The hardest critique of techno-humanism is that the compass is built by the same technology it is supposed to orient. Models trained on human data shape human behavior. The data feeds back into the training. The criteria by which we judge whether AI is expanding or contracting agency are themselves communicated through AI-mediated channels. The compass, in this reading, is a co-product of the territory, and the navigator is being navigated.
Hoffman is not unaware of this. His writing addresses the recursive problem repeatedly, usually under the heading of what he calls the alignment of the alignment — the question of who gets to decide what counts as good alignment, and how that decision is itself influenced by the systems being aligned. His answer is partial and pragmatic. He argues that democratic institutions, however imperfect, are still the best available mechanism for adjudicating contested values, and that AI development should be conducted in ways that strengthen those institutions rather than bypass them. The compass is built by the technology, yes, but the technology is built by humans embedded in political and ethical communities, and those communities have to remain the source of authority.
The pragmatism is honest. It is also vulnerable. The political and ethical communities Hoffman invokes are themselves under stress from technologies in which he has invested heavily. Social media has degraded the epistemic basis on which democratic deliberation depends. Recommendation systems have segmented populations into incompatible information environments. Generative AI threatens to multiply both effects at a scale that makes the early social media problems look modest. Asking those battered institutions to be the source of authority for steering AI is asking a lot.
Hoffman would say that it is also the only available answer. The alternatives — appointing technical experts to steer the technology unilaterally, or letting market forces decide everything, or freezing development until political institutions are ready — all fail their own tests. Technical experts have no special legitimacy on questions of value. Markets do not internalize externalities at the speed required. Freezing development cedes the field. The compass remains the imperfect tool we have, and the work is to make it as accurate as possible while accepting that the maker and the user are partly the same.
There is a deeper philosophical question hiding in this. The classical liberal tradition that Hoffman draws on assumed a relatively stable substrate of human nature against which technologies operated. Mill could write about liberty because he could assume a stable category of human person whose liberty was at stake. AI complicates that substrate. If cognitive labor is increasingly performed by machines, if memory is increasingly externalized, if even introspection is mediated by tools that suggest what to think — what does liberty mean? Hoffman's answer is that it means having the practical ability to choose differently than the suggestion. The compass, in this rendering, is the capacity for refusal as much as the capacity for direction.
The chapter that comes next moves from the abstract structure of the compass to its application in domains where Hoffman has placed concrete bets — biology, medicine, and the broader scientific enterprise. The compass is most testable where the territory is most measurable, and Hoffman has been increasingly active in fields where outcomes can be quantified and timelines can be checked. The techno-humanist wager is not a theory you can prove. It is a series of empirical bets whose verdicts will accumulate. Hoffman is placing those bets in public, and the next part of this book follows them.