Rabbit’s Reviews > A Citizen's Guide to Artificial Intelligence > Status Update

I’m determined to finish this book but it’s very entry level on tech ethics, and the technosolutionism that goes unexamined in chapter 3 (“Bias”) rubs me the wrong way. I may need to put this thing down for now and come back later.

I know we can’t expect every book to be systematically critical but I’m lost entirely on a lack of criticism for predictive policing as a body of work, not just of current predictive policing techniques.

“Predictive policing employs artificial intelligence to identify likely targets for police intervention, to prevent crime, and to solve past crimes. Its
most famous incarnation is PredPol, which began in 2006 as a collaboration between the Los Angeles Police Department and … the University of California … is now used by more than sixty police departments around the US…”

… • it is pseudo-objective;
it evades existing protections against discriminatory reasoning based on
race, gender, and other protected categories;
it obscures the often complex decisions made by developers about how
to interpret and categories facts about people's lives; and
it fundamentally distorts the nature of commerce, politics, and everyday life.

Its surprising, then, that the most persistent objections to the use of Al in
government, commerce, and daily life include allegations of unfairness and
bias. Here are the main ones:
it perpetuates or exacerbates existing inequalities;
it discriminates against minorities;
it hyper-scrutinizes the poor and disadvantaged;
its risk assessments in domains like justice and policing are fundamen.
tally unfair; …

Noted tools for problematic AI-based policing: PredPol, COMPAS

“Anti-discrimination provisions, for example, prevent the use of a person’s ethnicity, sexuality, or religious beliefs from influencing the decision to hire them. A company that uses AI probably purchased the software from another company. How can we be sure the AI doesn’t take these prohibited (“protected”) characteristics into account when making a recommendation to hire a person or refuse a loan? …”

Evergreen question: Is a programmer culpable for the black box of the AI model, and if so, to what extent?

Very nice demonstrative example of the black box AI problem here: a German horse named Clever Hans. Marvin Minsky, ed. Semantic Information Processing (Cambridge, MA: MIT Press, 1968).