was one of the first psychologists to focus on this. He coined the expression functional autonomy to refer to such motives.
Motivators are the basis of wanting. We want not necessarily because we expect the outcome or object of our wanting to be pleasurable, but because we are designed by evolution and learning to pursue certain behaviors and certain outcomes; and motive are necessary to achieving those things.
Take an example: the child playing with blocks. We might say that she is doing so because she finds it pleasant. But to echo Gilbert Ryle, this is just a manner of speaking; it does not mean that she is doing it for feelings of pleasure. She is doing it because evolution has designed her to play. Feelings of pleasure would get in the way of play. Ryle was proven right by the psychology of flow , the experience of flow, suggests that when we are in flow, we are focused not on ourselves, but on the activity out there.
Evolution has designed us to have mechanisms to generate many top level motives:
- to seek to be competent (i.e., to have effectance)
- to send “positive” social signals (that we are trustworthy, competent , etc.)
- to keep our blood sugar levels from falling below certain thresholds (hunger)
- to be attracted to some members of the opposite sex (homosexuality, etc., can also be explained in this way)
Not only are each of those bullets distinct from each other, implying manifold motivation. But each one of the items in those bullets is realized in the brain not as a single drive, but as a huge collection of motivators that get activated in certain circumstances. I.e., it’s not that we have a single hunger drive. The brain contains many motive generators that push us to eat. Similarly for drinking, social signalling, etc.
Nor do we have a single sexual drive. For instance, in human mating, the brain creates top-level motivators to make one interested in the other person. This motivator is intrinsically valueable. The teen-ager or adult does not necessarily realize that the “end game” is sex. Nor does the brain itself have an explicit or even implicit (and the latter is important) means-end link between the behavior, and sex. Again, we don’t need to explain interest in the other in terms of pursuit of pleasure. That extra step is not parsimonious. Interest can be decomposed into a number of motive generators, cognitive reflexes, executive processes, and so on —mechanisms that are required for autonomous agency.
That is the subtle concept of architecture-based motivation (PDF). I discuss effectance, or motivation for competence, in Cognitive Productivity: Using Knowledge to Become Profoundly Effective. It is obviously an important topic for CogZest, since we are mainly concerned with the development of competence, or meta-effectiveness, of which effectance is a component.
I think this view of what drives behavior is more biologically plausible and more promising as a way of designing autonomous agents.
For an example of an integrative design-oriented theory, see part 2 of my first book, Cognitive Productivity: Using Knowledge to Become Profoundly Effective.
Value pluralism in philosophy
The foregoing discussion of value pluralism was not drawn from mainstream philosophy. It comes from AI. However, value pluralism is discussed chiefly outside of AI, mainly philosophy and political science as far as I can tell. See the Stanford Encyclopedia of Philosophy entry on Isaiah Berlin’s section on value pluralism.
Isaiah Berlin emphasized the incommensurable of plurality of values. To quote from Wikipedia :
With his account of value pluralism, [Isaiah Berlin] proposed the view that moral values may be equally, or rather incommensurably, valid and yet incompatible, and may, therefore, come into conflict with one another in a way that admits of no resolution without reference to particular contexts of a decision. When values clash, it may not be that one is more important than the other: keeping a promise may conflict with the pursuit of truth; liberty may clash with social justice. Moral conflicts are “an intrinsic, irremovable element in human life”. “These collisions of values are of the essence of what they are and what we are.” For Berlin, this clashing of incommensurate values within, no less than between, individuals, constitutes the tragedy of human life. Alan Brown suggests, however, that Berlin ignores the fact that values are commensurable in the extent to which they contribute to the human good.
The discussion above of intransitivity of preference is relevant to this. I’ve just noticed another pertinent article from the Stanford Encyclopedia of Philosophy: Incommensurable Values (https://plato.stanford.edu/entries/value-incommensurable/). I’ve only read bits of it so far. I notice there is no mention of AI, but there is reference to Herbert Simon, who is one of the founding fathers of AI, in particular to his notion of satisficing. (That is a heuristic rule of decision-making. For other rules see the AI references below.)
(Aside: Gilbert Ryle, Isaai Berlin and Aaron Sloman were at Oxford at the same time. I am not an expert on Berlin, but I assume he was influenced by Ryle on conceptual analysis. Sloman’s theory of mind entails incommensurability of value —we used to talk about it together in the 1990s. At AISB-2017, I asked Aaron whether he ever attended a lecture by Berlin and he said no. Those three luminaries may have independently come to similar conclusions, which they pursued in very different ways.)
I realize many will dismiss this topic as being purely academic —worse (to them), just philosophical. I wouldn’t. It has many explanatory, practical and normative implications, in line with Plato’s call to “know thyself”. For instance, young people form beliefs about the purpose of living that shape how they make decisions throughout their lives. Reflections and attitudes towards purpose affect the major purposes we choose as worth pursuing.
