Why current-generation AI is 'mostly crap,' but still came along at the ‘perfect time’ to save the flatlining economy, according to a boutique VC firm

The wave of A.I. innovation and investment that has gripped Silicon Valley and Wall Street is coming to a close, according to one venture capital firm. It’s also “mostly crap” but nevertheless holds the key to fixing the big problem holding back the American economy. What happens next is the crucial bit, the firm adds.  

“We are quickly reaching the limits of current AI,” wrote Paul Kedrosky and Eric Norlin, both partners at SK Ventures, on their firm’s Substack. In a post titled “AI isn’t good enough,” they argue that less than a year after ChatGPT exploded into the public consciousness, “we are quickly reaching the limits of current AI, whether because of its tendency to hallucinations, inadequate training data in narrow fields, sunsetted training corpora from years ago, or myriad other reasons.” 

More provocatively, they argue, we are in a strange time where the technology is both too advanced not to challenge many people’s employment in the near future, and not even close to advanced enough to deliver genuine productivity gains. Describing a dynamic they call a “workforce wormhole” that is “eating the economy,” they say we either need much better or much worse A.I., but the current limitations of the technology leave us in a “middle zone” where A.I. is already able to rapidly displace huge numbers of workers, but it’s not yet providing enough wider economic benefits.

Here’s what they mean by the wormhole, the middle zone, and a “better” version of A.I.

The era of chronic shortages

The SK Ventures argument focuses entirely on the theme of the 2020s and the pandemic economy: shortages. 

Firstly, there’s a shortage in the building blocks that enable the technology—chips, training data, large language models. That scarcity has in turn driven up prices, making it harder for companies and startups to innovate in a cost-effective manner. Costs for chips, in particular, have remained exorbitant. Chip manufacturer Nvidia, whose market cap topped $1 trillion in June and reportedly holds an 80% market share, has a starting price of roughly $15,000 for its chips, according to the New York Times. Until costs come down, A.I. innovation will stagnate, Kedrosky and Norlin argue. “This wave has been terrific for a few companies, especially Nvidia, but it will be thought of in the future as mostly about piping the A.I. house,” they said.

There’s also a clear and obvious challenge of workers, they argue, pointing to the incomplete recovery in the labor force participation rate. This means the share of people working in the American economy, while obviously higher than it was in 2020, when there was the greatest loss of employment in modern times, will struggle to reach 2019 levels, let alone 2007 levels from before the Great Recession.

To understand what is really happening in the economy and the role of A.I. in potentially solving it, Kedrosky and Norlin use a wave metaphor. While many investors would associate A.I. with the explosive growth of ChatGPT in 2023, they argue this is actually the end of A.I.’s first wave that began back in 2017. That year saw the publication of an influential paper by a group of Google researchers, Attention Is All You Need, which became the foundation for how to train A.I. models. The current wave will last another year or two and will only end once costs fall across the board, the partners wrote. For that to happen, the world will need more and newer models, like tree-of-thought models, cheaper chips, and an “inevitable commoditization” of large language models that will get offered as a service. 

There is some indication that costs to access the infrastructure on which A.I. is built will fall. Amazon has already made clear its intentions to directly challenge Nvidia’s dominant position in chip manufacturing. Other major tech companies, like Meta and Alibaba, have made their large language models available for free to developers. 

The new era of cheaper A.I. computing systems technological advancements will last through 2030, the two theorize. More important, it will help the U.S. economy tackle the looming productivity downturns it faces as it struggles to find enough workers to fill all the open jobs it has.  

A.I. came at the perfect time to address the problems of a shrinking labor force 

Even though A.I. is moving into the next phase of its development, the birth of the industry came at the “perfect time” for the global/U.S. economy, the partners write. The U.S. economy is facing an existential problem whereby it risks not having enough workers to fill all of its jobs. Essentially, the current, extremely tight labor market will be a permanent fixture of the economy rather than a recent trend. “The U.S. workforce fell into a wormhole and disappeared,” Kedrosky and Norlin write. 

