Until
next week,Moral Meme Machines
Dateline: September 14, 1997
Note: For this and future articles, I shall dispense with ellipses in quotes. In omitting elements of a quote without flagging the omission, there is greater danger of misrepresenting an author’s intended meaning, but I’ll be careful . . . . [<- omission: "Yeah, right." :)] It makes for much smoother reading, this way, plus it’s one way new memes are created!]
HERE you are, devoting several hours to reading my book. Shouldn't we both be out raising money for Oxfam? Every day, while trying desperately to mind our own business, we hear a thousand cries for help, complete with volumes of information on how we might oblige. How on Earth could anyone prioritize that cacophony? Yet we do get there from here. Few of us are paralyzed by such indecision. By and large, we must solve this decision problem by permitting an entirely `indefensible' set of defaults to shield our attention from all but our current projects.
—Dan Dennett
Dennett is a philosopher of AI. Morality (or ethics; they are the same in essence) is one of philosophy’s enduring core topics, and Dennett writes about it in Darwin’s Dangerous Idea. Andrew Leonard’s splendid new book BOTS: The Origin of New Species provides some telling examples showing just why it would be wise to make morality a core topic of AI, too. This article draws primarily on these two works. The seeming emphasis on bots is just a reflection of the many quotes from Leonard’s book, so please take references to bots as applying to all software and hardware implementations of AI—that is, to Machina sapiens.
Three issues of morality should concern us in any discussion of intelligent machines:
This article examines the first of these questions. Future articles will examine the other two.
How does morality arise?
Philosophers have been all over the map on this question, but there is agreement at least among Jean-Jacques Rousseau, John Locke, and (in modern times) John Rawls (Theory of Justice). They all agree, says Dennett, "in seeing morality to be, in one way or another, an emergent [my emphasis] product of a major innovation in perspective that has been achieved by just one species, Homo sapiens, taking advantage of its unique extra medium of information transfer, language."
Morality has to do with questions of Good and Evil, Right and Wrong, Virtue and Vice. Modern evolutionists have a concept of "goodness" with a small g. An outgrowth of Darwinian theory called adaptationism holds essentially that a species will adapt in an optimal (best, "most good") way to its environment. On the face of it, adaptationism and goodness-with-a-small-g would seem to have nothing to do with morality; they are merely an evolutionary survival mechanism.
But so is morality itself; so is Goodness-with-a-capital-G.
Morality causes intelligent organisms to react against certain types of maladaptive meme (the gene of culture) which, while possibly offering short-term benefits for an individual or a sub-group within the species, are in the long term bad for the survival of the species. Those who react most strongly are generally known as saints and martyrs. Those who react in favor of, or knowingly create, such events and circumstances for selfish ends are generally known as assholes, if you’ll pardon my French.
But what causes morality itself? Dennett dismisses as nonsense the belief of entomologist E.O. Wilson and philosopher Michael Ruse that "Morality, or more strictly our belief in morality, is merely an adaptation put in place to further our reproductive ends," though that sounds about right to me, if you take "reproductive ends" as meaning species survival.
We transcend this, says Dennett. "Persons, according to the meme model, are larger, higher entities, and the policies they come to adopt, as a result of interactions between their meme-infested brains, are not at all bound to answer to the interests of their genes alone—or their memes alone." However, "A meme or complex of memes can redirect our underlying genetic proclivities." I think a good recent example of this can be seen in a report issued in September 1997 by a committee investigating sexual harassment and gender relations in the U.S. Army. The report recommended that a week of training in ethics and moral values be added to recruit training. This is a sign of progress—an example of memetic evolution in action; and it’s also a reversion to a bygone era when ethics were a staple ingredient of education.
So memetic evolution gives rise to morality, but how? Dennett says that "We can already be virtually certain that mutual recognition and the capacity to communicate a promise—stressed by both Hobbes and Nietzsche—are necessary conditions for the evolution of morality" [my emphasis]. If I understand this correctly, it means we have to know that others exist, and we have to be able to make contracts with others. I have a problem with the second condition: to me, it would be unethical not to act ethically to others even if we were totally unable to communicate with them. To our credit, we’re getting better at this (Save the Whales springs to mind), though some eastern religions and philosophies have practiced it for millennia.
So we’re still not close to the "how" of its emergence. What about the "how" of its operation? How do we deploy it? Apparently, it’s not programmable, so that’s out: "No remotely compelling system of ethics has ever been made computationally tractable, even indirectly, for real-world moral problems," says Dennett. Readers of my previous article on heuristics may not be surprised to learn that satisficing (a heuristic process of settling for the "good enough" rather than insisting on the best) "is the basic structure of all real decision-making, moral, prudential, economic, or even evolutionary."
We satisfice because ethical decisions often need to be made in a hurry and there is simply no time to sift, algorithmically, through the hundreds, thousands, or potentially billions of ramifications and alternatives actually open to us. A heuristic process quickly discards the majority of those ramifications and alternatives (at the risk of throwing out the baby with the bath water) and focuses deliberately, myopically, on what look to us like a few good bets. "Finding a moral algorithm," says Dennett, "is forlorn." It has to be heuristic.
In sum, Dennett seems to hold the view that morality/ethics is an emergent heuristic in organisms that have language. That means Homo sapiens. Since, as my faithful readers will know, I hold that Machina sapiens will be an organism operating on memetically-honed heuristics and possessing a language richer, faster, and more flexible than ours, then it is inevitable that it will be a moral machine.
Further evidence suggestive of (im)morality in machines is to be found in bots—software robots (see related article describing bots).
"Bots are just beginning to crawl out of the primordial digital ooze, and all the kinks haven’t been worked out of their genetic code. In the real world, they are imperfect beings," writes Andrew Leonard. "If bad bots run amok, good bots will appear [my emphasis; it suggests emergence] to counteract them, or the system as a whole will be redesigned to quash their delinquent outbursts." In other words, morality is an automatic species survival mechanism, as we said earlier. It is innate; it does not make a show of itself unless and until it is triggered by instances of immorality. It is a balancing mechanism.
But "Lost in all the hubbub over the question of whether we can achieve artificial intelligence is the more philosophical question of whether we should even strive to do so." I disagree. Only the first question is amenable to philosophy. The second is an empirical, historical, evolutionary fact: whatever we can do, we will do, if only conceptually—in our minds. When new memes come into existence, formed from bits of other memes, they will be expressed in some form or other, conceptual or physical. The expression may not last if it confers no survival benefit on the organism, if it does not contribute to an Evolutionary Stable Strategy (see previous article).
Instead of asking whether we should create Machina sapiens, knowing that we will, a far more interesting and important philosophical question is: What impact might Machina sapiens have on human morality? And that will be the topic of the article for September 28, after we take a brief detour next week to bone up on Bots.
Until
next week,
NEXT WEEK: Bots: What they are, what they do, where they've been, where they are going.