Responsible Lab Episode 3: Development Economics and AI Social Impact in Africa
AI and development economics
AI is starting to transcend developed countries such as Canada and the US, as tech companies start to go global.
Marc Andressen famously said “software is eating the world”, but I don’t think he was necessarily thinking about the developing world.
What this episode will do is outline a few recent developments in African tech, then pivot to outline the economics and policy implications.
PayStack, a Nigeria-based fintech company modeled on Stripe, was acquired by Stripe for around $200 million in October 2020. It’s the biggest acquisition in West Africa to date (I recall that PayStack gave a talk at the UC Berkeley Africa conference a few years ago, and even back then, it was clear they were going places). They provide payment services through an API (application programming interface) and just a few lines of code. Congrats to them!
Also worth noting is that, Amazon CEO Jeff Bezos’ personal investment arm, Bezos Holdings, is investing in Chipper Cash, a fintech company focused on remittances (sending and receiving money). Chipper Cash has raised a $30 million Series B round in late November 2020. Chipper Cash was founded by a Ghanaian and a Ugandan (so they are themselves as cross-border as their own platform). Well done!
The immediate goal is to expand into more African countries and offer more financial services.
There is actually an entire tech ecosystem on the African continent, although surprisingly little is known about them (even among development economists) and hopefully, such exciting developments will help raise the profiles of many companies.
Are these developments new? Not really.
In Ghana, the oldest story I am aware of is about a company called Soft Tribe back in the early ‘90s, which focuses on business applications such as payroll, accounting and other software. Computer scientists in Ghana used to call their approach “tropical tolerance”, which refers to software that is built to withstand the local sociocultural and physical environment.
They’ve always been there, but a problem such small firms have historically faced in Africa is that many larger clients understandably prefer to do business with established firms. Of course, the big tech players generally do not have anything remotely resembling tropical tolerance built into their software, and usually have no idea what it is, until it’s too late for it to matter. This has happened countless times and cost millions of dollars to companies, but the overall philosophy is relevant for international policy makers as well.
What I like about Soft Tribe’s current partnership with Microsoft is that Microsoft are at least trying to preserve Soft Tribe’s intellectual property. To the best of my knowledge, Soft Tribe owns 100% of what they do with Microsoft.
Soft Tribe has grown to the point where they also does some outsourcing to other firms in Poland.
The endogeneity problem faced by small tech companies is kind of like how you need work experience to get a job and a job to get work experience.
Development economics and technology
One interesting development in colleges around the world is the rise of computer science as one of the most popular majors. This might or might not have something to do with the 2010 film The Social Network. Although it is based on a book called The Accidental Billionaires (and thus technically not to blame for my critique), I guess the main thing I didn’t like about the film is that it gives the impression to young people that ethics do not actually matter at all. Still, it may be to computer science what Raiders of the Lost Ark was for young anthropologists a generation ago.
Let’s put our ethics and AI hats back on for a second. How should we expect young students in 2010 who were fans of The Social Network and internalized such values to act? There is a large literature in behavior economics suggesting that media can affect behavior, and which we need to be mindful of. The recent Netflix hit series The Queen’s Gambit (which I have not seen and only know that it’s about chess), is single-handedly selling out chess boards and books by itself, so it’s possible for tech to have good outcomes. I think economists will have to be a big part of that discussion.
There’s an entire movement of economists working in tech firms nowadays, focusing on strategy, pricing, platform, design and yes, program evaluations. There are counterpart computer science research labs like Google AI in Accra. You can read a broader profile about AI in Africa here.
Interestingly enough, at UC Berkeley, (the campus I am most familiar with in the United States) for example, computer science is neck-and-neck with economics as the most popular major.
An observation I have made while working with computer scientists (due to my research interests) is that the two disciplines seem much more similar than most economists and computer scientists may realize. There was a nice special issue in the Journal of Economic Theory some years back about this, and I personally believe that there are some synergies that can be exploited to accelerate international development and development economics as we know them.
It should not be too surprising that the popularity of computer science and a rivalry with development economics in particular may be on the horizon in developing country campuses as well, partly because most economists in developing countries are going to be development economists—and also partly because of the ease of building apps and the growing interest of tech in African economies. Of course, these are spaces traditionally associated with (and still associated with) development economics.
We know little about capacity building in the modern era in West Africa when it comes to the economics profession, but starting in the 1990s and 2000s, the great development economist Chris Udry used to nearly single-handedly support many African economics students into top US PhD programs while visiting the University of Ghana’s Institute of Statistical, Social and Economic Research (ISSER).
He is more known for basically creating what we call development economics today and making household production functions and solid empirical work standard, especially in the area of agriculture and technology.