In Beyond Pleasure and Pain, Tori E. Higgins argues that people want and need to be effective, and that this can and should often trump the desire for pleasure. While I am more pluralistic about value than Higgins’ is (cf. architecture-based theory of motivation), I of course agree with him that the pursuit of effectiveness is very important. Hence the subtitle of my first book, Using Knowledge to Become Profoundly Effective.
If Higgins and I are right then the pursuit of effectiveness is not only natural but is a route to happiness. And motivation for it, i.e., effectance, whether deliberate or implicit, should be nurtured.
Having said all this, I’ll be the first to admit that these issues get complicated very fast in professional psychological circles.
integrative design-oriented literature (AI+)
- architecture-based motivation by Aaron Sloman.
- Cognitive Productivity with macOS®: 7 Principles for Getting Smarter with Knowledge (search for “architecture-based motivation” , “meta-effectiveness” and “effectance”; also chapter 15).
- Emotional Agents (Ph.D. Thesis) by Ian Wright (who now has a second Ph.D., this one in economics). From the abstract: “A design hypothesis is proposed, the currency flow hypothesis, that states that a scalar quantity form of value is a common feature of adaptive systems composed of many interacting parts.”
- Goal Processing in Autonomous Agents. Chapter 3 contains a detailed design-oriented and conceptual analysis of goals. Chapter 4 specifies goal processes. Chapter 6 has an argument against utility theory.
- Minsky, M. L. (2006). The emotion machine: Commonsense thinking, artificial intelligence, and the future of the human mind. Hew York, NY: Simon and Schuster. As noted in footnote 3, this discusses pleasure in a way that is very different from the rather uniform story of dimensional affect theories of psychology and of the Componential Process Model.
- Higgins, E. T. (2011). Beyond Pleasure and Pain. OUP USA. Required reading.
- Dissecting components of reward: ‘liking’, ‘wanting’, and learning, Kent C Berridge, Terry E Robinson and J Wayne Aldridge. Required reading.
- Ekkekakis, P. (2013). The Measurement of Affect, Mood, and Emotion. Cambridge: Cambridge University Press.. Very helpful resource.
- BBS article: The Selfish Goal: Autonomously operating motivational structures as the proximate cause of human judgment and behavior. Useful, but the concept of goal is not the right one. They also need a theory of architecture-based motivation.
- Why Buddhism is True by Robert Wright. A long form argument for the impact of pleasure and pain. Also: an argument for why such impact should be minimized. And argument for a way to free oneself from it. I would say that Buddhism is in desperate need for a theory of architecture-based motivation. Compare ch. 15 of Cognitive Productivity. I will likely blog about that some other day.
- Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The Adaptive Decision Maker. That there are many rules for decision making. Decision-making is not just simple utility computation. Very helpful book.
- Value Theory (Stanford Encyclopedia of Philosophy)
- Value pluralism – Wikipedia
- Pleasure (Stanford Encyclopedia of Philosophy).
- Paradox of hedonism – Wikipedia
- Gilbert Ryle’s : Dilemmas and The Concept of Mind. Required reading.
- Pleasure – Lisa Shapiro (Editor) – Oxford University Press (I’ve not read all of this).
In line with my learning from art project, I try to always think of a set of works of art that are helpfully related to content I develop (or delve).
For some of this topic, see A Bit of Art for Value Pluralism – CogZest.
Thanks to Carol for proof-reading. Thanks to anyone else who will notify me of other infelicities in the text — I am sure there are many.
1. If a psychoanalyst were to read this post he might think it is just a defence mechanism of mine, attempting to deny what my id really wants (i.e., pleasure). A social psychologist might counter, “No! This entire blog post is just social signalling.” If there is truth to the latter, then psychological hedonism is false.
2 This stance is consistent with the argument in Simple Heuristics that Make Us Smart, which appeared a few years after my Ph.D. thesis, and The Adaptive Decision Maker (1993).
3 Section 4.4 of Marvin Minsky’s book The Emotion Machine deals with polysemous terms, which he calls “suitcase terms”, in psychology. He’s not the first to discuss suitcase concepts, but it’s nice to read this from an AI perspective, and in a book that deals with affect and that is very accessible. In Section 9-4, he points to four different contexts in which we use the word pleasure that differ significantly from each other: satisfaction, goal suppression, exploration, and relief. He argues that these different meanings are an important obstacle to the psychology of affect. I agree. (His book has 62 references to “pleasure”. If you’ve read this far, you might find his book worth reading. I have.)
- 2018-11-21 11-19 AM. extended my rant about factor analysis.
- 2018-11-21 4:34 PM. Added a footnote about and reference to Minsky’s very readable book, The Emotion Machine.