New White Paper Addresses “potential harm that AI can do”

By B.N. Frank

Artificial Intelligence (A.I.) technology has many powerful proponents (see 1, 2). Despite growing opposition to its use, A.I. is still being used for facial recognition technology (see 1, 2, 3, 4, 5, 6, 7, 8), gunshot detection technology (see 1, 2), and other applications that sometimes lead to the loss of human jobs. Of course, experts continue to warn against its use because it can be biased, be hacked, make errors (see1, 2, 3, 4, 5, 6). and it’s privacy invasive (see 1, 2, 3, 4, 5). In fact, last year a nonprofit founded by large tech companies launched an A.I. “Hall of Shame”. More recently, a white paper was published about ethics and A.I. use. Of course, this isn’t the first time ethics and A.I. use has been addressed.

From Gov Tech:

White Paper Offers Ethics Advice for Government Use of AI

Titled ‘AI’s Redress Problem,’ the white paper was published by the University of California, Berkeley, and it joins an accelerating cross-sector conversation about the importance of incorporating ethics as AI develops.

Zack Quaintance

A new white paper seeks to help government and other groups build a responsible future for artificial intelligence as the technology continues to evolve, specifically stressing the importance of creating redress mechanisms that can handle flaws as they emerge.

Published by the University of California, Berkeley, the paper is titled AI’s Redress Problem, and it joins an accelerating, cross-sector conversation about how to ensure that ethics and responsibility are part of artificial intelligence’s future. Government is no stranger to this conversation, with New York City, for example, having released a 116-page strategic vision for how to responsibly benefit from AI. This new white paper encourages all stakeholders — government among them — to consider potential harm that AI can do, and to plan for addressing that.

It was authored by Ifejesu Ogunleye, a graduate of the university’s Master of Development Practice program, and Ogunleye conducted this research at the Center for Long-term Cybersecurity’s AI Security Initiative.

In a recent conversation about the white paper with Government Technology, Ogunleye discussed some of her key findings, including the potential for incidental harm, often tied to data sources that have systemic or historical inequity issues.

“By and large, I don’t think you have companies or engineers sitting down and developing things they want to be biased or harmful,” Ogunleye said. “And if you have an AI system that is continuously learning, you haven’t mapped out all the ways it could potentially go wrong, either.”

For these reasons, one of Ogunleye’s key pieces of advice for government as well as private companies — including vendors who sell to government — is the idea of redress mechanisms within AI technologies. Essentially, what that means is that developers include mechanisms in advance that have the ability to stop harmful behaviors that AI might develop. This, Ogunleye notes, is of increasing importance to society writ large as more sectors become more reliant on AI, from health care to government to law enforcement to finance.

In terms of the practical, the paper goes on to cite some government measures that establish redress mechanisms in other technologies, specifically within data protection, those being Europe’s General Data Protection Regulation (GDPR) and the California Privacy Rights Act (CPRA), both of which have been much-lauded by advocates for ethical use of technology.

Higher-level legislation and regulations aside, there are things that lower levels of government can do in this area as well. For all levels of government, Ogunleye advises that it is important for decision-makers to consider members of the community as they make use of AI, and to do so in a meaningful way.

“The community is a very important stakeholder that the industry often hasn’t kept in touch with or engaged with in a meaningful way,” she said. “It’s not just about town hall meetings, it’s about taking in feedback in a meaningful way as you develop these systems.”

And, to be sure, these systems have vast potential for government, with proven capabilities to automate tasks formerly done by humans, improving governmental efficiency and clearing the way for real people to take on higher-level challenges.

It is not, as of yet, a technology that needs to be feared. Provided, Ogunleye said, it continues to be deployed with “the proper safeguards and guardrails.”