A flight simulator for AI fairness
We learn best by doing. This is a truism beyond a reasonable doubt. Few findings in social science give evidence to a claim as thoroughly and consistently as studies do this one.
Yet we don't currently learn AI fairness this way. Of course, concern about AI fairness is everywhere. But who's supposed to handle it, aside from the overworked experts? (Not the snake oil salesmen!). Knowledge about how to handle AI fairness must be much more widely distributed than it is today, and it will need to be in the hands of everyone who builds, deploys, or manages policy for AI tools.
How will we teach them?
Flight simulators
When I was a kid, I loved Microsoft Flight Simulator. After school, my dad and I used to "fly" from LAX to Las Vegas on maximum realism mode. No wonder I turned out the way I did...
Pilots learn to fly in fancier versions of these simulators, outfitted with realistic cockpit controls and force-feedback mechanisms. These proto-vVR experiences simulate the experience of being in a plane's cockpit. The benefit of these simulators is that pilots can fail in a safe environment. It's okay to make mistakes in a simulation; you'll live to tell the tale (and learn from it).
Flight simulators allow pilots to experience dangerous (but realistic) scenarios. Severe storms, engine failures, and water landings can all be practiced in simulation under the supervision of an instructor.
A flight simulator for AI fairness
Daylight AI's mission is to create safe environments to experience dangerous (but realistic) scenarios in AI fairness.
Are you a mortgage lender (or regulator) concerned about algorithms that redline? We can produce a lending algorithm and set you to work finding whether it's biased.
Are you interested in an algorithm that purports to improve healthcare outcomes, but concerned about how it treats people of color? We can build a simulated version of that algorithm and show you how it might fail.