There are other economists such as Jeremy Foltz that have done similar things in Mali over the years. Similarly, Ted Miguel has run the EASST program which invites visiting African students to UC Berkeley for many years. There are obviously many more economics helping with capacity building in more ways than space will permit me to list, but they are all appreciated.
As brilliant, talented and generous as they are, however, these are individual economists, whereas computer science has an entire set of tech companies trying to diversify their fields (they’re not getting as far as they would like for a variety of reasons that are more in line with the social relative to the computer sciences, but they’re at least trying). The closest thing we have in economics is the policy world, but with the exception of the World Bank, the links are generally not quite as tight. Doing a PhD with one organization in mind is worse than joining a program because of a single potential advisor, obviously.
I believe what we’re seeing and will continue to see more “trade” with computer science for some students that would have traditionally gone into economics graduate programs in West Africa. I’m referring to young people with a technical and quantitative (i.e. mathematics and/or statistics and/or economics) social science background. Again, in my opinion, there is as much difference between the two fields of CS and Econ from a technical standpoint. Perhaps an exception might be if you study hardware in computer science, but then again, I’m told that for better or worse, is becoming as rare as studying pure economic theory. The arguments for or against sound rather similar those referring to the empirical wave in economics research, but I digress.
Computer science is becoming synonymous with machine learning and AI in much of the same way as economics is becoming known for applied econometrics and program evaluations.
Obviously, much of the mobile data is owned by mobile network providers, which tend to sometimes partner with academics that express an interest in their data. This should sound familiar to economists who have or are interested in doing work with tech companies in the Bay Area or in Toronto and using their data for research. One of the exciting developments is the applications of machine learning in economics. There was some initial push-back due to thoughts about things like the underlying data-generating process, but my sense is that it is increasingly accepted. As people never tire of saying, the Nobel winner Hal White was talking about the econometrics of neural nets in the ‘90s. Of course, you can count on someone wanting to point out that deep learning is not that different from a logistic regression, or to “just use OLS”, which is something of a meme on #EconTwitter. However, one thing seems increasingly clear:
AI methods are here to stay.
Since causal inference is the order of the day, this paper gives a nice explanation of using deep learning for semi-parametric causal inference and treatment effects, which dovetails nicely into what many policy makers and commercial stakeholders are thinking about. Similarly, structural models also hold much promise when integrated with neural nets.
To just give a sense of what we’re up against in terms of policy impact: a big problem in economic development is informational: it can be something as basic as knowing actual addresses. African locations are generally not as well-represented on Google Maps, although it’s certainly much better than one would think. The idea that location data is important for basic economic activity is baked into every paper ever written on trade and economic geography, especially this one, which is the heir to Paul Krugman’s classic work, in my view).
I’m a little torn about young economists going into computer science relative to economics (being an economist myself), but I ultimately think that we’re still a ways away from an actual trend.
Fast-forward a bit, and there was a contract Google won a few years ago to use Google Maps to enable digital addresses to be available to all Ghanaian households on their phones (yes, there are analyses of the natural experiments on the way as I type this). It would have helped to have had an economist help with this kind of thing from the beginning (and this is something I’ve been advocating to tech companies with some encouraging success), but Google did the best they could at the time.
I also think that is time for economics training to involve actual programming, given the speed at which research data needs are evolving in the profession. There are only so many hours in the day, so I don’t want to belabor the point, other than to say it will organically happen as the need becomes clearer to more of us over time.
There is an entire Ministry of Monitoring and Evaluation in Accra now, and I personally know of many evaluations that have been done by Ministry research officers (although not in this newer Ministry). The dilemma is that most evaluations done by governments never see the light of day. That is, it might not be very transparent.
To be fair, the counter-argument is that this is already the case with commercial research labs in tech companies, so it shouldn’t be very surprising.
As far as I know, the focus is still on randomized experiments for the most part, and this is probably true in most African countries where these are being generally done on the quiet. Of course, the bottle neck is trained staff, and many tasks can be automated to make the process much easier.
At least for now, I think as is the case with government officials in other countries, my sense is that there is less hesitation to rely single-handedly or even just mostly on algorithms for decision-making than there was in the past. However, the hesitation is still there, and rightly so: for reasons that will be familiar to readers of this newsletter: ethics, responsibility and social impact. For example, the sorts of algorithmic biases Africa-based experts will be trying to avoid would be ethnic biases, which would be not just toxic, but literally explosive when coupled with weak institutions.
Of course, technology is always improving and it is up to economists and AI research scientists to make a convincing case to stakeholders and policy makers as the state of the art continues to evolve in both fields.
This is a discussion still in progress.
There still is not much space for more nuanced thoughts to percolate, but hopefully this newsletter edition can play some small role.