Hello!

I’ve written a version of this post several times but have never published it. Today, I decided to change that, even though I think this will be a difficult read for some. It’s where I think Responsible AI as a field is going (spoiler: not in great directions) and what that means for actually building AI responsibly. You can start reading the post below.

As always, feedback is welcome. Please do read the footnotes before sharing your rebuttals, though. :)

Trey

Responsible AI, as a field with a capital R, is in trouble¹. While layoffs have affected many, many organizations across the tech industry over the past few years, my perception is that very few with “AI” in the name have been cut as deeply as RAI teams have. I saw two RAI layoffs announced in my LinkedIn feed this morning, after I was already in the midst of writing this piece.

This presents a bit of a conundrum. If every company is becoming AI-first, as nearly every CEO has now stated, one would not be irrational to think that companies should be increasing the staff tasked with ensuring that AI development and usage goes well. Perhaps this is a temporary blip, a symptom of the broader macroeconomic and political environment, and we could see a huge spike in RAI team growth in the next few years. But, I’m not willing to bet on that.

What explains the disconnect between the demand for AI professionals and the lack of demand for RAI professionals? In my view, it boils down to three areas, which I’ve listed in no specific order (violating a personal rule). Note that I’m focused primarily on things that RAI, as a field, could have done differently and not what their parent organizations could have done differently because a) that is what I know best and b) I’d prefer to offer a higher-agency view of potential ways forward. This is in no way an assertion that parent organizations have been perfect partners or sponsors. As you’ll see, I don’t think any of these areas are monocausal, nor do I think solving any of them is likely to be sufficient to getting the field of RAI back on track, hence the random ordering.

• Insufficient technical proficiency

• Visible failures, invisible successes

• Epistemic overreach and lack of product-market fit

Continue Reading