Ben Goertzel
October 25, 2004
Clearly most of the ethical principles that people live by (or pretend to live by) are merely culturally-specific conventions of behavior. But it’s tempting to look for an ethical core that goes beyond any particular culture or belief system. In a transhumanist context, one even wants to find an ethical core that goes beyond any particular species, or any particular type of mind or organism.
One strategy for seeking an “objective” grounding for ethics has been the evolutionary approach. Researchers in this vein try to explain why apparently-altruistic behavior would emerge in organisms evolving via a “selfish-gene” dynamic. This approach has led to a host of interesting insights, many of which are reported in Matt Ridley’s excellent book The Origins of Virtue. Ultimately, however, I think the evolutionary approach fails to get at the essence of ethical philosophy. I do think evolution is relevant -- but I think that, to get at the crux of the matter, one has to look at evolution and ethics as intertwined manifestations of the same deeper phenomena.
In this essay I discuss ethics, and the relation between ethics and evolution, from the perspective of “pattern philosophy” – the philosophical framework that states, very simply, that the universe is a collection of patterns evolving over time. I’ve written extensively on pattern philosophy in my prior publications so I won’t dwell on it here in any general way.
I’ll describe here a general property of the universe called “continuous pattern-sympathy,” and I’ll present an argument that, because of this property, an ethic of compassion is a natural and inevitable part of the universe. This argument not only explains the existence of compassion but it also explains some of the particular properties of compassion as we observe it among humans. I’ll then explore the relation between ethics and evolution by analyzing natural selection itself as a manifestation of continuous pattern-sympathy on the genotypic and phenotypic levels. Compassion as we humans experience it in our lives thus emerges as a particular example of continuous pattern-sympathy, intertwined with the continuous pattern-sympathy embodied in the evolution that gave rise to us.
What this all adds up to is an argument that compassion is a kind of “universal ethic” -- in that its existence follows naturally from a very abstract property of the universe. This argument doesn’t imply that we should all be maximally compassionate – it just argues that some degree of compassion is natural and inevitable.
To turn this analysis into a normative ethical theory, one must specify some goal, some value system. Suppose one decides, for example, to value “freedom, growth and joy,” as I’ve advocated in a recent essay. Then one may ask: What is the optimal degree of compassion that should exist, in order to maximize the amount of freedom, growth and joy in the universe? And one may ask: What is the optimal degree of compassion that I myself should manifest, in order to maximize this goal? The answer, pretty clearly, is that the optimal degree of compassion is neither zero nor maximal: rather, an intermediate level of compassion is almost certainly going to be optimal. Of course, quantifying this kind of conclusion in any useful way is far beyond the scope of contemporary science and mathematics. But even the qualitative conclusion seems worthwhile.
My own feeling is that the level of compassion displayed by the average human is somewhat below the optimal level for maximizing my pet goals of freedom, growth and joy. I believe that if humans were a bit more compassionate, we’d move toward these goals faster. But of course I can’t prove this – and to mount a careful argument for the point would require a detailed analysis of human history and human affairs, which I’m not going to undertake here.
At one time, I puzzled at great length over the paradoxical peculiarity of Buddhism, which both asserts that the world doesn’t exist, and asserts that we should be compassionate to other beings. If the beings aren’t real, I wondered, then why the heck should be bother being compassionate to them? I answered this question for myself in a personal sense a long time ago, but the ideas in this essay seem to form a crisper answer than I’ve come up with before. The crux is that “the world” as such doesn’t exist, but patterns exist, and patterns persist and grow gradually over time (“continuous pattern-sympathy”), and compassion arises naturally from this process.
OK, let’s move on to the detailed arguments. In this section I’ll define my key concepts of pattern-sympathy and continuous pattern-sympathy – and then show what these have to do with the ethic of compassion.
