Today’s new technologies, from virtual reality to nuclear fusion have recently received record investments from venture capitalists, but their revenues are not growing as fast as technologies of past decades. Startup losses are unprecedented — far larger than in past decades. Share prices and private valuations have also been collapsing in 2022.
Optimists mostly focus on the good news and ignore these facts. They believe that the heavy funding for these new technologies is a good measure of potential and thus any criticism is unjustified. Here is their typical argument:
Paul Krugman and other “experts” criticized the Internet, personal computers, and other technologies in their early years. But these technologies succeeded. Therefore, criticisms of the new technologies are unfounded — or so goes the argument for optimists.
My response is simple: just because some people were wrong about some new technologies in the past, doesn’t mean that every new technology will succeed. In fact, many, if not all, might fail.
What determines whether a technology succeeds or fails?
Any optimistic argument about a new technology should first detail the benefits for a specific set of early users, the more specific, the better. Unfortunately, proponents of today’s new technologies often focus on the benefits to suppliers and not to customers, because most people think in terms of value capture and not value creation. But value creation is the more important of the two because suppliers must create more value than they capture, with one part of this created value offered to the customer in the form of a value proposition. Some examples will, I hope, demonstrate my points:
Within a few years of the first offer of desktop computers in 1975, their software enabled users to edit documents and do calculations in spreadsheets. The value of these applications was easily appreciated because most people understood the limitations of typewriters and pocket calculators — the latter were themselves a recent replacement for adding machines.
The first users of Amazon quickly recognized the benefits of buying books (and music) on the Internet because they had wasted much time searching through bookstores. By ordering from Amazon, readers could avoid driving to different bookstores just to be told that the book wasn’t available. Books were the perfect killer app because of their huge variety.
Today’s technologies should offer equally large benefits. If they can’t be described, then the new technology probably won’t have many sales over the next few years, and we should be suspicious of optimistic forecasts.
A second issue is whether there is progress towards providing or extending the type of benefits described above. I have written extensively in academic journals on topics such as rapidly improving new technologies prior to commercialization, either by themselves or in new combinations such as the smart phone.
For instance, personal computers benefitted from Moore’s Law, because dramatic increases in processing power enabled computers to process more complex and feature-filled software for word processing and spreadsheets. These rapid improvements also contributed to a booming market for enterprise software that helped manage factories, procurement, engineering, and sales by the early 1990s, a result that purportedly enabled the largest productivity increases of the last 50 years in the early 1990s.
Improvements in Internet speeds also made this software better. They also enabled more products to be sold online as sellers used these improvements to include an increasing number of pictures, videos, and Java files on their websites. For instance, fashion and cosmetic suppliers were initially resistant to offering their products online because the poor aesthetics of webpages, mostly text or slow loading pictures, degraded their product.
It’s different today. Several of today’s new technologies are experiencing slow improvements in performance and cost. Thus their revenue growth is slow and their suppliers experience big losses. Here are some examples:
What’s holding virtual and augmented reality back? A physics law
The two biggest problems of VR today are the large physical sizes of files and, for some users, the nausea that comes from using them. Progress in solving these two problems is slow. Sizes are not getting smaller and one outcome is that the Department of Defense cancelled Microsoft’s contract of AR for soldiers because of nausea. Worse, many neuroscientists believe that AR-induced nausea cannot be eliminated.
Reducing size is hard due to a physics theory known as the law of etendue: There is a tradeoff between size and field of view: We can either have small size or large field of view (FoV), but not both. Google Glass has a horizontal FoV of only 12.5°. The FoV is 43° for the larger Hololens goggles from Microsoft. But without progress towards smaller sizes, VR and AR will likely never experience rapid growth.
What’s the future for delivery drones?
Little progress has been made in handing a package to a tenth story resident through a window, a common challenge in high density areas where the economics of home delivery are the most promising. In less denser areas, there are often above-ground power and telephone lines that can cause accidents. Progress in making drones that can maneuver these lines is also slow. Perhaps it is faster than the progress towards making handoffs through a tenth-story window, but slow nonetheless.
And self-driving vehicles?
Safety is understandably important. It took decades of improvements in vehicles and roads for the safety of human-driven vehicles to reach the low levels of mishap that we experience today. For self-driving vehicles to accomplish the same level of safety, they face the ”devilish problem of edge cases,” the thousands of situations that humans learn to handle through years of “becoming an adult.” Progress in teaching machines all these edge cases is “devilishly” slow.
An even bigger problem is the negative impact that ride hailing vehicles, and thus robo-taxis, have on congestion. Ride hailing suppliers such as Uber and Lyft have made no progress along this metric, because drivers spend a significant amount of their time searching for customers. Robo-taxis face the same problem and no amount of training an AI system will solve it.
What about satellite internet?
These services provide value to a slowly growing number of subscribers, but the question is whether their progress is sufficient to enable them (now 500,000) to obtain the current number of global subscribers for cellular phone services (8.6 billion), or even the 12 million that cellular had obtained by 1990.
Large numbers of satellites are needed to provide reliable service because the low-altitude satellites move quickly across the sky and are often hidden behind buildings, trees, and mountains. The number of satellites is growing and hopefully will continue to grow despite NASA’s reservations about potential collisions. The problem is that speeds are slowing due to congestion from the growth in subscribers, even though the number of subscribers is still very small. Speeds should be rising not falling. Even America’s Federal Communications Commission rejected Starlink’s proposal to provide rural services because, in the FCC’s view, the speeds are too slow. If Starlink cannot handle today’s 500,000 subscribers, how can it handle millions of subscribers much less billions to have a big impact?
And artificial intelligence?
AI is the most successful of today’s new technologies because value is being created. But most of this value has come through augmentation of — rather than replacement of — workers. Will this change? It all has to do with progress.
The cost of training an AI system for a specific accuracy is dropping, mostly from using better computers. But when trying to increase the accuracy from say 90% to 99% or to 99.9% there are exponential increases in computing resources, costs, and green-house gas emissions. For instance, GPT-3, the world’s largest language model, still requires human inputs, even with 175 billion parameters and trained on 45 Terabytes of text data.
Tech bubbles are worse when engineers and entrepreneurs don’t understand what value is, what it means to create value, and whether progress is being made towards offering that value to customers.
When I taught my course on the economics of new technology at the National University of Singapore from the late 2000s through the mid-2010s, I talked about these issues for many technologies including nuclear fusion, superconductors, displays, bio-electronics, neuromorphic computing, and other new technologies. I always saw my course as the start of a conversation about a new technology, not the end. But unfortunately, for many participants in today’s bubble, few have even started having this conversation. And that makes me very pessimistic.