Editor's Note: Can it be that I can do three in a row? But don't get used to it. The faiV will be off for the next two weeks as I'm traveling. Any faithful faiV readers in Milan who might want to meet up?
I also want to thank a variety of readers who have sent me either encouraging notes on the return of the faiV or items to include in the faiV. Please do send on anything you think should be included, it really helps!
Last week's most popular link was the one about the “Death of Behavioral Economics”—indicating that I'm not the only one thinking about that, but interestingly I didn't get any feedback on what people are thinking about that. Please do share.
Finally, a personal rant that's kind of related to the first item below but only tangentially, so I'll put it here. Yesterday I logged into a webinar that I was really looking forward to based on the content that was going to be presented. But I lasted less than 15 minutes as the presenters were doing something very close to reading from a script. In an age of generative AI that can produce summaries or annotated transcripts or likely soon, virtual presenters with charisma, "performative" presentations of research are just a huge waste of everyone's time. I'm not suggesting that the human element be dropped, but there are people who are good at interviewing people or presenting information. My plea: use them, or don't do the performative thing at all. Just send everyone the transcript.
- Tim Ogden
1. Our AI Overlords
I used to have a regular item called "Our Algorithmic Overlords" but that seems so 2022 now. I'm going to endeavor to keep this from becoming an AI newsletter (if you're interested in something like that, albeit not from a development perspective, I recommend Understanding AI from the tastefully named Timothy B. Lee.) But it's certainly the case that you can't talk about many of the topics we talk about around here without featuring it regularly.
VoxDev is already on the ball with a VoxDev talk on an experiment with a ChatGPT-4-based "virtual business mentor," and the paper is here. I honestly have no idea how they could have already done this experiment and gotten results and written up a paper, so clearly I'm already falling behind. They find no average effect, but heterogeneous effects consistent with various findings from other papers about benefits for the upper part of the ex-ante distribution. You may also want to compare and contrast with another paper from Kenya on text-based business coaching (HT: David McKenzie) which finds small effects early which quickly fade out as people stop engaging with the content.
That could weave nicely into this op-ed from some folks at Linked In, arguing that the human skills around connection with other humans—and with AI models—are more important for the future than the technical skills in building and using AI. It also weaves well into a new paper from David Autor, arguing that AI "can be applied to enable a larger set of workers possessing complementary knowledge to perform some of the higher-stakes decision-making tasks that are currently arrogated to elite experts." David's focus is on how AI could help rebuild middle-skill jobs in the US, but I think that applies just as well, if not more so, to the people running small firms in developing countries.
2. Biases (and Nudges?)
Of course, the outcome of helping those workers with complementary knowledge via an AI assistant depends on what information the AI provides and how it is framed. Ideas42 has taken a look at biases in ChatGPT about poverty, finding that it is more likely to implicitly or explicitly endorse biased ideas about poverty (for instance, that poverty is the result of bad choices and moral failings). Of particular note, the biases become more evident the further you dig into the details.
The source of those biases is the data that ChatGPT was trained on, and the source of that data was human beings who held these biases. So it's not enough to just say that the AI is biased without recognizing that it's just amplifying existing biases. But amplifying existing biases is a big problem!
Where else do we see these biases? In textbooks from around the world. CGDEV looked at sexism embedded in textbooks used as part of national curricula in anglophone countries. They found that those used in low- and middle-income countries include fewer women and girls in their examples and discussions (e.g. a math word problem only has boys), and when they are present are presented in domestic and/or passive roles. I'm betting that those textbooks are part of the training data being used by generative AI efforts focused on those countries.
Should we think about these biases as nudges? Or are they something else?
3. Fraud
It appears that, like Our AI Overlords, Fraud is going to be a regular item. Many of you clicked on the story last week about how Cory Doctorow got phished. And then this week, this story from New York Magazine's Consumer Finance writer popped up. My jaw literally dropped as I started reading this and I had to check several times to make sure that this was being related as a true story. And as far as I can tell, it is, but I still have a hard time believing it.
