AI Is Not Like Electricity. It Invents, Too.
The upside gets bigger, Mohamed El-Erian says, and so does the danger, because AI helps create new inventions.
Global News Podcast · Israel holds first election since October 7th attacksThe question came up after Anthropic filed for a stock market debut that could value the company behind the Claude chatbot at nearly $2 trillion. According to reporting on the filing by correspondent Valerie Sanderson, the company listed about $16 billion in revenue for the first six months of the year, up more than tenfold from a year earlier, alongside $80 billion in losses. Sanderson noted that some of the loss figure reflected a one-time charge and was about five to six times the prior year level. Sanderson also reported that the filing flagged concern that the most advanced AI systems could bring catastrophic or existential dangers for humanity if developed or used without proper safeguards and oversight, especially if misused or turned into weapons.
Asked what to make of that mix of promise and peril, Mohamed El-Erian, a University of Pennsylvania economist, said AI does not fit the usual story about new technology.
Most innovations, he argues, are what he calls 80-20. “80 is good. 20 has to be risk managed,” he said. He pointed to cars, where society tried to enjoy the benefits while managing the risks.
AI, in his view, intensifies both sides. It is not only, in his words, “a general purpose technology like electricity,” a tool that spreads across the economy. It is also, in his phrase, “an IMI, an invention machine for inventions.”
What he means is that AI helps produce further inventions, and that added inventive power is what enlarges both the benefit and the danger. That, he argues, “takes the 80 to the power of X and it takes the 20 to the power of X.”
The same logic shapes how he thinks about who needs the most powerful systems. He distinguished everyday uses from work at the frontier, by which he meant the most capable AI systems. Most users, El-Erian believes, "don't need to be at the frontier." If a clinic in rural Malawi were given round the clock access to a whole medical library, rather than relying on intermittent visits, that would not require the frontier, he said. National security, he added, is different, and officials there will insist on having the most advanced systems.
He also offered a prediction about how investors will weigh that future. He said he expects the market to value Anthropic at a large figure somewhat below $2 trillion, closer to $1.2 trillion, which he noted would still be enormous.
Questions people ask
What does Anthropic actually do?
Anthropic is an American AI company best known for Claude, a chatbot that answers questions, writes code, and summarizes documents. It was founded by former researchers from OpenAI and has focused on building very large AI systems while publicly stressing safety. It is still private, so filing to sell shares would let ordinary investors buy a stake.
What is a general purpose technology?
A general purpose technology is one that spreads across the economy, like steam power, electricity, or the internet. Economists use the phrase because the economic impact comes less from the invention itself than from all the other uses people find for it. El-Erian’s point is that AI goes one step further by helping produce those follow-on inventions faster.
Why would a rural clinic not need frontier AI?
Frontier models are the most capable AI systems at any given time, usually run in giant data centers at great expense. They can handle harder science, coding, and reasoning problems, but they cost far more to build and use. For many everyday jobs, like searching medical guidance or drafting letters, a smaller and cheaper system is accurate enough.