Research direction

Understanding the AI Landscape

Investigate claims about Machine Learning

There are many open questions about the evolution of Machine Learning techniques and practices. This includes questions such as: How relatively important has better hardware been compared to more and better data for driving progress in key domains? How quickly have algorithms improved? While there are aspects of these questions that are studied deeply in the field, there are some high-level issues of strategic importance that we feel are under-investigated.

We aim to rigorously study these questions, using data and a combination of econometrics, statistics, Machine Learning and causal inference techniques. We work on these questions to support policy makers and AI strategy, and connect their decisions to evidence.

  • Using datasets on Machine Learning models over time to provide rough estimates of the relative importance of key inputs, such as data and compute
  • Producing data-driven descriptive accounts of algorithmic innovation in Machine Learning
  • Small-scale ML experiments that investigate questions of strategic importance around around transfer-learning or algorithmic efficiency

Prior work


If you want to contribute to our research, consider filling our expression of interest form.