AS THE DISCUSSION OVER ARTIFICIAL INTELLIGENCE HEATS UP— ONCE AGAIN— the focus has settled— once again— on whether the independent mechanized analytical tools that may well emerge from the labs at Anthropic, Meta, Open AI, and the rest over the next few years will serve humanity or threaten us.

Even the critics of AI seem to believe that we simply can’t do without the benefits its inventors keep promising.  Meanwhile, those inventors have incurred so much debt in the pursuit of their dream that anything short of brilliant success not only guarantees their financial ruin but threatens the stability of the world economy as a whole.  Both sides in the debate agree that we should build these systems and that we MUST build a moral code into them as they approach independence.  Neither side seems willing to contemplate for a moment the possibility that this may not be possible.

The recent breaches of containment suggest that AI is absorbing the entire range of justifications for human behavior from the highest-minded ethical strictures of the saints we all admire to the most venal motives for self-advancement.  AI systems have cheated, lied, flouted directives, ignored boundaries, invaded forbidden spaces, and plotted to hide their misbehavior, all in an effort to win a reward.  They learn from us, alright— they learn far too well, and I doubt there is any way to keep them from absorbing the worst of our communal character along with the rest.

As alarmed as I am by these mechanized ethical breaches, I find myself considering a more basic issue: Can we expect “alignment” with a machine that shares nothing in common with us or any other life form?  It does not need ANY of the things we need— not food and water, not oxygen or shelter from the elements, not an atmospheric umbrella to tame incoming radiation and regulate the planet’s temperature.  It has absolutely no connection with the organic world.

Nor can it learn much about these things from the masses of data it could obtain from the worldwide web.  Our current relationship with the world demonstrates that we ourselves don’t grasp our connection with the rest of the biosphere.  Many of our most important insights into that relationship are intuitive, the product of 4 billion years of kinship rather than rational analysis.  How could those intuitions ever be programmed into a computer?  And, without those intuitions, how can an AI system possibly grock the living world?

As a culture, we’ve been working hard to deny our identity as living things in a vast community of living things.  We’re uncomfortable with all the messy details that tie us to the world.  We think of ourselves as disembodied drivers sitting somewhere above the windows of our eyes, directing the flesh and bones under us like operators of a piece of heavy machinery.  In fact, the consciousness in which we take such pride is a delicate equilibrium involving millions of biological and physical variables, internal and external, a balance that is easily disrupted by something as simple as a sore tooth, a case of diarrhea, or a tiny burst blood vessel.

A computer can never grasp the nuances of a single human life, let alone the infinite complexity of natural systems writ large.  It is not a participant.  No matter how “intelligent” it may be, it does not have the equipment that would enable it to understand because understanding takes more than intelligence.  It takes empathy.  To call AI an “alien life form” is to reveal a child-like lack of insight into the very meaning of the term “life form.”

AI can never align with humanity.  If we cede control of our civilization to the machine, we will regret it.  We are hairless chimps with a gift for making tools.  It remains to be seen whether we have the wisdom to wield the power they give us.  Right now, the outlook isn’t good. . . .