Notes on AI, Labor, and Democracy
I was asked to speak at two AI-focused events on Tuesday. Here's what I said
Hello!
Earlier this week I had the privilege of being invited to speak at the Alan Turing Institute’s ‘AI UK’ conference, and then, later that evening, at an event hosted by the Collective Intelligence Project. Before I share my latest article, below, I want to share (roughly) what I said at each.
At the Turing, I moderated a panel on “AI, Data and Labor.” I was joined on stage by Karen Gregory, a sociologist of work, and Matt Buckley, a member of a tech workers’ union. But the guest of honor, joining us virtually from Kenya, was Mophat Okinyi. As I first revealed in an investigation for TIME last year, OpenAI used outsourced workers in Kenya, some of them paid $2 per hour or less, to detoxify ChatGPT. They helped build safeguards into the AI to stop it producing racist, sexist, or otherwise disturbing content. I wasn’t able to disclose this at the time, but Mophat was one of my brave anonymous sources for that story. He’s since given me permission to waive that anonymity, and has shared his devastating personal story in more detail with other great journalists.
On the panel, Mophat spoke eloquently about the injustice he felt at having contributed such vital labor toward an app which, as we all know, subsequently went hugely viral, helping OpenAI surge to a reported $80+ billion valuation. Mophat told us how he has since gone on to establish Techworker Community Africa, an organization devoted to helping data workers in Kenya and beyond, many of whom are struggling with low wages and PTSD as a result of their work. (He’s also a founding member of the African Content Moderators Union, which I reported the establishment of last year.) He's called the job a form of “digital slavery.”
Later the same evening, I spoke at an event hosted by the clever folks at the Collective Intelligence Project, who have been doing fascinating work on aligning AI tools to our collective preferences. (It’s similar work to the team at OpenAI trying to make AI more “democratic,” which I wrote about in this newsletter a few weeks ago.) The question they asked each of us to address during our short lightning talks was: What concrete steps need to be taken to build towards democratic AI?
With Mophat’s words fresh in my mind, I took a different tack to the (great) speakers who came before me. Most of them had discussed different technical and normative questions around soliciting human preferences, but I wanted to talk about money and power. Here’s the gist of what I said:
The conversations you often hear in London and San Francisco about democratizing AI are often at the level of: What values should a chatbot hold? How should it act on those values?
But what I hear from data workers in the global south, when I speak to them for my reporting, is that their central concern is economic justice. Why are their wages so low? Will they still have a job in a month? Model behavior is very far down their list of worries, if present there at all. The same is true for many of the artists and writers whose work AI companies scrape to train their models, often without compensation.
Even if companies could find a way to collect human preferences with 100% accuracy, (putting to one side the complicating factor that humans tend to disagree on a lot of stuff,) we would still exist in a world where ~all the wealth and power that derives from these systems accrues to a very tiny group of individuals.
This distribution of power is not conducive to democracy.
To be clear: finding ways for the public to have an input into the rules that govern AI is necessary. But it's nowhere near sufficient if your goal is to “democratize” AI.
In a democracy, citizens have a stake in the state. If the state does well, citizens share in its rewards.
Right now, when it comes to AI, workers do not have a role analogous to citizens in a democracy, because they have no stake (or a negligible one) in AI’s success. To paraphrase Meredith Whittaker, workers are not AI companies’ citizens, but their subjects.
If their goal is to truly democratize AI, then companies would need to explore mechanisms for distributing (if not ownership of, then at least) a degree of inviolable control over, and perhaps also a percentage of proceeds deriving from, their AI systems as well.
If you’re not following the CIP, I urge you to read up on their fascinating work. Big shout out to Divya Siddarth, Saffron Huang and Flynn Devine for putting on one of the most mind-expanding events I’ve been to in a long time.
What I’ve written
As I wrote in here a couple weeks ago, I was recently in Dubai for a TIME event. While in town, I took the opportunity to interview several leading Emirati officials leading the UAE’s push into the world of AI. As I write in TIME:
In the realm of AI, the U.S. and China are the world’s undisputed heavyweights. But, sandwiched between the two superpowers, the United Arab Emirates is beginning to punch above its weight. The tiny Gulf nation of some 10 million, which appointed the world’s first AI minister in 2017, is betting big on the technology as an engine for diversifying its economy away from oil, and for projecting geopolitical influence beyond its borders […] The UAE casts its autocratic, state-capitalist government as a plus — giving it the ability to quickly marshal its significant resources to achieve what it sees as an epochal project. “What we have in the UAE that’s going to give us an advantage is the decision-making power to make it happen,” says [Faisal] Al Bannai, the official in charge of the [Abu Dhabi government’s AI R&D group]. “Yes, you need some checks and balances. But in many places it is overdone.”
Read the rest in TIME: The UAE Is on a Mission to Become an AI Power
What I’ve read
Disrupted: work in the age of AGI
By Seb Krier & AI Policy Perspectives
If AI adoption mainly takes place in developed countries and leads to changes in global trade patterns, the consequences for developing economies could be more severe. For example, if AI enables advanced economies to automate and reshore tasks that were previously outsourced to developing countries (such as call centers or manufacturing), it could significantly reduce the global demand for goods and services produced in these countries. This could render some developing economies increasingly irrelevant in the global marketplace, leading to reduced export revenues, slower growth, and higher unemployment.
Are you thinking deeply about this problem? Get in touch. I’d love to bounce ideas on or off the record. I'm on Signal at billyperrigo.01, or simply reply to this email!
The Zone of Interest Oscar Winner Jonathan Glazer Said What No One Else Dared to Say
The Zone of Interest is, as Glazer noted, a portrait of dehumanization taken to a genocidal extreme […] It illustrates how numbness to the torture, starvation, and death of innocent people you’ve convinced yourself are less than human turns even the most passive enablers of that violence into monsters […] Glazer was taking a real risk by speaking up for Palestinians on Hollywood’s biggest stage […] How refreshing to see a director honored whose humanism trumps Hollywood taboo.
The Housing Theory of Everything
By John Myers, Sam Bowman and Ben Southwood in Works in Progress
The aggregate, countrywide effect of housing being so limited in supply has been that economic growth in most Western countries has accrued more and more to landowners and less to everyone else. Economist Thomas Piketty famously demonstrated an increase in the share of national income that flows to owners of capital, rather than to labour. But what was less widely acknowledged was that, at least in the US, it was really an increase in the share of income going to landowners, driven by increases in the cost of housing, and that this effect was particularly strong in states that have highly restrictive rules against building more homes.