Spoiler alert: I’m not going to talk about how ChatGPT responds when asked about economic development strategies. Basically, he replicates the reasonable but mediocre ideas seen in his training set. But ChatGPT’s design, which has given it much better capabilities than its creators expected, provides a valuable lesson in dealing with the complexities of economic development.
For more than ten years, deep neural networks (DNNs) have outperformed all other artificial intelligence technologies, driving significant progress in the fields of artificial vision, speech recognition and translation. The emergence of generative AI chatbots such as ChatGPT follows this trend.
In order to learn, AI algorithms require training, which can be achieved through two main strategies: supervised learning and unsupervised learning. In supervised learning, people present the computer with a set of photos labeled “dog,” “cat,” “hamburger,” “car,” and so on. The algorithm is then tested to determine how well it can predict labels associated with images you haven’t seen yet.
The problem with the supervised approach is that it requires people to go through the tedious process of manually tagging each image. In contrast, unsupervised learning does not rely on labeled data. But the lack of labels raises the question of what the algorithm should learn. To solve this, ChatGPT trains the algorithm simply predict next word of the text used to train it.
Predicting the next word may seem like a trivial task, similar to the auto-complete in Google Search. But ChatGPT’s model allows you to perform highly complex tasks, such as passing the bar exam with a score higher than the most outstanding law students.
The key to these feats lies in the stupendous power of this simple learning process. For this to predict the next word, the algorithm is forced to develop deep understanding of context, grammar, syntax, style and much more. The level of sophistication it reached took everyone by surprise, including its designers. DNNs proved to be able to do a much better job without attempting to incorporate the theories that linguists had been developing for decades into language learning models.
The lesson for economic development is that policymakers must focus on a task that may seem mundane, until excelling at it indirectly forces them to learn more complex development challenges. .
In contrast, the prevailing strategy in the field of development economics has been to distinguish between the immediate causes and the deeper determinants of development, and to focus on the latter. This strategy is equivalent to saying, “Understand the context and meaning of the entire book, rather than trying to predict the next word.”
For example, in their 2012 book Why Nations Fail, Daron Acemoglu and James A. Robinson argues that institutions, by influencing the structure of incentives in society, are the ultimate determinants of economic outcomes. Brown University economist Oded Galore has taken a different approach: emphasizing the complex Demographic and technological change that threw humanity out of Malthusian equilibrium and longer life expectancies, lower fertility rates, and increased investment in education. Taken together, these trends boosted women’s participation in the labor force and increased the availability of skills needed to support technology adoption and economic growth.
Now, do these theories line up with the facts? In the past 40 years, many of the radical changes described by Galore have taken place in the developing world. As the late physicist Hans Rosling observed, the gap between developed and developing countries in life expectancy, infant mortality, fertility, education, college enrollment, female labor force participation, and urbanization has narrowed markedly. Following Acemoglu and Robinson’s argument, developing country institutions could not be so bad if they managed to make progress on so many fronts. In Galore’s plan, progress on all these fronts should explain why developing countries are so close to the developed world in terms of income.
Only this wasn’t the case: The middle country is no closer to US income levels than it was 40 years ago. How is it possible that even the narrowing gap in education, health, urbanization and women empowerment has not narrowed the income gap? Why has progress on deep fixation assumptions not lived up to expectations?
To make sense of this puzzling result, economists call for a Ever-widening technology gap. Rather than an explanation, it is a mathematical necessity: if more inputs do not produce more outputs, then something must be making the inputs less effective.
To explain this unexpected result, it is useful to note that some countries shared two distinctive characteristics: their exports grew much faster than their GDP, and they expanded their exports by focusing on more complex products. Diversified in
To achieve this, these successful countries must adopt and adapt better technologies, adjust the provision of public goods and their institutions to support emerging industries, and reduce inefficiencies and costs by increasing productivity and training workers. should be reduced. In that process, he may have fixed many other problems.
A ChatGPT-inspired growth strategy would focus on one simple goal: improving the competitiveness, diversity and complexity of exports. To figure out how to do this, policy makers will need to learn how to do important things, just as predicting the next word allows ChatGPT to learn context, grammar, syntax and style Is.
Like early AI programmers who were distracted from their complex theories by linguists, policy makers have been distracted by too many goals, such as 17 United Nations Sustainable Development Goals, But applying the ChatGPT strategy to economic development could make things easier: just as the language model only tries to predict the next word, policymakers could try to focus on facilitating the next export, As countries have done. While this may seem like a small step, it can produce surprisingly relevant results.