Willem writes*
Presidential elections recently took place in Nigeria. The build-up to these elections saw tragic violence between groups of pastoral Muslims and farming Christians that was sparked largely by fake news circulating on Facebook. This fake news was designed specifically to instigate ethnic violence. Facebook is active in Nigeria with their initiative Internet.org, which aims to realize global internet connectivity, and meanwhile gives users ‘free’ access to Facebook. In Nigeria alone, 53 million new mobile internet users are expected to come online in the next seven years, a profitable market for Facebook.
The spread of fake news is but one example of the disruptive impact that digital technologies can have on society. In particular, artificial intelligence (AI) is expected to have a profound influence on societies in the near future.
Studies on the potential impact of such AI technologies on economic growth are remarkably unanimous in their conclusions: AI is great for growth. A model by the McKinsey Global Institute predicts that between now and 2030, AI will deliver additional economic output of a whopping $13 trillion. A question that has received much less attention, however, is how this growth can translate into economic development.
AI has already made positive contributions to economic development in numerous areas: early detection of epidemics, disease detection in crops, and providing financial services to rural areas, to name a few. However, particularly for poorer countries, it seems that it will be very difficult to actually realize these benefits.
The social effects of technologies are shaped by institutional, political, economic context in which they are rolled out. The example from Nigeria shows that in a society that is already characterized by ethnic divides, AI can reinforce or aggrevate existing systemic biases and lead to inequalities and even violence against certain social groups. Evidence from the U.S. suggests that algorithms and AI applications systematically aggrieve those in lower social classes and those with lower digital ‘literacy’. In poorer countries, which generally have more social inequality and lower quality insitutions, these social effects of AI are likely to be even larger. Furthermore, persisting differences within many developing countries between those who are able to participate in the design and implementation of AI and those who are not, could result in an uneven distribution of the benefits from AI.
Mitigating these negative effects presents a daunting task to policymakers, and demands a wide range of economic and institutional reforms. Even for the richest countries, regulating tech giants like Facebook is proving very difficult. However, the speed at which AI evolves and the potential benefits it promises make that this should be a top priority, particularly for poorer countries. Tragedies like those in Nigeria may unfortunately be a new reality, but inequality, social unrest and political instability resulting from AI must be minimized if development is to occur.
Bottom line: AI may be great for economic growth, but not for economic development. Poorer countries need to implement drastic economic and institutional reforms if they are going to benefit from AI.
* Please help my Growth & Development Economics students by commenting on unclear analysis, alternative perspectives, better data sources, or maybe just saying something nice :).
Nice post, Willem! I had a question regarding the passage:
“The example from Nigeria shows that in a society that is already characterized by ethnic divides, AI can reinforce or aggrevate existing systemic biases and lead to inequalities and even violence against certain social groups. Evidence from the U.S. suggests that algorithms and AI applications systematically aggrieve those in lower social classes and those with lower digital ‘literacy’”
How does this happen exactly? Do you mean that AI causes wealth inequality which in turn leads to (ethnic) tensions? Or are you referring to other mechanisms, such as racial bias of AI bots that are used for criminal sentencing in the us? I tried the link you provided, but it seems to be broken.
Hey Willem! Your example on Nigeria is very interesting, and makes a good case in favor of what you are arguing. However, I think that for your draft/final essay, you would need a little more evidence in favor of the idea that AI is not that great for development. As demonstrated by the recent past, the AI also promises to have very positive impact on the poor countries in industries like agriculture, international development and education. Let’s not forget that many poor countries experience lack of effective schooling/teaching, and AI has been effective in providing online personalized teaching to those in need. This can be added to the list of other benefits like smart agriculture for developing countries, or AI’s contribution to Nepal’s 2015 post-earthquake relief programmes. [https://www.forbes.com/sites/forbestechcouncil/2018/08/23/ai-for-humanity-using-ai-to-make-a-positive-impact-in-developing-countries-2/#601cc8ba1b08].
I am sure you will find more evidence in your favor. The purpose of my comment was to express that your topic is highly debatable/controversial, so you have to be cautious to not end up ‘cherrypicking,’ and excluding the relevant drawbacks to your whole argument. Good luck!
Thank you both for your comments! Kato, you bring up a very good point. The main intention of my blog post was not, however, to provide a balanced perspective on the issue. I think that we are often exposed to the benefits that technological innovation has/can have. My view is that this is partly also because the parties that have a stake in this positive perception (Facebook, Google, etcetera) have a strong influence on our opinion. In this blog post I intended to shed light on an aspect of AI that I think is underexposed.
I fully agree with you that the potential benefits of AI (especially for development) are fantastic, and my point is not to question the usefulness of AI. Rather, I try to highlight the challenges that are involved with adopting AI in developing countries, and show what can go wrong if we don’t accompany AI with the appropriate regulations.
To further expand on some of these potential consequences, and get to Jeroen’s question, the source that I intended to direct to is the book ‘Weapons of Math Destruction’ by Cathy O’Neil (there is a review of the book on aguanomics, I believe, and if you want to read it I can lend it to you). There are many examples in the book, but to highlight one: for-profit colleges in the U.S. used machine learning algorithms to target single mothers and persuade them to enrol in their college. They did this by posting fake advertisements that promised to help people get food stamps and medicaid coverage. The people who responded to these ads, however, were called by for-profit colleges and persuaded to go back to school and borrow large sums of money to pay for the tuition fees (which they often could not actually afford). The colleges specifically target the most vulnerable and desperate people, because they know that richer kids and their parents will see through their marketing and choose e.g. a community college instead. Thus, vulnerable people are targeted and end up accumulating debt to get below average education.
Both examples that you mention yourself are valid. On the one hand, the benefits of technological innovation tend to be distributed unequally, and the middle class in particular loses (see this talk by Daron Acemoglu: https://events.technologyreview.com/video/watch/daron-acemoglu-mit-ai-economic-impact/). Additionally, this effect is probably even worse in countries that lack a reliable social safety net. On the other hand, embedded biases in AI technologies tend to further marginalize groups that are already marginalized (like the predictive policing you mention). Besides this, political manipulation, such as the fake news, also causes social fragmentation.