Researchers propose alternative AI agenda focused on human augmentation over automation

Arvind Narayanan and Gabriel Unger argue that the dominant AI agenda—centered on automating human labor—is overly narrow and risks overlooking technical and economic realities. They contend that AI can and should be developed to augment human capabilities, fostering collaboration between humans and machines rather than replacing workers entirely.

The authors trace the intellectual roots of this debate to human-computer interaction and note that a minority of businesses are beginning to embrace collaborative AI after economic disappointments from pure automation efforts. However, they argue that computer science lacks a clear, systematic research program to advance this alternative agenda.

The essay proposes four illustrative research directions to steer AI toward collaboration. The first is the development of 'white-box agents'—systems designed to reveal their intermediate processes, assumptions, and uncertainties to users. This approach contrasts with the dominant 'black-box' paradigm, which prioritizes final outputs without transparency into how those outputs are generated.

To build white-box agents, the authors highlight the need for process trace data across professions such as law, medicine, and the sciences—domains where such data is currently scarce. They argue that the availability of process traces, akin to open-source software repositories for coding, could enable AI systems that amplify human expertise rather than replace it.

The second direction focuses on developing agents capable of real-time interaction, with low latency that mirrors human conversation. The authors suggest that today’s AI systems often fail to meet this standard, which is critical for meaningful collaboration.

The essay also situates this agenda within broader debates in economics and policy, arguing that a market-driven, bottom-up approach could help steer AI development toward collaborative outcomes. It calls for a collective, experimental research program to systematically explore these directions.