We need better robots

I have a story in the Sunday NYT this week about robots, and how they’re working their way up the corporate ladder and into the kinds of white-collar jobs we never thought could be automated. (Like mine! And maybe yours!)

It was a fun article to write – I especially dug the designer’s illustration of a suit-clad USB drive – but it’s a dicey topic to wade into, in part because there isn’t a lot of room for nuance. When you write about robots and AI, people want you to be either full utopian (robots are going to save us from boring and degrading jobs!) or full dystopian (robots are going to create mass unemployment and then kill us, probably!).

My own feelings on AI and automation evolved a lot in the course of writing my book. I started out pretty skeptical of the Andrew Yang-type claims that millions of jobs would soon disappear to automation, and pretty convinced that the people warning about an AI apocalypse were just being paranoid.

I was particularly stuck on what economists refer to as the “productivity paradox.” If AI and automation were actually destroying jobs left and right, the economists say, we should be seeing massive productivity gains in U.S. economic data, as companies replace slow, error-prone humans with hyper-efficient computers. Instead, productivity growth has been declining for years.

But as I began to interview AI and automation experts, and look at the ways automation was actually being deployed inside companies, I started seeing another explanation for the productivity paradox.

I touched on this explanation in today’s article, and I’m glad my colleague Conor Dougherty caught it:

Basically, there’s growing evidence that the reason we’re not seeing huge productivity growth is that the robots companies are using to replace human workers these days are kind of shitty.

What I mean by “shitty” isn’t that the robots don’t work correctly, or that they’re not capable of replacing human workers. (They do! They are!) It’s that they’re the kind of machines that don’t actually save us much time or energy, or make the companies who use them insanely more productive. They just do a worse version of what a human used to do, at a slightly lower cost. Think of a grocery store self-checkout machine, or an automated customer service line – do you experience these things and think, “wow, technology is amazing!” or “wow, this sucks, I’d rather deal with a human”?

I love the work of Daron Acemoglu and Pascual Restrepo, two economists who turned me on to this idea. They use the term “so-so automation” to describe the mediocre kind of robots that don’t make the economy more productive, but do displace workers. And their hypothesis, which I find pretty persuasive, is that the reason we don’t see massive productivity gains, even as corporate automation accelerates, is that we’re not getting the right kind of robots.

Most old-line companies aren’t using AI to radically transform their businesses, develop new products, or make themselves massively more productive, because that’s hard and expensive and probably requires more engineering talent than they have. Instead, they’re spending $100,000 on an off-the-shelf R.P.A. package that can replace a dozen people in Accounts Payable, and congratulating themselves on a successful “digital transformation.”

If you believe Acemoglu and Restrepo, then the problem isn’t too much automation. It’s mediocre automation. And the solution isn’t fewer robots, it’s more capable ones that could generate new jobs and lead to massive productivity gains and economic growth, rather than just getting rid of back-office workers by the millions. We need more stuff from Elysium, and less stuff from Up in the Air.

– My book comes out on Tuesday (preorder here!) and I’m doing the podcast rounds, as well as some TV and radio and print stuff. I won’t put every link here, but there’s a nice excerpt in the Irish Times, a fun podcast with Peter Kafka, and an incisive write-up from Kevin Delaney at Reset Work.

– Did you know that book publishers now create “lifestyle photos” showing new books sitting in Instagram-friendly environments for authors to promote on social media? I did not, but now I do. And here, courtesy of Random House, is a lifestyle photo of my book, which is basically an influencer now.

– Last book-related note: I’ve gotten some inquiries from people who want to bulk-order enough copies for their school/company/book club, and have me appear on a Zoom call to talk about it. I love that, and am trying to figure out if there’s some kind of systematized way to do this. In the meantime, if you’d like to do something like this, email me (kevin@kevinroose.com) and let’s chat.

– I’m hoping to use this newsletter as a way to do extended Q&As with some of the fascinating people I interview, which often get whittled down to a quote or two in a print story. Soon, I’ll be posting a conversation I had with Jack Clark, a co-author of the AI Index, whose Import AI newsletter was a big part of what got me interested in AI as a topic. Who else do you want to hear from?

– I got my first vaccine shot! What a long-awaited relief. I really liked Anne Helen Petersen’s essay this week about people who are not ready or eager to get back to pre-pandemic life. I don’t think that’s me, exactly – I miss my friends and family and cubicle! – but I think we’re underestimating how weird this is all going to feel, as things open back up and we start emerging from our bunkers.

– I’m trying to write about NFTs (I know!) and am wrapping my head around the possibility that the future of art and creative expression involves paying hundreds of thousands of dollars for drawings of Spiderman.

– Cal Newport’s new book about email, which is already making me rethink this newsletter thing.

– Lauren Harris’s story in CJR about a pair of teachers who won a giveaway for a small local newspaper in Alaska, and proceeded to run head-first into the pandemic. (Running an obscure small-town newspaper has always been my eject-button career fantasy, so this was a nice reality check.)

– This paper (linked to in Jack Clark’s newsletter) about how AI researchers have figured out how to make robots better at folding clothes.

– I’m plotting my annual re-read of Jenny Odell’s “How to do Nothing,” the best book I know of about resisting the attention economy.

– I’m looking forward to getting my copy of my colleague Cade Metz’s book “Genius Makers,” which comes out a week after mine. Cade is a genius and a mensch, and I’m excited to get a glimpse of what he’s been working on.