On Lesswrong, when we talk about artificial intelligence, we tend to focus on the technical aspects, such as potential designs, specific developments, and future capabilities. From an engineering perspective, this focus makes sense. But most people here aren't interested in artificial intelligence because they want to know how AI will be designed; the reason we're here is because AI has the potential to radically reshape the world around us.

Longtermists have often emphasized the role economic growth plays as perhaps the most important phenomena of human history. In a quite real sense, economic growth is what distinguishes 21st century humanity from our distant ancestors who had no technology or civilization. Nick Bostrom summarizes this point well,

You could argue that if we look back over history, there have really only been two events that have fundamentally changed the human condition, the first being the Agricultural Revolution some 10,000 or 12,000 years ago in Mesopotamia, where we transitioned from being hunter-gatherers, small bands roaming around, to settling into cities, growing, domesticating crops and animals. [...]
The second fundamental change in the human condition, Industrial Revolution, where for the first time, you have the rate of economic and technological growth outstripping population growth, and so only when this happens can you have an increase in average income. Before that, there was technological growth and economic growth, but the economy grew 10%, the population grew 10%, everybody's still in a Malthusian condition.

Many theorists anticipate that there will be a third fundamental change in the human condition, roughly timed with the development of advanced artificial intelligence. In line with these predictions, economic growth is the primary specific benchmark people have used to characterize potential future AI takeoff.

If economic growth is the essential variable we should pay most attention to when it comes to AI, then our understanding of AI takeoff will be woefully incomplete without a grasp of what drives economic growth in the first place. To help mitigate this issue, in this sequence I will explore the underpinnings of modern economic growth theory, and then try to relate economic theory to AI developments. In doing so, I aim to identify crucial pieces of information that may help answer questions like,

  • How much technological progress in the past has been bottlenecked by investment as compared to insights?
  • How soon after advanced AI is created and turned on should we expect rapid economic progress to follow? Is there typically a large lag between when technologies are first demonstrated and when they heavily impact the economy?
  • What are the key factors for why AI is different from other technologies in its ability to induce rapid growth? Is it even different at all?

To provide one specific example of how we can import insights from economic growth theory into our understanding of AI, consider the phenomenon of wealth inequality between nations in the world. Wealth inequality between nations is ultimately the result of historical economic growth inequality, but things weren't always so unequal. Before the industrial revolution, per-capita wealth was approximately equal for all civilizations--at subsistence level. This state of affairs only changed when economic growth began to outstrip population growth in some nations during the industrial revolution.

AI takeoff can also be described in terms of growth inequality. A local (foom) intelligence explosion could be defined as an extremely uneven distribution of economic growth following the creation of superintelligent AI. A global (multipolar) takeoff could therefore be defined as the negation of a local intelligence explosion, where economic growth is distributed more evenly across projects, nations, or people.

Before we answer the important question of which version of AI takeoff is more likely, it’s worth recognizing why historically, growth inequality began after the industrial revolution. The factors that drove growth in the past are likely the best keys for understanding what will drive it in the future.

Organization of the sequence

Below, I have included a rough sketch of this sequence. It is organized into three parts.

The first part will provide the basic mechanics behind models of economic growth, and some standard results, with an emphasis on the factors driving technological innovation. Upon some research, and a recommendation from Alex Tabarrok’s blog, I have chosen to summarize the first several chapters of The Economics of Growth by Philippe Aghion and Peter Howitt.

The second part will dive into a recently developed economic model under the name Unified Growth Theory which the creator Oded Galor claims is the first major attempt to model the deep underlying factors driving economic growth throughout human history, cohesively explaining the onset of the industrial revolution and the emergence of the modern growth era. To provide some credibility here, the book introducing the theory has been reviewed favorably by top growth researchers, and Oded Galor is the editor in chief of the Journal of Economic Growth.

The third part will connect economic growth theory to artificial intelligence. Little research has been done so far examining the key economic assumptions behind the AI takeoff hypothesis, and thus it is possible to get a comprehensive survey of the published work so far. I will review and summarize the main papers, hopefully distilling the main insights generated thus far into a few coherent thoughts.

Other ways economic growth is relevant

Besides being a fixture of how people characterize AI takeoff, economic growth is potentially important for effective altruists of all backgrounds. For instance, in an effective altruism forum post, John Halstead and Hauke Hillebrandt argue that effective altruists have given short shrift to evidence that the best way to reduce poverty is to spur economic growth, rather than to distribute medicine or cash directly.

Economists have characterized the impacts of climate change primarily by its effects on growth, which has important implications for how much we should prioritize it in our longtermist portfolio. Similar statements can be made about the relative priority of pandemics, recessions, and in general a wide variety of global issues.

Economic growth is also just a critical piece of the human story. Without a basic understanding of growth, one's understanding of history is arguably horrible. From Luke Muehlhauser,

Basically, if I help myself to the common (but certainly debatable) assumption that “the industrial revolution” is the primary cause of the dramatic trajectory change in human welfare around 1800-1870 then my one-sentence summary of recorded human history is this:
"Everything was awful for a very long time, and then the industrial revolution happened."
Interestingly, this is not the impression of history I got from the world history books I read in school. Those books tended to go on at length about the transformative impact of the wheel or writing or money or cavalry, or the conquering of this society by that other society, or the rise of this or that religion, or the disintegration of the Western Roman Empire, or the Black Death, or the Protestant Reformation, or the Scientific Revolution.
But they could have ended each of those chapters by saying “Despite these developments, global human well-being remained roughly the same as it had been for millennia, by every measure we have access to.” And then when you got to the chapter on the industrial revolution, these books could’ve said: “Finally, for the first time in recorded history, the trajectory of human well-being changed completely, and this change dwarfed the magnitude of all previous fluctuations in human well-being.”

New to LessWrong?

New Comment