🔬 Research Summary by Nur Ahmed, an assistant professor at Walton College, University of Arkansas, and a digital fellow at MIT’s Initiative on the Digital Economy.
[Original paper by Nur Ahmed, Amit Das, Kirsten Martin, Kawshik Banerjee]
Overview: This paper examines the engagement of AI firms in responsible AI research compared to mainstream AI research, revealing a limited engagement in responsible AI, along with a notable gap between the research and its application by industry. Overall, this study underscores the need for greater industry participation in responsible AI to align AI development with societal values and mitigate potential harms.
Introduction
“Ethics is arguably the hottest product in Silicon Valley’s hype cycle today”
– Jacob Metcalf, Emanuel Moss, danah boyd (2019) “Owning Ethics”
As artificial intelligence (AI) continues to mediate various aspects of our lives, from personalized content recommendations to enhanced medical diagnostics, a critical question arises: Are the companies developing these powerful technologies adequately addressing their ethical implications? This study investigates the extent to which the AI industry engages in responsible AI research, assessing the depth and breadth of their involvement.
We analyzed over 6 million peer-reviewed articles and 32 million patent citations from 2010 to 2022. We used multiple machine learning classification methods and validated our classification manually. Our sample includes every firm with at least one AI patent with the United States Patent and Trademark Office.
Our findings reveal a concerning trend: approximately 90% of AI companies (having at least 1 AI patent) show no engagement in responsible AI research. This is in stark contrast to their dominant role in conventional AI research. Furthermore, leading AI firms produce significantly less output in responsible AI research than academic institutions and their mainstream AI efforts.
Patent citation analysis further underscores this disparity, showing that responsible AI research has limited influence on industry’s commercial innovations. This suggests that the rapid commercialization of AI technologies is progressing with insufficient regard for potential ethical and societal consequences despite increasing pressure from regulators, activists, and employees. This study raises crucial questions about the future trajectory of AI development and its societal impact.
Why Should Industry Engage in Responsible AI Research?
Industry plays a pivotal role in the development and deployment of AI technologies due to its access to vast resources and talent. Engaging in responsible AI research allows firms to enhance their “absorptive capacity”– the ability to better integrate ethical considerations into AI development. Public engagement in this field also promotes transparency and accountability, building trust among stakeholders, including customers, regulators, and the general public.
Industry’s Limited Engagement in Responsible AI Research
Our comprehensive analysis shows a significant lack of industry engagement in responsible AI research. Among the 1,771 companies holding AI patents, only 3.3% show meaningful engagement in this area. Similarly, of the 519 companies publishing AI research, only about 11.2% are involved in responsible AI research through peer-reviewed publications. Furthermore, industry presence in responsible AI conferences is also notably low compared to conventional AI conferences, where participation has increased over the years.
Differences in Research Priorities
Employing multiple natural language processing methods, our analysis reveals distinct differences in research priorities between industry and academia. Industry tends to focus on technical advancements and market-driven aspects of AI, while academia emphasizes ethical concerns, societal implications, and human rights issues. This divergence suggests that industry-led AI research may prioritize technical solutions at the expense of broader societal insights, potentially exacerbating the risks associated with AI.
Limited Adoption of Responsible AI Research in Commercialization
Patent citation analysis indicates that responsible AI research has minimal impact on industry’s commercial products. Over 32 million patent citations show that industry patents rarely cite responsible AI research, highlighting a significant gap between responsible AI research and its application in industry. We found that only 88 and 3 industry-authored responsible AI papers were cited in generic and AI patents, respectively. In contrast, generic and AI patents cited over 7,500 and over 1,400 conventional AI papers, respectively, from industry over the same period. This limited integration suggests that AI firms may not be dedicating adequate attention to responsible AI research or lack the capability to incorporate and apply such knowledge effectively.
Between the lines
Our findings suggest that the rapid commercialization of AI is proceeding without sufficient consideration of its potential consequences. Given its leading role and compute advantage in AI development, industry’s limited engagement in responsible AI research is concerning. This limited engagement in research and its adoption in development indicates that the current AI trajectory might not be socially optimal.
The lack of meaningful participation in responsible AI research suggests that firms are largely reactive to AI-induced challenges. Despite mounting pressure to address these challenges and having substantial resources, industry participation remains limited. This indicates that existing incentives might not be sufficient.
Our results highlight the urgent need for industry to prioritize and deepen its involvement in responsible AI research, ensuring that AI development aligns with societal values and mitigates potential harms. Failure to do so may undermine public trust, hinder the realization of AI’s benefits, and exacerbate the risks posed by this powerful technology.