What Everyone Is Getting Wrong About AI And Jobs

**Garry Tan** (0:00)
Is AI going to make human labor obsolete? Right now, the loudest voices on both sides of the AI jobs debate are in hysterics. On the one hand, you've got doomers who are convinced we're a couple of years away from near-universal unemployment.

**SPEAKER_2** (0:18)
Over a five-year period, it could wipe out half of white-collar entry-level jobs. Unemployment could spike to 10 to 20 percent in the next five years.

**SPEAKER_3** (0:27)
So we're looking at a world where we have levels of unemployment we've never seen before.

**Garry Tan** (0:31)
On the other hand, you've got people who think AI is an overblown hype that won't fundamentally transform the economy.

**SPEAKER_4** (0:39)
Sam has been telling us we know how to build AGI for years. This just isn't AGI. We're not going to get to AGI next year. Probably not going to save as much money for various workplaces as we thought.

**Garry Tan** (0:51)
The truth is, both perspectives are flawed. All the best indicators we have from history, industry, and common sense suggest AI is going to transform the economy, but not destroy it. Let me explain why.
I want to begin by telling you the strange story of radiologists. Back in 2016, Jeffrey Hinton, a Turing Award winner and one of the godfathers of AI, declared that people should stop training radiologists now.

**Jeffrey Hinton** (1:26)
It's just completely obvious that within five years, deep learning is going to do better than radiologists. Students are going to be able to get a lot more experience.

**Garry Tan** (1:33)
Hinton is one of the pioneers of neural nets, someone who understood better than almost anyone else what the emerging technology was capable of. But he was wrong. Almost 10 years later, demand for radiologists hasn't gone to zero. It's actually at an all-time high. This is despite the launch of dozens of new state-of-the-art AI products that can detect and classify hundreds of diseases faster and more accurately than humans. What explains that? Well, there are a few reasons that are specific to the medical industry, like malpractice concerns and insurance regulation that requires humans in the loop. But more fundamentally, it turns out that when we gave radiologists the tools that sped up one aspect of their job, demand for their services actually exploded. Cheaper scans means more scans, and more scans means more demand for complex diagnoses and treatment planning from radiologists. In other words, when we use technology to push down the cost of using a resource, in this case MRIs and other imaging techniques, demand for this resource and the services associated with it skyrocketed. This is what economists call Jevons Paradox. Jevons Paradox was first proposed in England in the mid-19th century, when the economist William Stanley Jevons observed that technological improvements that increase the efficiency of using coal increased coal consumption across many industries. This ran contrary to the assumption of many at the time, that increased efficiency would lower consumption. In fact, what Jevons showed was it can just as often reveal latent demand, and this new demand in turn can create entirely new categories of work. There are lots of historical examples of this. When containerization made shipping 90% cheaper in the 1960s, some dock workers were initially laid off. But global trade exploded, and this led to the rise of billion-dollar empires in freight forwarding, logistics, and warehouse distribution. Similarly, when cloud computing made infrastructure 10x cheaper in the 2010s, traditional IT roles transformed. Server admins became DevOps engineers and cloud architects managing infrastructure at scales that previously would have seemed impossible. And most recently, as algorithmic improvements have pushed down the cost of inference, demand for GPUs has skyrocketed, not cratered. NVIDIA stock recently hit an all-time high. So what does this mean for how we should think about how AI will affect our labor economy?
Well, as Aaron Levy, the CEO and co-founder of Box recently wrote, we should expect that efficiency increases will actually mean more, not less, demand for services in a bunch of fields. As Aaron writes, when the cost of doing work goes down, the demand for it goes up. And usually, there's a far more pent up demand than we realize. In other words, as AI makes it cheaper, faster, and easier to do things like analyze MRIs, draft legal documents, and write code, we should expect that the demand for radiologists' treatment plans, lawyers' counsel, and engineers' expertise will broadly increase, not decrease. This doesn't mean jobs aren't going to change, and in some cases disappear. In the future, many roles that might have previously involved manual human involvement will probably look more like supervising teams of agents. Humans will still be in the loop. Andrei Karpathy, one of the co-founders of OpenAI, had a similar take. Karpathy argues that AI will first transform jobs that are rote, require little context, and are forgiving of mistakes. Things like customer service agents and data entry. But even then, he thinks many of these jobs will be refactored into manager or supervisor roles, rather than disappearing entirely. We're already seeing this in our companies at YC. Avoka, which is an AI-powered sales agent for service-based industries like plumbing and HVAC, is freeing up customer service agents to do higher value work. Tenor, which is automating the flow of paperwork between healthcare providers, is transforming admin roles from data entry to patient care coordination and complex case management. Often these are horribly boring, rote jobs that suddenly can be a lot more interesting when you're manning an army of AI agents. Lots of the tasks AI is automating for these employees, like dealing with inpatient customers or filling out routine forms, are unenjoyable. And though some of these jobs will disappear as with the Internet, we can expect generally more engaging ones will take their place. So if you're thinking about starting a startup with AI, what should your takeaway from all of this be? First, the AI transformation is absolutely real and advancing as we speak. Don't be like Paul Krugman, who compared the impact of the Internet to a fax machine in 1998 Don't underestimate that change. Second, this isn't the time to indulge in fantasies about fully automated luxury communism or the imminent collapse of the entire human economy. Don't just sit on your couch waiting for a UBI check. AI is the next thing. As big as, if not bigger, than the Internet itself. The future that you're going to build isn't waiting for a permission slip to start. It's being built right now by people who see things that other people don't, just like you.