Topics: artificial intelligence

Essay by Arthur Holland Michel: “Many assumptions about artificial intelligence (AI) have become entrenched despite the lack of evidence to support them. Basing policies on these assumptions is likely to increase the risk of negative impacts for certain demographic groups. These dominant assumptions include claims that AI is ‘intelligent’ and ‘ethical’, that more data means better AI, and that AI development is a ‘race’.

The risks of this approach to AI policymaking are often ignored, while the potential positive impacts of AI tend to be overblown. By illustrating how a more evidence-based, inclusive discourse can improve policy outcomes, this paper makes the case for recalibrating the conversation around AI policymaking…(More)”

Essay by  L. M. Sacasas: “To take a different class of example, we might think of the preoccupation with technological fixes to what may turn out to be irreducibly social and political problems. In a prescient essay from 2020 about the pandemic response, the science writer Ed Yong observed that “instead of solving social problems, the U.S. uses techno-fixes to bypass them, plastering the wounds instead of removing the source of injury—and that’s if people even accept the solution on offer.” There’s no need for good judgment, responsible governance, self-sacrifice or mutual care if there’s an easy technological fix to ostensibly solve the problem. No need, in other words, to be good, so long as the right technological solution can be found.

Likewise, there’s no shortage of examples involving algorithmic tools intended to outsource human judgment. Consider the case of NarxCare, a predictive program developed by Appriss Health, as reported in Wired in 2021. NarxCare is “an ‘analytics tool and care management platform’ that purports to instantly and automatically identify a patient’s risk of misusing opioids.” The article details the case of a 32-year-old woman suffering from endometriosis whose pain medications were cut off, without explanation or recourse, because she triggered a high-risk score from the proprietary algorithm. The details of the story are both fascinating and disturbing, but here’s the pertinent part for my purposes:

Appriss is adamant that a NarxCare score is not meant to supplant a doctor’s diagnosis. But physicians ignore these numbers at their peril. Nearly every state now uses Appriss software to manage its prescription drug monitoring programs, and most legally require physicians and pharmacists to consult them when prescribing controlled substances, on penalty of losing their license.

This is an obviously complex and sensitive issue, but it is hard to escape the conclusion that the use of these algorithmic systems exacerbates the same demoralizing opaqueness, evasion of responsibility and cover-your-ass dynamics that have long characterized analog bureaucracies. It becomes difficult to assume responsibility for a particular decision made in a particular case. Or, to put it otherwise, it becomes too easy to claim “the algorithm made me do it,” and it becomes so, in part, because the existing bureaucratic dynamics all but require it…(More)”.

Article by Mike Barlow: “…Today’s conversations about AI bias tend to focus on high-visibility social issues such as racism, sexism, ageism, homophobia, transphobia, xenophobia, and economic inequality. But there are dozens and dozens of known biases (e.g., confirmation bias, hindsight bias, availability bias, anchoring bias, selection bias, loss aversion bias, outlier bias, survivorship bias, omitted variable bias and many, many others). Jeff Desjardins, founder and editor-in-chief at Visual Capitalist, has published a fascinating infographic depicting 188 cognitive biases–and those are just the ones we know about.

Ana Chubinidze, founder of AdalanAI, a Berlin-based AI governance startup, worries that AIs will develop their own invisible biases. Currently, the term “AI bias” refers mostly to human biases that are embedded in historical data. “Things will become more difficult when AIs begin creating their own biases,” she says.

She foresees that AIs will find correlations in data and assume they are causal relationships—even if those relationships don’t exist in reality. Imagine, she says, an edtech system with an AI that poses increasingly difficult questions to students based on their ability to answer previous questions correctly. The AI would quickly develop a bias about which students are “smart” and which aren’t, even though we all know that answering questions correctly can depend on many factors, including hunger, fatigue, distraction, and anxiety. 

Nevertheless, the edtech AI’s “smarter” students would get challenging questions and the rest would get easier questions, resulting in unequal learning outcomes that might not be noticed until the semester is over—or might not be noticed at all. Worse yet, the AI’s bias would likely find its way into the system’s database and follow the students from one class to the next…

As we apply AI more widely and grapple with its implications, it becomes clear that bias itself is a slippery and imprecise term, especially when it is conflated with the idea of unfairness. Just because a solution to a particular problem appears “unbiased” doesn’t mean that it’s fair, and vice versa. 

“There is really no mathematical definition for fairness,” Stoyanovich says. “Things that we talk about in general may or may not apply in practice. Any definitions of bias and fairness should be grounded in a particular domain. You have to ask, ‘Whom does the AI impact? What are the harms and who is harmed? What are the benefits and who benefits?’”…(More)”.