A new report by researchers from Stanford University has shed light on the challenges tech companies face in upholding their commitments to the ethical development of artificial intelligence (AI). Despite making promises to prioritize AI ethics, the reality within these organizations often falls short of these lofty goals.
The Gap Between Promise and Practice
The report, titled "Walking the Walk of AI Ethics in Technology Companies," authored by Sanna J Ali, Angele Christin, Andrew Smart, and Riitta Katila, highlights a significant discrepancy between companies' public commitments to AI ethics and the actual implementation of these principles.
Many companies have embraced the idea of AI ethics, publishing principles and hiring social scientists and engineers to research and develop ethical AI solutions. However, the Stanford Institute for Human-Centered Artificial Intelligence notes that the adoption of ethical safeguards is not a priority for most of these companies.
Based on interviews with 25 AI ethics practitioners, the report paints a picture of a workplace environment where individuals tasked with promoting AI ethics face institutional resistance. They often find themselves isolated within their organizations, despite promises of integration and support.
Cultural Indifference and Hostility Towards AI Ethics
One of the most striking findings is the reported culture of indifference, and at times hostility, towards AI ethics within these companies. Product managers are often seen as perceiving ethics as a hindrance to productivity, revenue generation, or product launch timelines.
An individual surveyed for the report mentioned, “Being very loud about putting more brakes on [AI development] was a risky thing to do. It was not built into the process”.The report further reveals that ethical considerations are frequently an afterthought, brought into the picture late in the development process, making it challenging to integrate changes into new apps or software.
Moreover, the focus on engagement metrics and AI model performance often overshadows ethics-related recommendations, especially when these recommendations might negatively impact those metrics. The difficulty in quantifying ethics or fairness, compounded by companies' existing data infrastructures not being tailored to such metrics, adds another layer of complexity to the issue.