Beyond computers and AI, significant technological breakthroughs are becoming increasingly rare. What explains this phenomenon? Is it a concern? And can Artificial Intelligence provide a solution? We posed these critical questions to Carl Benedikt Frey from the University of Oxford.
Mr. Frey, whether we consider fusion energy, flying taxis, or autonomous driving, we seem to be perpetually waiting for major breakthroughs. Some economists even describe this period as one of technological stagnation. Do you share this diagnosis?
Carl Benedikt Frey: The data quite clearly indicate that stagnation is real. Over the last two decades, productivity performance in Europe, and even in the US, has been largely dismal, apart from a brief surge between approximately 1995 and 2004. There has been a distinct slowdown in productivity across nearly all leading economies. The central question is, why?
Precisely: Why?
Carl Benedikt Frey: That’s where the complexity lies. It’s quite astonishing: the computer revolution, encompassing personal computers and the internet, granted humanity access to the world’s entire body of knowledge. People no longer needed to visit libraries; access to information became significantly easier, potentially even to knowledge that would have been previously inaccessible. All of this should have provided a tremendous advantage for research and innovation. Furthermore, the internet connects top scientists and inventive minds globally, vastly optimizing the research process. Yet, the outcome was merely a decade of modest productivity growth, predominantly confined to the United States.
So, perhaps the digital revolution didn’t quite live up to its promises?
Carl Benedikt Frey: No, the technology itself is absolutely revolutionary; everyone uses it. However, a powerful productivity tool can be utilized in two ways: you can use it to dig deeper, or you can use it to drill more holes. In academia, for instance, the “publish or perish” incentive encourages us to pursue a multitude of projects rather than focusing on a select few. The consequence is that our attention is spread across an increasing number of endeavors. And the more projects you concurrently pursue, the lower the probability of achieving a significant breakthrough in any single one.
Therefore, we’re creating the wrong incentives.
Carl Benedikt Frey: Exactly. Beyond that, rules and regulations make it more difficult to translate innovative ideas into tangible products and processes. Even if Artificial Intelligence proves to be an exceptionally valuable tool for drug development, clinical trials are still mandatory. This process is enormously expensive, typically necessitating collaboration with a large pharmaceutical company. While the pharmaceutical industry might be an extreme illustration, it highlights a broader trend: regulation has significantly increased costs. Another example: cloud technology has substantially reduced the cost of starting companies. Despite this, since the 2000s, we have observed a decline in business dynamism.
What is the connection between this and technological stagnation?
Carl Benedikt Frey: The waning dynamism impacts the types of technologies we develop. Large companies benefit from economies of scale; they are far more inclined to leverage technologies like AI for automation and process optimization. In contrast, smaller or nascent companies are more likely to develop novel products or even pioneer entirely new industries. The issue is this: if, from 1800 onwards, we had exclusively invested in automation, today we would indeed have highly productive agriculture and very inexpensive textiles, but we would lack antibiotics, vaccines, aircraft, computers, or rockets. Growth and material progress emerge when we undertake new and previously unimaginable things. Thus, the decline in startup dynamism is particularly worrisome.
Nonetheless, the past 30 years have been rich with astonishing innovations; consider the smartphone, for instance. Couldn’t this be attributed to a type of measurement error? There’s an argument that conventional methods of measuring productivity are no longer suitable for the digital age.
Carl Benedikt Frey: It is true that traditional tools struggle to measure everything occurring in the digital economy. However, it is also true that we have always failed to measure certain things. Economists have long debated whether measurement errors have intensified over time. If anything, the slowdown from the 1990s to the 2000s appears even more severe. I am convinced that this deceleration is not simply a matter of measurement errors. We should seriously consider an alternative explanation.
And what would that be?
