With Senate Majority Leader Chuck Schumer releasing a sweeping “roadmap” for AI legislation today and major product announcements from OpenAI and Google, it’s been a big week for AI… and it’s only Wednesday. But amid the ever-quickening pace of action, some observers wonder if government is looking at the tech industry with the right perspective. A report shared first with DFD from the nonprofit Data & Society argues that in order for powerful AI to integrate successfully with humanity, it must actually feature… the humanities. Data & Society’s Serena Oduro and Tamara Kneese write that social scientists and other researchers should be directly involved in federally funded efforts to regulate and analyze AI. They say that given the unpredictable impact it might have on how people live, work and interact with institutions, AI development should involve non-STEM experts at every step. “Especially with a general purpose technology, it is very hard to anticipate what exactly this technology will be used for,” said Kneese, a Data & Society senior researcher. Data & Society has some sway in high-level Washington conversations about AI, as its executive director, Janet Haven, serves on the White House’s National AI Advisory Committee (and participated in one of Schumer's Insight Forums in December 2023, on "Safeguarding Against AI Risks"). Still, when it comes to this push to integrate the humanities and AI, the two researchers believe today's "roadmap," developed separately from the White House, is a missed opportunity. “The roadmap reduces AI research to technical assessments and safety practices," they wrote to DFD. "It guides us toward a fragmented approach driven by national security fears, rather than focusing on protecting Americans from the technology’s harms and ensuring evidence-based AI development.” (Asked for comment a representative of Sen. Schumer pointed to today’s press conference announcing the roadmap, where he said “We know there are harms that can come from AI, whether it be, hurting labor in the workforce, bias that’s built into the systems... dealing with intellectual property," and called for "innovation to come up with ways to minimize those liabilities.") Oduro and Kneese see an opportunity to accomplish their goal of including “sociotechnical” analysis in AI development via Biden’s AI executive order. They write that federal hiring programs on AI should post jobs for humanities and social science experts; agencies adopting AI should require such research as part of their procurement process; and people writing federal guidelines on tech should incorporate humanities and social science methods. There are longstanding, important arguments that technologists should place the human experience first as they design and roll out their products. The report points to the pioneering work done by ethnographers at Xerox’s PARC lab in the 1970s and 1980s; the Canadian physicist and public intellectual Ursula Franklin warned in 1989 as part of her “Real World of Technology” lecture series that “patterns laid down in the practice of technology become part of a society’s life.” Oduro and Kneese say computer technology, and now increasingly AI, have fulfilled that prophecy. They believe Silicon Valley has become such a financial titan that companies may not be able to step back and examine that relationship between their tech and society — that is, unless policy forces it. “There are still pockets of people who are thinking in the longer term and doing ethnographic fieldwork even within large tech companies,” Kneese said. “But by and large, the financial pressure of having to produce quarterly results … makes it much harder to justify doing that kind of work in the budget.” Another group pondering how to bake humanities into AI is the AI Governance Lab at the nonprofit Center for Democracy & Technology. In a report published today in parallel with Oduro and Kneese's, they make a sweeping list of recommendations for implementing social science expertise that includes applying ethnographic research to algorithm development and not just user experience, integrating humanities experts in AI safety teams and having those experts monitor AI systems after their deployment. The CDT authors write that while experts in sociology, anthropology, political science, law, economicse and psychology “already exist in a wide range of technical and non-technical roles in AI companies,” they “are often relegated to siloed roles in AI ethics or governance, compliance, or pre-deployment user interface testing.” Amid the AI gold rush to get products like OpenAI’s powerful, free-to-use new GPT-4o voice assistant into consumers’ hands, civil society critics say relegating humanities experts to the sidelines risks repeating the worst mistakes of the most recent digital era. “The general success of AI development and innovation hinges on seeing it as a socio-technical system and having that expertise integrated throughout,” Oduro said, adding that “hopefully [the report will] have an influence on industry, that companies can't have contract work with a government or be seen as ethical without those people involved.”
| 
Comments
Post a Comment