Let’s say that a dynamical system displays “pattern-sympathy” to the extent that, when a pattern appears in the system, this pattern tends to continue into the future (with a probability greater than would have been the case if the pattern had not appeared in the system in the past). Some systems display generic pattern-sympathy; others display pattern-sympathy only with respect to patterns in some particular category (so that when a pattern in the right category appears in the system, this pattern tends to continue into the future).
Furthermore, let’s say that a dynamical system displays “continuous pattern-sympathy” if, among patterns in the system, continuance over the long-term is reasonably well-predicted by continuance in the short-term. This is a little more than just saying that patterns tend to propagate themselves. It’s saying that patterns tend to propagate themselves, and they tend to do so gradually over time.
Compassion, the root of all ethics, appears to be a trivial and natural consequence of the property of continuous pattern-sympathy. Suppose we have a population of minds, each one considered as a bundle of patterns. And suppose that the dynamic by which these minds change over time obeys the principle of pattern-sympathy. This means that each pattern in each one of the minds “wants” to perpetuate itself. Then it follows that a mind will want to ensure the survival and productive activity of other minds that share a lot of its patterns. In other words, it’s not that “I” am compassionate toward “you” – it’s that a large number of the patterns in my mind are compassionate toward their clone or near-clone patterns in your mind. This explains very neatly why compassion, in practical human life, is roughly proportional to similarity.
Continuous pattern-sympathy means: of all the patterns that exist right now, the ones that are going to persist into the long-term future are generally going to be the ones that persist into the short-term future. And this means that survivor patterns are going to be ones that gradually increase their intensity as patterns.
In the case of patterns that achieve their intensity via repeatedly occurring in several different substrates – for instance patterns that occur in many different organisms, or many different minds – then clearly the long-term persistence of a pattern is going to be achieved via, in the short term, maximizing the number of organisms or minds containing the pattern. Ergo, in a dynamical system displaying continuous pattern-sympathy and patterns that achieve their intensity via repeated instances, one will see pattern-bundles (minds, organisms, etc.) that seek the continuation and flourishing of similar pattern-bundles.
But of course, similarity is not the only driver of compassion. We are also generally compassionate, for example, toward beings with whom we’ve been substantially involved, whether or not we are similar to them. This is because our involvement with the other being has led to the emergence of patterns between that being and ourselves, and these emergent patterns also want to persist.
Now I’ll tie these themes in with evolution – first reviewing the pattern-theoretic perspective on evolution, then tying pattern-theoretic evolutionary theory in with pattern-theoretic ethics as discussed above.
Evolutionary systems tend to display continuous pattern-sympathy. This is because “natural selection,” at bottom, is nothing other than a particular instance of the general phenomenon of continuous pattern-sympathy. To see this, consider the paradigmatic evolutionary phenomenon: the evolution and adaptation of species via natural selection. I’ll treat the evolution and adaptation of species in two stages: first purely genetically, and then in terms of genetic/environmental interaction. (The discussion in the next couple paragraphs follows the ideas given in my 1993 book The Evolving Mind, but my present formulation is different in detail and I believe crisper and clearer. However, my 1993 treatment is worth looking at due to the more expanded discussion and the many relevant, concrete biological examples discussed there.)
Firstly, I define the “potential” of a certain gene at a certain time as the number of instances of the gene that will exist in the long-term future, divided by the number of instance of the gene that exist at the present time. This may be roughly estimated by: the number of organisms containing that gene that will exist in the long-term future, divided by the number of organisms containing that gene now. It is then quite clear that one way for a gene to have a lot of potential is for organisms containing that gene to have a lot of offspring. This follows from the exponential mathematics of reproduction: more offspring now leads to lots more offspring later. Historically, Malthus highlighted the exponential mathematics of reproduction, and Darwin extrapolated the consequence of this: the set genes with long-term potential can be roughly estimated as the set of genes leading to short-term reproductive success. Or in the language I’m using here, genes – and sets of genes, and patterns among genes – are patterns that display continuous pattern-sympathy. The insight of natural selection theory is not that genetic patterns tend to persist over time (which is true but completely obvious), but rather that, due to the exponential nature of population growth, the genetic patterns that will persist over the long term can be predicted by looking at which genetic patterns proliferate over the short term. This distinction between time-scales is somewhat blurred by the common formulation of evolution as “survival of the fittest.” A better formulation, in my view, would be “long-term survival of the short-term survivors,” where the “survivors” in both the short and long term are not organisms but genes and patterns of combinations of genes.