Current numbers show a plateauing U.S. workforce after decades of steady growth. The Great Resignation brought on by the COVID-19 pandemic exacerbated this trend, bringing it on all at once rather than having it play out over time. Kedrosky and Norlin estimate that if the U.S. labor market had continued to grow in line with GDP and its pre-pandemic trends, there would be an additional 5 million workers currently employed. They’re not the only ones pointing out these trends in the population. The Census Bureau also recorded data that points to a near future in which the number of workers aging into the workforce doesn’t offset those who are retiring. 

Kedrosky and Norlin believe demographic trends will ultimately lead to a drastic decline in overall productivity as industries like retail, manufacturing, and health care struggle to fill open positions. There’s some indication that those trends are, in fact, here to stay. The overall labor force participation rate is still about a percentage point lower than it was in February 2020. In a workforce the size of the U.S.’s that can equal several million workers. 

A.I. could lower productivity until it gets good enough to truly replace knowledge work

The absence of human workers means that labor has become more expensive than capital, according to Kedrosky and Norlin. And when economies find themselves in such a situation, they turn to automation to solve their labor shortages. In the past, prominent tech CEOs like IBM’s Arvind Krishna and Google’s Eric Schmidt have also pointed to demographic trends in the developed world as a reason to back A.I. innovation. The big difference between the oncoming round of automation caused by A.I. is that it is primarily targeted at jobs that include “tacit knowledge.” Others have made similar claims, saying A.I. will come for white-collar jobs first, a marked change from most previous types of innovation that affect agricultural and industrial manufacturing primarily. 

The issue, though, is some of these jobs are too complex to be automated by the current state of A.I. “We are in a middle zone, however, with A.I. able to displace huge numbers of workers quickly, but not provide compensatory and broader productivity benefits,” the two write. 

That risks creating a world in which some workers are displaced, leaving them out of a job, but by a technology that is still too rudimentary to generate meaningful productivity gains—leaving everyone worse off because workers lose their jobs while the economy remains unproductive. “Not all waves of automation create jobs as speedily as they displace them,” Kedrosky and Norlin write. “And, even more importantly for our purposes, not all waves of automation deliver bursts of productivity that compensate for the displacement.”

This sort of middle-ground innovation, which replaces workers without spiking productivity, was called “so-so automation,” according to a research paper the two VCs cite. A recent example Kedrosky and Norlin cite is call center workers being replaced by generative A.I. that can answer customer service questions. The workers often aren’t redirected to another division of the company; they’re usually just laid off. And the customer ends up having a worse experience because without the exact right prompt for the A.I. chatbot service rep they risk not getting any helpful advice, creating a lose-lose situation for everyone involved. “These products and services will, inevitably, displace a huge number of people, but they will not drive human flourishing,” Kedrosky and Norlin write.

Kedrosky and Norlin’s fellow venture capitalist, Marc Andreesen, remains convinced of A.I.’s benefits to productivity. “Technology doesn’t destroy jobs and never will,” Andreesen wrote in a widely circulated blog post about the benefits of A.I. And even if A.I. did somehow take every single job from a human, that would actually be a good thing, he argues. “It would be a straight spiral up to a material utopia that neither Adam Smith or Karl Marx ever dared dream of,” Andreesen said of a world in which A.I. did all labor.   

A report from consulting firm McKinsey would seem to back up the Andreesen school of thought that A.I. will lead to productivity growth. McKinsey’s research estimates productivity will increase between 0.1% to 0.6% a year through 2040, although it does acknowledge that part of the variance in its projections depend on the economy’s ability to funnel workers who lose their jobs into new positions. 

Complicating Kedrosky and Norlin’s thesis is the possibility that once a generation of workers becomes trained in A.I. they could further displace existing workers who lack those skills. Whether that possibility would spur productivity as more skilled employees took over from less skilled ones or neuter it as a result of mass unemployment remains to be seen. 

What the two feel is certain is that automation without “high productivity gains” results in “economic disruption,” according to Kedrosky and Norlin, especially when workers are as hard to come by as they are in the current labor market. 

The next wave of A.I. innovation will be paramount to helping the U.S. fix the macro problems it will face as a result of its shrinking workforce, argue Kedrosky and Norlin. Without it, the U.S. economy risks falling head first into the “workforce wormhole” that risks eating it alive. “We need to look past the limits of current AI technology if we are to break free from the past few decades of automation and compensate for the gravitational forces dragging the U.S. workforce into that wormhole.”

Source link

About The Author

Scroll to Top