From the legal blog at Wired News: AI Cited for Unlicensed Practice of Law

Frankfort Digital Services set up a web-based AI expert system to automatically fill out bankruptcy forms, which included citing relevant case law. The U.S. Bankruptcy Court took a dim view. Digging a bit deeper, apparently the way the software filled out the forms caused Frankfort to be classified as a “bankruptcy preparer” under 11 U.S.C. 110, and as such the company was required to list itself as a preparer and have someone sign the forms. (So, not quite as fascinating as the headline suggests.)

But this brings up the issue of how to apportion responsibility (and liability) when an expert system is making important decisions with little human involvement. If something goes wrong, should the creators be blamed? Or the handlers providing day-to-day oversight? Or, more fancifully, is there a way to punish the AI itself?

The debate about this has been warming up for decades now in medicine, automated trading on Wall Street, tax preparation, etc. I expect it will get more interesting as AI technology continues to improve. An interesting up-to-the-moment SF take is Charles Stross’s Accelerando.