The De-democratization of AI: Deep Learning and the Compute Divide in Artificial Intelligence Research

Citation: Nur Ahmed, Muntasir Wahed The De-democratization of AI: Deep Learning and the Compute Divide in Artificial Intelligence Research.
Internet Archive Scholar (search for fulltext): The De-democratization of AI: Deep Learning and the Compute Divide in Artificial Intelligence Research
Wikidata (metadata): Q101703434
Download: https://arxiv.org/abs/2010.15581
Tagged:

Summary

Authors construct dataset of 171,394 papers from 57 leading computer science academic venues for AI and non-AI, with 2012 ImpageNet contest used as a shock that demonstrated utility of GPUs, and using non-AI venues as a synthetic counterfactual, in order to answer the following research questions.

1. Are we observing an increased concentration of AI research among a few actors since deep learning’s rise? 2. Who are the key contributors to “modern AI” research?

Yes. Increased participation by firms, and by top tier universities.

3. What are the implications for organizations that have been previously active in AI research?

"we still do not have any concrete evidence of how the industry's increased presence will affect the future of AI research"

Paper makes two contributions to the innovation literature:

1. first study that finds evidence that an increased need for specialized equipment can result in “haves and have-nots” in a scientific field 2. contradict trend in innovation research showing decreased corporate research; the opposite is true for AI