While many proponents like to see the field of Machine Learning as value-neutral and/or universally beneficial, work in this field tends to be animated by a narrow set of motivations and concerns that shapes what projects are chosen, what problems are addressed and how outcome are conceived.

Abeba Birhane and colleagues show that ML projects tend to be driven primarily by a narrow set of values and concerns over improving technical performance, efficiency and/or generalisability of ML systems. AI researchers and developers are often motivated either to build on their previous work and understanding, *or* the perceived novelty of the application.

In contrast, societal implications and possible negative consequences are loosely conceived and considered (if at all).

All told, ML projects are driven by values that support the centralization of power, commercial agendas of big tech actors, and the framing of ML as a technical context-free computational challenge.


Birhane, A., Kalluri, P, Card, P., Agnew, W., Dotan, R. and Bao, M.  (2021)  The values encoded in machine learning research.  https://arxiv.org/abs/2106.15590