But even this only gets at part of the phenomenon of the origin and adaptation of species – because it leaves out ecology. What ecology says is that in some cases the “survivors” are not just genes or patterns of combinations of genes, but they may be purely phenotypic patterns. Of course some phenotypic patterns may display continuous pattern-sympathy simply because they correspond with particular genetic pattern. But phenotypic patterns may also display continuous pattern-sympathy for another reason also: co-adaptation. When a genetic pattern persists and gives rise to a certain phenotypic pattern, this phenotypic pattern then forms part of the environment to which various organisms adapt, and it encourages the creation of other phenotypic patterns that “complement” it (in the sense that the original phenotypic pattern and the new ones give rise to significant emergent patterns). But if genetic pattern G has given rise to phenotypic pattern P, which has encouraged emergence of complementary phenotypic pattern Q, then even if genetic pattern G should disappear, there will be a tendency for a new genetic pattern H to appear, giving rise to a new phenotypic pattern P’ that is also roughly complementary to Q. For example, the existence of predators with certain properties may lead to the evolution of prey with certain properties – but then even if the original predators are extincted, the continued existence of these prey may lead to the evolution of new predators sharing properties with the original ones, in spite of genetic dissimilarity to the original ones. The relevant properties of the original predators are a phenotypic pattern that has persisted itself over time, without corresponding persistence of the underlying genetics, via co-adaptation. Co-adaptation dynamics leads to a tendency for phenotypic patterns to propagate themselves independently of their genetic basis – and this dynamic has the same exponential-growth dynamic that Malthus observed on the organism level and Darwin extrapolated to the gene level (though arguably with a considerably smaller exponent). So phenotypic patterns, independently of genotypic patterns, may display continuous pattern-sympathy as well.
So what evolution by natural selection really comes down to is: Long term-survival of the short-term surviving genotypic and phenotypic patterns.
Now, what does this tell us about ethics and compassion in humans and other evolved organisms? Ridley and other classical evolutionary-ethicists focus on the genetic roots of human ethics. Basically, the argue that apparent altruism in humans and other organisms is a consequence of the phenomenon in which a gene or gene-combination, existing in multiple organisms, maximizes its short and long term survival potential by inducing cooperation among the various organisms that host it. This is a valid observation and is doubtless a critical part of the origin and development of human and animal ethics, but I think these authors tend to underestimate the importance of purely phenotypic pattern-sympathy induced via co-adaptation.
The co-adaptive aspect of compassion is simply that, once a community of compassionate entities exists, there is a strong bias for the evolution of new compassionate entities. In a community of nasty bastards, patterns that are stuck in an organism displaying compassion will likely get screwed. In a community of compassionate individuals, patterns in an organism displaying compassion will get rewarded, because they’ll get to enter into a community-wide exponential pattern-spreading dynamic. Thus, compassion itself – independent of any genotypic basis for compassion – will tend to propagate itself. Compassion itself, as a trait of organisms, is a particular pattern that spreads exponentially according to the logic of continuous pattern-sympathy. And it spreads faster than would be predicted via looking at gene-level dynamics alone.
Normative and Transhumanist Implications
I’ve argued for the ethic of compassion as a natural consequence of the fundamental nature of the universe. Not as a consequence of the particular nature of our physical universe, but as a consequence of the nature of basically any sensible universe – any universe displaying the property of continuous pattern-sympathy. I’ve also argued that compassion, as a property of humans and other biological organisms, emerges naturally from natural selection, which is itself a manifestation of continuous pattern-sympathy.
