Transcend

Creating policies to shape the future of AI requires clarifying what humans value most.

AI is ubiquitous in much of everyday life. It’s involved in everything from the facial recognition software you use to unlock your phone to the songs your music streaming app plays next. As AI technologies evolve, many are justifiably concerned about the profound effects these emergent technologies could have on important facets of life, such as human agency and planetary health. Humanity must establish clear priorities about how to direct and govern AI’s use and development by reflecting on two questions: What does humanity value most, and thus need to protect?; and what should people value, despite the fact that they appear to overlook it?

Clarifying human values and uniquely human capacities is of vital importance when training large language models (LLMs): People must find some degree of alignment in terms of what constitutes a “good pattern” and is, therefore, worth reproducing at scale. For example, if you were training an LLM to speak, you’d want to identify the patterns of “good” human speech. Unlike...