The problem AI poses is political through and through. While today we often characterize debates around the ratification of the Constitution as centering on questions of power, actually the Founders were much more concerned about the value of responsibility.

More than a generation ago, Steve Jobs enthused that the computer represented “a bicycle for the mind,” boosting our efficiency and our mental muscles at the same time. And so it may have been in the early days. Before long, however, the advent of high-speed internet so accelerated access to information that our devices increasingly became automobiles for the mind, reducing us to a state of mental obesity that soon demanded a “mobility scooter for the mind,” which has now arrived in the form of generative AI. Such devices do not merely compensate for our collapse of personal competence, but condemn us to a state of perpetual dependence upon them.

Educators are thus lamenting the accelerating collapse of literacy, while employers sound the alarm about new hires unable to function without ChatGPT; and even Anthropic has revealed that when their engineers use Claude to code, they become less skilled for future work. And yet the impending crisis of human competence continues to be overshadowed in the national conversation by more sensationalist hopes and fears around AI.

When I recently had the pleasure of moderating a conversation on this theme with leading AI policy minds from the frontier labs and DC thinktanks, few seemed to quite share my concern. Haven’t we all already outsourced our navigational competence, one asked, to GPS? Many of the participants seemed less concerned with human competencies than human incompetencies: things our species is so poor at doing that we had better hand them off to the machines before we make more of a mess of things. Driving, for instance: Human driving incompetence accounts for over one million deaths per year worldwide, so why not save millions of lives by giving the wheel to self-driving cars?

The optimism about AI automation goes well beyond cars, however. Many techno-optimists have suggested handing over large swaths of imperfect human decision-making to the new “machines of loving grace,” which promise justice, fairness, and safety. Judicial sentencing, for instance, can be notoriously inconsistent and subject to human bias; why shouldn’t we want such matters of life and death to be handled by an all-knowing algorithm? I’m not so sure, and not just because I’ve seen Minority Report.

While the techno-optimists are liable to see society’s hesitancy to embrace the Waymo as an irrational phobia, perhaps this hang-up points to a deeper level of rationality: one that values accountability over infallibility, responsibility over efficiency. Self-driving car accidents may be rare, but human beings care less about “never getting it wrong” than knowing whom to blame if it does. If a drunk driver kills an innocent bystander, a great hole is torn in the fabric of the moral universe; and yet, by naming the offender, bringing him to justice, and holding him accountable for his crime, we are, in some measure, able to patch it, restoring some sense of sanity and order to this vale of tears. If justice miscarries, we have mechanisms of appeal, or in the last analysis, a sense of comfort in knowing that the corrupt judge will someday face a higher judge. But what if both driver and judge are faceless, bodiless algorithms residing in an impenetrable black box? Accidents will still happen (even if more rarely), and judicial decisions will still be hotly disputed (even if decided on by impeccable logic). The difference will be that we will no longer know whom to blame, whom to hold accountable, how to patch the hole that has opened in our moral universe.

Read More