🚨I am excited to share my first-ever peer-reviewed paper, which ended up at @ScienceMagazine! 🚨. In this paper, we systematically quantify the growing influence of industry in AI Research. 📜 A thread 1/n

What an excellent paper! Congratulations @yian_yin, @dashunwang & team.

Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.

RT @miaomiaozhang20: @AOM_TIM @AijaLeiponen @klakhani @sstern_mit @lynnwu02 Please also consider joining @KevinJBoudreau @H_DigInnovation @….

Please join us at #AOM2024 . This is our 2nd year on "Science of Science". We are taking it to the next level.

Our recent preprint highlights that AI firms might underinvest in responsible AI development without a significant external push. There is no regulation in AI and the industry seems to be moving ahead without seriously considering the downsides of it. #ResponsibleAI #AI.

This work is jointly done with an amazing team. @SapientDas, @KMac_Martin,@Kawshik2024 . Please reach out for feedback and other follow-up work! .

A more recent example is from OpenAI and its safety team. It seems that OpenAI is choosing speed over safety. Just like the vast majority of AI firms. ⏭️Overall, our results showcase that the commercialization of AI is happening with limited consideration of its consequences

What are the consequences of this limited engagement of industry in safe and responsible AI development? . @GaryMarcus & others have highlighted such issues for a long time. In particular, firms do not pay much attention to the negative externalities.

It's important to note that additional analyses (and ongoing work) suggest a participation varies among firms. We are studying the causes and consequences of this heterogeneity. But overall, industry presence is rather limited in this space despite mounting pressure.

Finally, we examine to what extent industry utilizes responsible AI research in its commercialization (in patents). Our large-scale patent citation analyses suggest that the vast majority of AI firms rarely cite responsible AI even tho they build on conventional AI research

Next, we document that the industry's research portfolio is narrower than academia's, with limited focus on critical issues such as environmental costs 🌲, human rights 🔎, and moral considerations ⚖️

These results are consistent with the fact that industry is prioritizing commercialization over safety and caution

Then, we show the % of industry papers in conventional AI & responsible AI conferences over time (Fig B). Then we show a distinct keywords-based sample (Fig. C). In both cases, industry is not improving its engagement in responsible AI research

After that, we focus on leading conventional and responsible AI conferences. We document a strong disparity in industry engagement in 10 leading conventional AI conferences & 3 leading responsible AI conferences

Next ⏭️, we compare the leading 100 AI firms with the leading 100 universities. This graph showcases a stark difference in their participation in conventional AI research compared to their engagement in responsible AI research . 2nd graph considers the quality of the research.

1⃣, we show that among all AI firms (firms with 1 or more AI patents), nearly 90% do not have any single responsible AI paper. 2⃣, we show that even within AI research firms (firms with 1 or more AI papers), nearly 65% have no engagement in responsible AI research

To classify responsible AI papers, we use *both* machine learning 🤖 (SciBERT-based model) and keywords-based approach 🧍‍♂️. We also manually validate our method and have used multiple methods to produce consistent results. (Please see the paper for more details).

We used 5 different comprehensive datasets that go beyond top-tier outlets. This data includes more than 6 million peer-reviewed papers (2010-22) and 32 million USPTO patent citations (1947-2022)

In this study, we ask: .1⃣To what extent does industry engage in responsible AI research?.2⃣How do the priorities of responsible AI research differ between industry and academia? .3⃣To what extent does industry integrate responsible AI research into its commercial inventions?.

However, evidence on the industry's engagement in responsible AI research is conflicting. Some argue there is "ethics washing," meaning all talk and no real action. Others assert that significant, genuine activities are taking place.

Other scholars & researchers have called for more engagement in this space too (eg, @privitera_) . Additionally, regulators including the White House have urged more investment in R&D in this area so that we can maximize the benefits and minimize the negatives from AI