Open Tech Institute

For over a year, @OTI has pushed @FTC for stronger vertical merger guidelines. Now, progress: @FTC voted to rescind its 2020 guidelines, which @joshuastager says "glossed over critical issues, ignored key harms in digital markets & denied public participation."

Audience Q: Does growing use of #AI in govt/biz seriously threaten our democracy? @ChristineCustis: We don't want to forfeit the benefits of breakthroughs just because they could be used for harm. We’ve just got to be thinking thru risks, mitigation, and all scenarios. #AIRisk

.@nyulaw's Catherine Sharkey is most drawn to requiring algorithmic impact assessments & how that fits into an administrative law mechanism. @ChristineCustis: Impact assessments are super important; thinking about #AIRisk, being bold in the way we distinguish issues & risks.

.@SarkBites asks: is there one method of mitigating #AIRisk that is most important? @spandi_s: No! There are benefits and disadvantages to each approach, and they can supplement one another.

.@ChristineCustis: "Risk" is not necessarily the way we would categorize #AI itself. I’d say it’s the way we should categorize our visibility into it. At some vantage point, it’s all high risk. We just need more visibility and transparency. #AIRisk

.@nyulaw's Catherine M. Sharkey: I worry about these lines in the sand... there’s a human in the loop, but the AI actually makes a determination. We should define these concepts along a continuum so we don’t create “safe harbors” that let certain tools evade scrutiny. #AIRisk

.@ChristineCustis: There are so many different ways to document an ML system… the way you choose is not necessarily as important as your consistency. "Every way is not right for every organization." #AIRisk

.@SarkBites asks abt proposed gov't requirements for #AI audits. @spandi_s: Good to see policymakers thinking thru these issues, but many of these proposals "lack teeth." Legislation also needs to address the broader ecosystem that allows high-risk AI to be so harmful. #AIRisk

.@ChristineCustis says documentation is important throughout the machine learning lifecycle: who did what, and what the outcomes and impacts were for the system, the users, and non-users. If we don't get it right, we can go back and figure out the source of the harm. #AIRisk

.@spandi_s: Audits, impact assessments & more can shed light on opaque high-risk #AI by evaluating variables we're concerned about, like privacy, bias, fairness, & human rights, & they can examine harms of a system & help an entity create a plan to mitigate these issues. #AIRisk

We have to be careful thinking about off-the-shelf, private sector tools being imported in terms of governmental use, says @nyulaw's Catherine M. Sharkey. “We need to get the lawyers, policists and technologists in the room together while developing these tools.” #AIRisk