Does AI adoption widen the chasm? Causal evidence on the effects of education on AI adoption in Kenya
Image from Kenya Life Panel Study data collection.
Study Context
Research has begun to emerge — predominantly from high-income countries — showing potential for wide-reaching effects of AI: from labor productivity and demand, to educational outcomes, to personal health. Yet, determinants of AI adoption, AI usage, and downstream effects of AI on household welfare are largely unexplored, especially in low and middle income countries (LMICs), where over 80% of the world population lives.
Understanding the expected impacts of AI is particularly crucial in these contexts as AI could bear both large risks and economic potential for the Global South. AI has potential to increase worker productivity, but if predominantly adopted by more educated individuals, access to AI could worsen inequality. Conversely, if education is critical for more economically disadvantaged users to productively use AI, then public investments in basic education could be more important than ever to address inequality over the coming decades.
Study Activities
To address these questions, the Kenya Life Panel Survey (KLPS), one of CEGA’s longest running projects, started collecting data on a sample in Kenya of more than 2,000 individuals capturing usage patterns of AI. KLPS is a unique longitudinal panel survey spanning nearly three decades and capturing a rich set of variables including information on respondents’ health, education, demographics, and employment among other outcomes. Using these data, the KLPS team will study which individual characteristics best predict AI adoption and productive use in Kenya. Furthermore, exploiting a natural experiment in education can provide causal evidence on how much increases in education affect AI adoption and patterns of usage. Working with OpenAI to better understand use of ChatGPT — one of the more widely used AI tools in Kenya — the team will also unpack potential mechanisms to explain patterns found in KLPS data.
Expected Contributions
This work will produce novel causal estimates of increased education on AI adoption and usage in a LMIC setting and allow us to better understand the extent to which AI is expected to affect inequality.