Carl Benedikt Frey: One of the primary drivers of impressive growth during the post-war era was the emergence of entirely new industries built around the internal combustion engine and electricity. The automotive industry, along with its vast network of suppliers, represented the largest industrial undertaking the world had ever witnessed. While computers and the internet have indeed brought forth new sectors and business ventures, these have been concentrated in a few specific locations, such as the San Francisco Bay Area, and they have not achieved the sheer scale of the industries that characterized the Second Industrial Revolution.
Could it be that it simply takes time for these foundational new technologies to manifest their full effects?
Carl Benedikt Frey: Certainly, everything requires time. The First Industrial Revolution took nearly a century to deliver the majority of its benefits. The Second Industrial Revolution, centered on the internal combustion engine and electricity, took approximately four decades to register significantly in productivity statistics. But then, its impact persisted for three or four decades. That’s the key difference with the IT revolution: we observed its effects in productivity statistics for about a decade, and then the impact quickly faded.
Could Artificial Intelligence solve the problem of stagnation?
Carl Benedikt Frey: Yes. Without a doubt, AI represents our best hope for a revival of productivity. However, its success will critically depend on how we choose to employ AI. If we predominantly use this technology for automation, we will achieve a one-time surge in productivity growth, because you can only automate a task once. This would then resemble the computer revolution, where we experienced a boom that then rather quickly subsided. For growth to be sustained, we would need to leverage AI to develop entirely new products, services, and types of industries, on a scale comparable to the Second Industrial Revolution. I am not suggesting this is impossible. But we are not yet seeing it. And almost every AI application I can conceive of aims to automate something we already do, rather than enabling us to do something we couldn’t do before.
So, you see no indications of more fundamental breakthroughs facilitated by AI?
Carl Benedikt Frey: There’s always hope. But the decline in business dynamism certainly doesn’t help. As I’ve already mentioned, it’s typically new companies that are inclined to develop novel product types. Of course, startups like Anthropic, OpenAI, or XAI have emerged. However, players from the tech industry alone will not generate a broad-based productivity surge. For that to happen, AI would need to be integrated into new product development across a wide array of industries.
What would be the ideal framework conditions for AI to unleash productivity once again?
Carl Benedikt Frey: We need to re-establish greater economic dynamism. In part, regulations are impeding us. Regulations such as GDPR were well-intentioned but have enormously increased compliance costs in Europe, which disproportionately impact smaller and newer companies compared to larger ones. Therefore, we should reduce regulatory barriers to market entry. We need to make it easier for companies to experiment—to hire and dismiss personnel, for instance, following the principles of Denmark’s Flexicurity labor law. This empowers companies to adapt their business strategies as new technologies emerge. Last but not least, in Europe, we could do more to enhance companies’ scaling opportunities by strengthening the role of venture capital. We still largely operate with a bank-based financial system, which is better suited for investments in stable companies and physical infrastructure.
In your recent book, you identify developments in both the United States and China that could hinder progress: tech monopolies in the US and the autocratic government in Beijing. Could this present an opportunity for Europe to bridge the gap?
Carl Benedikt Frey: Traditionally, Europe has excelled at catching up. During the First Industrial Revolution, Continental Europe caught up to Great Britain; after World War II, the focus was on catching up to the United States. What remains puzzling, then, is why Europe hasn’t truly been able to catch up since the computer revolution of the 1980s. Especially considering there have been successful industrial policy initiatives in the past, such as Airbus. But that’s where the distinctions begin: when Airbus was formed, aircraft technology was relatively mature. Catching up to Boeing was therefore a comparatively static objective. In developing its own semiconductor industry, European industrial policy proved considerably less successful, precisely because it involved a much more dynamic and fluid target. Computers and AI follow a similar pattern: nobody knows whether the future of AI lies in large or small language models, or something entirely different. This makes it challenging to know where to place one’s bets.
What, then, remains to be done?
Carl Benedikt Frey: Europe should concentrate on its foundational strengths, particularly the harmonization of its single market: a fragmented market inherently limits the returns derived from innovation. Both the US and China share the advantage of possessing large, harmonized domestic markets. Moreover, they are generally less regulated, which facilitates market entry.
