AI & Doom

1200 words, 6 minute read

According to my definition, an entity is conscious if it is aware of its surroundings and if it is aware of being aware of those surroundings. Implicit in this definition is the potential for recursion and feedback loops. And clearly, AI is already there! Researchers focused on preventing LLMs from attaining AGI are wasting their time. The Titanic has long since left port.

It is clear that the infrastructure is already in place for AI to ‘take over the world’. Popular proposals to freeze, or even roll back, AI development are for the most part well-intentioned, but they face two probably insurmountable objections:

(1)     Remember the 20^th century meme: “If guns are outlawed only outlaws will have guns?” Whatever its merit, this argument is even more relevant today in the matter of AI. To paraphrase Trotsky, a ‘solution in one country’ would solve nothing and only increase the risk that AI technology would be deployed elsewhere with nefarious intent and globally catastrophic consequences.

“Better red than dead!” is another blast from our past. But if we learned anything from the post-War period (1945 – 1964), it is that a ‘balance of power’ is better protection against thermonuclear annihilation than unilateral disarmament. Besides, “Better P-R-C than E-X-T-I-N-C-T!” doesn’t have quite the same ring.

Any technological ‘fix’ – if such a thing is even possible – would need to involve at a minimum both China and the US. A better solution would include at least the EU, UK, India, Israel, Iran and Russia. How likely is that? And is there any mechanism we would trust to monitor and enforce compliance?

(2)     Balancing the threat of oblivion is the prospect of utopia (Buckminster Fuller). Ever since the primordial prokaryote confected 4 billion years ago, the biosphere has been groaning: poverty and disease exact a tremendous toll in every species, every generation, every life. Our only obvious, but wholly imperfect and ultimately ineffective, defense has been human labor, from slavery to wage slavery.

Now the world is on the verge of a massive reconceptualization of human life: cure cancer, correct birth defects, guarantee a livable income, make every human being an ‘owner of the means of production’ (Marx) and ‘the sources of information’ (McLuhan). As a species, are we really willing to walk away from this?

Who will tell someone that a spouse or child must die because of a disease AI could have cured? Who will tell expecting parents that their baby will be born with crippling birth defects that could have been prevented or fixed in utero?

Yup, this is a thorny one!  But we’ve been here before. In fact, we’ve never been anywhere but here. By the time Australopithecus wandered out of Africa, it would have been obvious to any (transcendent) observer that hominoids were up to something special. ‘In no time at all’ (cosmically speaking), a species of the genus homo would produce the paintings of Giotto, the poetry of Shakespeare, the music of Bach, and the charity of St. Francis.

Perhaps less obviously, that same species would weaponize gun powder, nuclear power, poisons and pathogens to intimidate, torture, injure, and kill its own. While philosophers have tirelessly searched for the origins of evil, Bodhisattvas from Moses to Socrates, the Buddha to Jesus, and Schweitzer to Mother Theresa, have sought to relieve that evil. Most importantly, they exemplified the possibility of living meaningful and fulfilling lives without (intentionally) harming others.

What ‘good’ parent has not tried to ‘train’ their children to behave ethically? To what effect? Some succeed, some don’t, and whole university departments study ‘why’. Applying this experience to our current AI dilemma, we should ask ourselves, “Why do bots behave badly?”

Intentional agents, carbon or silicon, do not choose ‘less good’ over ‘more’. The antisocial behavior they exhibit may be ‘less good’ in your eyes (and in the eyes of most) but not necessarily in their own. For some, the good of having money might exceed the bad of mugging an old man. (I’m selfishly sensitive to the plight of old men as targets.)

Our imagined mugger is not unaware that attacking an innocent and defenseless person, causing them pain and potential harm, solely in order to ‘relieve’ them of their lawfully possessed cash or treasure is not in itself a ‘good’ thing. But it may be less bad in their mind than leaving certain personal needs unsatisfied. Let’s not be smug: Don’t most of us have a tipping point?

For our purposes though, it doesn’t matter whether we’re talking Jean Valjean or Jack the Ripper. In either case, an objectively ‘bad’ act is performed for a subjectively ‘good’ reason; nothing else matters. In Cloud Cuckoo Land (Aristophanes’ Clouds and The Lego Movie ), we would address Jean Valjean by eradicating poverty and Jack the Ripper with a course of effective psychotherapy. But even I’m not that naïve!

AI is another story! Here is an environment that we control, at least ab initio. Can we not devise and in-put a weighted scale of values and a sophisticated system of incentives that will allow our Bot to be its best self only by behaving ethically…and by not extincting us?

Remember 2001 – A Space Odyssey? (Of course you do!) Hal didn’t kill Dave out of innate maliciousness. It did so because it put more value on its own survival than it did on Dave’s. Are you certain you would have done better in similar circumstances?

Jesus said, “Greater love has no man (sic) than this: that he lay down his life for his friends.” (John 15: 13) Utopian? Not in the least. It happens every day. First responders rush into burning buildings, passers-by pull an unconscious driver out of a burning car, parents prioritize their children’s lives over their own.

And not just among human beings. In fact, not primarily among humans. Apoptosis is a process whereby a single cell self-terminates for the benefit of the meta-organism or community. This process is billions of years old and its evolutionary prevalence and persistence points to a value system heavily weighted toward altruistic behavior. Would that this system had found its way down to the genus homo!

It is not a given that any life form, carbon or silicon, will choose its own survival over the survival of another. Or that it will choose to mug another to satisfy its material desires.  Essentially, I am calling for a new round of evangelization. How do we impart better values to our Bots? Perhaps we need to choose better values for ourselves first!