The Gen AI Frenzy: What’s Hype, What’s Real, and Where’s the Productivity?

Today, I read two contrasting articles. One posited that we are near the peak of investor hype in Gen AI. It argued that productivity gains from this new technology will be incremental rather than transformative. Another article suggested the opposite. It made the distinction between good bubbles and bad bubbles, highlighting that Gen AI has some attributes of a good bubble.

I believe both articles contain elements that are likely to unfold as predicted, as well as aspects that may turn out to be wrong. To tell these apart, we have to drill down to see what those respective bits are. Let’s start with the distinction between good bubbles and bad bubbles.

Bad and Good Bubbles

Bad bubbles involve huge investments in unproductive assets. One of the earliest examples in recorded history is the tulip mania of the 1600s. During the peak of this frenzy, tulip bulbs—the underground organs of a tulip plant—could sell for prices comparable to those of a luxury home. However, investments in tulips did nothing for the real economy.

Good bubbles, on the other hand, may also be fueled by hype and mania. But unlike bad bubbles, they can accelerate the adoption of new technology that drives economic productivity. In her work on the topic, the economist Carlota Perez highlights the railroad tracks in the 1800s (Railway Mania) and fibre-optic lines in 1990s (dotcom bubble) as one of many examples. Speculative bubbles partially (and meaningfully) financed transformative technologies in those cases.

Gen AI could be another transformative technology financed by a bit of a frenzy (much like the Internet and World Wide Web were in the 1990s). For example, I use LLMs to code, and the gains are evident in my work. However, Gen AI’s potential isn’t equally applicable to all areas of the economy. This is why I think some of the sceptics are correct.

The Hype of Gen AI

So far, the biggest winners from LLMs are people who write code and work with text and images (e.g., translation, summarisation, marketing, and other knowledge work). However, they make up a small fraction of the economy.

For instance, the tech sector in the USA employs about 5.6m people. That’s about 3% of the workforce who can do their jobs better and faster with Gen AI. Now, compare that to labour-intensive areas such as retail, leisure, hospitality (20% of the workforce), healthcare (12% of the workforce), and construction and manufacturing (13% of the workforce). These are big chunks of the economy for whom we have yet to see big productivity gains from Gen AI.

In light of this context, I see why the prediction by economist Tyler Cowen is modest relative to what you’ll read in the press about Gen AI. His current estimate (a rough one at that) is that AI will add 0.5% growth per year to the world's economy. This is no small feat since, even at that small rate, you can grow significantly over a horizon of decades. However, this rate is lower than what AI influencers abuzz on Twitter would have you believe.

Amusingly, anyone hyping Gen AI ought to read through some comments on Cowen’s blog for a reality check on how technology spreads through the economy. A few of my favourites are:

• One commenter notes how the promised "paperless workplace," discussed since the 1990s, took nearly two decades due to inertia among decision-makers and practical implementation challenges.
• Another person observes that even though GPT-3.5 launched over two years ago, customer-service chatbots have barely improved.
• This reader adds that technological capability often isn't the main limiting factor. For instance, although Moore’s Law doubled transistor density roughly every 18 months for decades, economic growth still hovered around 2% annually, not 100% every 18 months.
• Also, video calling was widely adopted only after COVID-19 forced its growth, despite being a mature technology prior to that, as this commenter points out.

Separating Hype from Productivity

Gen AI might have elements of hype, but it’s undoubtedly a powerful general-purpose technology that has transformed how we work with text (particularly unstructured data), code, and media. Does this mean it will drive rapid and massive productivity gains across the entire economy? No. The potential gains will not spread equally.

While knowledge workers already see benefits, particularly at the individual level, translating these Gen AI gains into broader organisational efficiency will take time because of institutional inertia and other bottlenecks. Meanwhile, sectors which make up a large share of the economy, such as healthcare, retail, and manufacturing, will need more than Gen AI to become more productive.