Few would disagree that ideas are important to innovation and productivity growth. They are needed for new products, processes, and methods for their conception, introduction, and diffusion.

One challenge is how new ideas fit together to enable positive outcomes. Is the initial idea for the concept the most important, the ideas for the implementation, or those for the many problems that must be solved over the course of a technology’s lifetime in order that the technology becomes better in any way we define better?

An economics paper by researchers from Stanford and MIT tries to make sense of these issues. Published in the American Economic Review, the paper analyzes the number of researchers needed to achieve improvements in the number of transistors per chip, sometimes called Moore’s Law, crop yield, usually measured in output per area of farmland, or a new drug. The paper found that the number of researchers needed to achieve these outputs has increased over the last 50 years, thus suggesting that researchers are becoming less effective in finding new ideas either with the same or a new technology.

The paper is controversial because it restricts ideas to the achievement of these improvements. Many people would claim that our world is filled with ideas, even if they might not be associated with the metrics used by the researchers from Stanford and MIT.

For instance, funding of startups by venture capitalists reached records each year between 2017 and 2021 not only in the U.S., but in Europe, China, India, and other countries. This funding went to a wide variety of product categories, both low-tech and high-tech. Would VCs be giving money to startups if those startups didn’t have new ideas? Most people involved with the startup eco-system and university research would answer with a resounding yes.

Let’s go back to the beginning of the startup explosion in 2013. In a negative review of a book titled The Creativity Crisis, a prominent social science researcher wrote:

“In the book, [Roberta] Ness promises to offer new thinking on the topic of scientific innovation, but I did not find much creativity either in her analysis or in her proposed solutions. Is there really a “creativity crisis” in universities? From where I sit, as a professor who teaches entrepreneurship and innovation, the answer is no. Today, in university labs throughout the United States, there are new ideas in abundance. The word “crisis” strikes too dire a note. And the notion of “reinventing science” (to use a phrase from Ness’s subtitle) suggests a kind of hubris that seems inappropriate.”

Although this review seems like it is from a different era, many of today’s Unicorn startups, those valued at $1 billion or more before they went public, were founded around or before 2013. The review captures what many believed back then, and it promoted the type of hype that was to follow, particularly from tech leaders about a long list of new product categories that were being commercialized by startups.  

Let’s return to the paper and consider the reasons researchers might have used their analysis of chips, crop yields, and drugs to test whether ideas are getting hard to find. A key point is that the researchers were concerned with good ideas, not just ideas, good ideas that should lead to positive economic outcomes, certainly a reasonable concern.

Second, although one could choose any number of sectors to focus on, the ones they chose are undoubtedly important ones. The chips used in our phones and computers have enabled increases in productivity, convenience, and other benefits. Increased crop yields have enabled less expensive food, and thus more people to survive. Similarly, new drugs have also enabled longer lives.

But what does this have to do with ideas, one might ask? Improvements in chips, crop yields, and new drugs require many new ideas, and thus cumulatively, we are looking at thousands if not millions of ideas. Most of these ideas were for small and not noteworthy improvements but together these ideas have had a huge cumulative impact on our lives.

Nevertheless, when many readers hear the question: “Are ideas getting harder to find?” they probably do not think of these small ideas, most of them highly technical, that enabled the improvements in chips, crop yields and new drugs. They think of the product categories that are being funded by VCs, the same types of product categories that the social science researcher was pointing to in her review of “The Creativity Crisis.”

Why aren’t these types of new product categories mentioned by the researchers from Stanford and MIT? They certainly exist and could easily be listed. They don’t mention them because we should be more concerned with good ideas, those with positive economic impacts, than just ideas for new product categories, no matter how detailed someone is able to make those categories. After all, many of the VC funded Unicorn startups are losing money, 90% of the publicly traded Unicorns by my count, and thus most cannot be defined as good ideas, at least right now.

What does the paper’s title and conclusions tell us about the technologies being challenged by today’s startups and incumbents? A lot. My forthcoming book addresses the types of improvements being experienced by today’s technologies, and they aren’t very rapid.

For instance, virtual and augmented reality devices aren’t getting much smaller despite the importance of electronics. Drones are not becoming better at delivering to high-rise apartments or navigating power and telephone lines. Speeds of hyperloop vehicles or satellite Internet services aren’t getting faster, and electrical vertical takeoff and landing devices (eVTOLs) aren’t getting cheaper or better. One could argue about which metric is the best, but it is hard to find evidence that these improvements are occurring.

Even generative AI doesn’t seem to demonstrate rapid improvements, despite the heavy hype and optimistic predictions. As one scientist says:

“OpenAI’s innovation wasn’t so much in AI, but in the project of providing access to genAI tools to the public and to developers.” He continues: “the real transition we’ve experienced in the past year has been about the transformation of AI research from private to public. The public is reeling, but not because AI technology itself suddenly accelerated.”

Behind the scenes there were improvements, mostly driven by improvements in chips, those same chips that the Stanford and MIT researchers concluded are getting harder to improve. Neural networks replaced the expert systems of the 1980s, those used when I was a PhD student at Carnegie Mellon, because improvements in chips enabled neural networks to easily ingest and analyze vast amounts of data. And those improvements enabled OpenAI, Microsoft, and others to build bigger models, at least until they couldn’t. Sam Altman, the CEO of OpenAI seemed to agree when he said in April that the “age of giant AI models is already over.”

The Stanford and MIT researchers were pointing us in the right direction when they argued that good ideas can be found in technological improvements that bring society large benefits. We should be looking for those improvements in today’s new technologies, and if we can’t find them, we should be dialing back our expectations. New product categories with fancy names can’t be called good ideas until they bring large benefits, and many of them haven’t, and perhaps never will.