One thing I haven’t done is to give any kind of particular ethical guideline. OK, compassion is good – but are some kinds of compassion better than others? How bad is eating meat, as opposed to failing to send your extra dollars to Darfur and spending it on a night at the movies instead? How important is it to ensure that humans survive the next million years, as opposed to promoting the ongoing development of more and more advanced mind- and life-forms? Unfortunately or fortunately, I don’t think that abstract philosophical or scientific analysis will ever be able to answer this sort of question.
What I’ve given so far is basically a descriptive ethics: I’ve described the kind of ethics that seems to exist in the universe, and explained why it’s here in terms of other, non-explicitly-ethical aspects of the universe. To turn these ideas into a prescriptive ethics one needs to articulate some particular value system. As I noted above, my own value system centers around the notions of freedom, growth and joy. Given this particular value system or any other, one may ask how much compassion, or what kinds of compassion, are likely to provide maximum value. And in the context of my own value system, it seems clear to me that if we assume the universe can only contain a finite number of patterns, then there is some optimal degree of compassion which is greater than zero but less than the maximum.
Growth, in a finite universe, is clearly not maximized by maximal compassion: maximal compassion breeds stasis, because it means that patterns that exist will very strongly reinforce themselves, preventing new patterns from forming. As Nietzsche observed, in a finite universe, progress requires some degree of hardness: it requires letting old patterns die so that new and better ones may form.
Freedom also seems incompatible with maximal compassion: in order to provide choice, we must allow minds to choose to be evil if they want to, at least to an extent. At very least, we must allow minds to be evil to themselves -- otherwise we’re not really allowing any kind of choice at all.
Of my three basic values, it seems that only joy is compatible with maximal compassion. If joy is the only goal, then everything in the universe can be maximally compassionate to everything else, no matter whether the universe is finite or not – there’s no growth and no freedom, just a pulsing field of radiant bliss!
We thus draw the conclusion that the root of “evil” – if evil is defined as incomplete compassion – is either the finite capacity of the universe, or the ethic of growth that seems to be embodied in the universe (alongside the ethic for real but partial compassion). Only in an infinite universe can we have growth and compassion side by side without contradiction.
This line of thinking has interesting consequences for the puzzle of “Friendly AI.” Put briefly, this puzzle asks how one might create a superhuman artificial intelligence, with the ability to modify itself freely and to assimilate all the matter in the universe if it so wishes – yet create this AI in such a way that the survival of the human race isn’t threatened. The good news is that any AI one creates is likely to be compassionate in a sense, since compassionate is intrinsic to the universe. The bad news is that compassion, as embodied in the universe, is a highly abstract thing. The fact that patterns tend to compassionately cause like patterns to persist isn’t much consolation if you’re one of the patterns that happen not to persist. There seems to me little hope of creating an AI that, once it becomes unboundedly intelligent and powerful, will care more than a little bit about persisting the particular patterns that are human beings. So the key to Friendly AI may well be engineering a situation in which caring a little bit is enough. In other words: make sure the AI has a really big universe to play in, so that it doesn’t need to annihilate our patterns in order to make room for its. In this case just a little bit of specially-focused compassion on us humans will be enough to keep us around.
The issue of Friendly AI highlights the weakness of any approach to ethical philosophy as general as the one I’ve given here. To answer any particular ethical question – even one as seemingly general as whether it’s ethically desirable for humans to survive the next millennium – requires one to get a lot more fine-grained than seems possible to do in the language of “universal ethics.” But even so, it seems valuable conceptually to observe the ways in which the particular ethics by which we guide our behavior have a general root in the dynamics of the universe. Particular ethical rules and systems are cultural and psychological manifestations that fill in the blanks left by the powerful but abstract ethics of compassion that is an intrinsic consequence of the basic nature of the universe’s dynamics.