While this story is extreme (and extremely extreme at that!), the amount of fraud happening out there is staggering. The US Federal Trade commission reports that they received reports totaling $10 Billion in scams in 2023, and of course that has to be a vast undercount given that I think a significant portion of fraud victims would never report it to the FTC. The median amount lost is $500, which when you think about that in the context of the 30% of Americans who wouldn't cover a $400 emergency with cash, is a lot to be siphoned from people. I return to what I said last week—we are really behind the curve on this issue in the financial inclusion space. Protecting people from financial fraud has to be a central part of the inclusion conversation right now, and we can't waste any time or money on the worse than useless "financial literacy" approach to the problem. Take a look at this screenshot from a news story about Super Bowl security and tell me with a straight face that we can teach the average consumer to protect themselves. Or consider how much we are training people to be vulnerable to fraud through such things as government payment systems that look like they are scams.
There are other varieties of fraud than just outright scams and they will become increasingly important in middle income countries. Remember that the median loss reported to the FTC is $500. In India, consumers are being encouraged to engage in high frequency options trading, and the average retail trader has lost ~$1500 (per capita GDP $2800), with 90% of traders losing money. If you're in the US, you might be thinking, "how could they let this happen?" But what you should be thinking is, "things are worse here, where people are betting on the color of Gatorade used in the Super Bowl."
And I can't move on without noting an entirely different kind of fraud that seems to be proliferating. Here's a story about the Dean of Engineering at a state university in Nevada publishing fraudulent papers (no, it's not faking data, it's much worse than that).
4. Financial Inclusion Potpourri
I'm feeling a bit guilty relying on the "grab bag" approach, rather than finding some way to pull things into an at least semi-coherent item, but I'm still building up my stamina, so cut me some slack please.
Thanks to Matthew Soursourian who sent it on, I'm really enjoying this paper on regulators' "financial inclusion trilemma"—yes, it's about the United States, but no it's not really about the United States, it's applicable globally.
Who does the regulation and how those trade-offs are made depends on country-context, and regulator capability. Which raises the question of where "the poor" are. Lee Crawfurd at CGD has a blog post on whether or not the often heard phrase that half the world's poor live in fragile states (which presumably will have an especially hard time with the trilemma) is true, and how changing poverty lines affects the answer.
Here's a new paper about how to offer agricultural insurance that actually works for smallholders: "Can Revenue Index Insurance Outperform Yield Index Insurance?" I feel like insurance is one of those areas of financial inclusion where we keep shrugging our shoulders and thinking we'll get to that later because it's just so hard.
Friend of the faiV Ryan Edwards and colleagues have a three-post series on the cost of trans-national remittances in the Pacific, looking at actual costs of remittances in the Pacific, the role of consumer choice in keeping costs higher than they could be, and why people don't switch to lower cost providers. Ryan and I will have more to say about Pacific remittances in the near future as we look at data from trans-national household diaries in the Fiji/Australia corridor.
5. Other Things I'm Paying (Limited) Attention To
Yes, it's another potpourri item, but I can't bring myself to write that headline more than once. So here are some other things that are interesting, I hope not just to me.
Dustin Moskovitz, co-founder of Facebook, co-founder of Asana, Elon Musk troller, and creator with his wife Cari Tuna of Good Ventures (and by extension Open Philanthropy, and supporter of GiveWell, where, full disclosure, I'm chair of the board) has a long post about their "long journey to doing good better".
Here's a paper about forecasting long-run causal effects found in RCTs of at least 5 years, finding that experience forecasters do better than academics, but neither does particularly well. What I found most interesting/discouraging/amusing is that neither "details about the RCT intervention, or the local study context improves forecasting accuracy" which kind of flies in the face of a lot of critiques of RCT-maximalism, but also not in a good way?
Since it's a three-day weekend in the US, I'm going to assume that people will have some time to recover from ending on some not-so-good news, like we're out of cholera vaccine and conditions are getting worse in places with cholera outbreaks. And it's pretty discouraging when consumer packaged goods companies are withdrawing from the largest consumer market in sub-Saharan Africa because conditions are just too hard.
Video of the Day
I couldn't really leave you on bad news, so here's something that made me smile this week. I'm pretty sure this wasn't generated by AI. Not any less appropriate for work situations than the average current pop song.
The faiV is written by Timothy Ogden and produced by the Financial Access Initiative at NYU's Wagner Graduate School of Public Service
Email: fai-wagner@nyu.edu
To read this in your inbox, subscribe to the faiV.
