How Poland and Its Peers Can Advance Their Economies Using Artificial General Intelligence
How Economies on the Low End of High-Income Can Use Artificial General Intelligence to Reach the Frontier
A friend recently asked me whether Poland and its peers can meaningfully incorporate artificial general intelligence into their respective economies. This piece is adapted from my response.
Yes, the Polish economy would gain from exposure to artificial general intelligence. However, the Polish economy can only expose itself to some areas in the industry of artificial general intelligence. First, let's map out the artificial general intelligence industry, examining different scientific and engineering questions asked as well as which economies are able to insert themselves in different areas.
I am not very familiar with hardware matters in artificial intelligence, e.g. the creation of more advanced GPUs which are able to perform tensor operations more quickly, etc. I will focus on the areas with which I am familiar—I recommend learning more about hardware matters from someone better versed in that area.
Not considering hardware, artificial general intelligence can be divided into three major areas1: the field of scientific research into more advanced techniques which will hopefully achieve artificial general intelligence; the field of engineering which applies already-known techniques to create artificial intelligence models; and the field of business, engineering, and design which uses existing artificial intelligence models to create software which solves business problems. Of these three areas, the first two are out of Poland's reach, but the third is arguably not only achievable but necessary.
The first area, the scientific field which aims to discover the techniques necessary to achieve artificial general intelligence, poses a high barrier to entry. For an economy to be competitive here, it needs a strong academic community, something incredibly difficult to foster. Germany, for example, had arguably the strongest academic community of any economy prior to the Second World War, but the war destroyed the German academic community, especially because German academics were disproportionately Jewish. Germany still has not recovered its academic edge, and there is no indication that it will.
Currently, only a handful of economies have academic communities able to make meaningful strides towards artificial general intelligence: the United States, mainland China, the United Kingdom, Canada, the Republic of Korea, and the State of Israel. Of these, the United States and mainland China are dramatically ahead of the remaining, with the United States having an edge over mainland China. As previously established, it is almost impossible for an economy to nurture an academic community competitive on the world stage, so it is not feasible for Poland to achieve this.
A similar story presents itself for the second area. Here, we consider the field which applies existing scientific knowledge about techniques towards artificial general intelligence to engineer artificial-general-intelligence-like models. As in the previous area, this area presents a high barrier to entry, one even higher than the previous. While work in the first area is present in economies such as the United Kingdom, Canada, the Republic of Korea, and the State of Israel, the second area only meaningfully sees two economies: the United States and mainland China, though the United Kingdom adds a small addition through Google DeepMind.
This barrier is arguably even higher as training frontier models now costs around one billion USD, a figure which I expect to rise rapidly. In particular, I expect that training a frontier model will cost around one hundred billion USD in five years' time. Building a technology industry and ecosystem is already incredibly difficult, but the incredible training costs associated with frontier models turn this from difficult to effectively impossible without continental support. Even if Poland and other European economies are able to attract the necessary engineering talent and nurture their respective business environments, the cost of the compute, data, and electricity necessary would require a non-negligible portion of Europe's GDP to be spent on training frontier models. While the United States and mainland China have governments prepared for this, European institutions generally have not been as friendly. Unless Europe as a whole changes course to facilitate massive, continent-wide endeavours in building frontier models, it is not possible for Poland to establish a domestic industry focused on engineering frontier artificial-general-intelligence-like models.
This brings me to the third area, which is not only much more reachable for an economy like Poland but, arguably, necessary. The landmark World Bank report China 20302 argues that, for an economy to advance from upper-middle income to high-income and, onwards, to the economic frontier, it needs to grow what is often referred to as its 'innovation economy'. The report does not use that terminology, and I generally avoid using that term as well because it derives primarily from applied work in industry rather than from academic sources in economics and political science, but those whose expertise derives primarily from applied fields can certainly use that term if it is helpful.
The argument raised in China 2030, which is fairly widely held within the academic community, is that advanced economies, especially those at the economic frontier, e.g. the United States and Taiwan, derive their economic strength from their respective scientific and technological industries. That is to say, Taiwan is an advanced economy because there exist organizations on the island which make discoveries and advancements which are not made elsewhere. For example, TSMC is the world's most advanced chipmaker for CPUs; when they discover a novel technique for increasing CPU speed, for example, they are the first in the world to make this discovery, so they issue a surcharge for anyone who wants to purchase a chip they manufacture with this advancement. Through scientific and technological advancement, Taiwanese firms are able to achieve particularly-high revenues, which keeps the Taiwanese economy at the frontier. To advance from the lower levels of high-income, which is where the Polish economy is currently, to the economic frontier, Poland needs to nurture its scientific and technological industry.
Specifically, it would behoove an economy like Poland's to have a strong industry focused on creating software which applies existing artificial-general-intelligence-like models to solve business problems. The recent few Y Combinator batches have a large portion of their startups focus on applying artificial-general-intelligence-like models to, for example, improve medical insurance processing or facilitate education by streamlining the educators' toil tasks3. An economy like Poland's is much better suited to address these problems. They require far fewer upfront capital costs, generally only the standard single-digit million USD or equivalent figures expected of a tech startup, in stark contrast to the punch-a-hole-in-your-nation's-GDP figures seen when training frontier artificial-general-intelligence-like models. Poland would benefit from a business ecosystem and industry where startups and enterprises regularly apply already-built artificial-general-intelligence-like models to create software which solves business problems. For example, perhaps a startup in Kraków will be formed which aims to use artificial-general-intelligence to help large enterprises manage their supply chain, responding to blocks and relieving bottlenecks.
From here, the next step is to consider the specifics of Poland and its economy to form policy recommendations which can facilitate the growth of a nascent Polish startup ecosystem and industry in this way.
This is just one useful way to structure the field; others are valid depending on the context.
World Bank and Development Research Center of the State Council, P. R. China. China 2030: Building a Modern, Harmonious, and Creative Society. Washington, DC: World Bank, 2013. https://documents.worldbank.org/en/publication/documents-reports/documentdetail/781101468239669951/china-2030-building-a-modern-harmonious-and-creative-society.
I do not know if there are any examples from the recent few Y Combinator batches which aim to achieve these exact directions. These are just examples I pulled from the aether.