Since then, the use of AI has spread considerably: Today, there are nearly 500 AI-generated news and information websites, in at least 14 languages; music-streaming platforms are teeming with songs composed by AI; and Amazon offers thousands of AI-authored books. The tech giants Google, Meta, Microsoft, and IBM have announced their plans to release new artificial-intelligence systems. All of which has been accompanied by forecasts of massive employment dislocations and a transformation of the global economy.

Yet the performance of the AI chatbots that have generated to much excitement about the technology is still uneven, sometimes generating wrong answers or citing nonexistent sources. Traffic to the ChatGPT website fell by about 10 percent in June; downloads of its iPhone app have also declined. And a recent survey found that only 18 percent of U.S. adults had ever used ChatGPT at all. So how much economic impact is AI really having?

Daron Acemoglu is a professor of economics at MIT and a co-author of the recent book Power and Progress: Our 1,000-Year Struggle Over Technology and Prosperity. So far, according to Acemoglu, artificial intelligence has had only a minimal effect on the economy, both in the U.S. and globally. And companies that have adopted it are using it mostly for automation and surveillance. On its current path, the technology’s development has been limited by Big Tech companies’ dominant business model—which depends on creating maximum hype around new products that don’t necessarily have a lot of long-term value—along with a vision of artificial intelligence, dominant in popular culture and the tech industry alike, that sees the ultimate purpose of AI as supplanting human workers and creators.

Ultimately though, Acemoglu believes that AI holds tremendous potential for humanity. But that potential won’t be realized by the technology itself; it will be realized by us shaping it to do what we want it to do.

Before 2017 or so, production processes in the U.S. economy used almost no artificial intelligence at all. Then AI started to take off in very narrow sectors and occupations—mostly for fairly straightforward tasks in white-collar jobs, some in manufacturing.

At that point, the use of AI was heading primarily toward automation—meaning, companies or factories that had tasks they could automate were starting to use emerging AI technologies to do that. Which had the effect of slowing the hiring of new employees who’d otherwise do the jobs they were automating. But the effect was limited. I worked with researchers from the U.S. Census Bureau and found that in 2018, less than 2 percent of American companies were using AI in any way. In 2019, anyone who told you that AI was transforming the economy was exaggerating.

At the same time, some AI-like technologies, including machine learning, were becoming central in certain sectors and for well-known tech companies like Facebook and Google. For example, everything you did on Google was subject to algorithms that fall broadly within the AI domain. Facebook relies on AI for its recommendation algorithms and content moderation.

Bluhm: What’s changing?

Acemoglu: There’s a lot of uncertainty about the capabilities of generative AI eventually, but there’s a lot of investment in it now—and there’s no doubt that investment will grow.

Two areas seem to dominate the use of generative AI now: The first is search-style tasks, like those incorporating ChatGPT. The business model there is monetization through digital ads—so you can see that fundamentally as a continuation of the model developed by Google and Facebook, though Microsoft might become a bigger player in applying AI to this model.

The second is in white-collar jobs. These applications are a bit more varied, though, as before, they’re mainly to automate tasks—as with the U.S. digital-media company BuzzFeed, which uses artificial intelligence to generate targeted content for consumers at scale. Some customer-service companies are trying to be more creative with generative AI.

Still, hype might be AI’s worst enemy. There are claims that AI can do things that just aren’t feasible—or in some cases, advisable. One example is in the education sector: There’s tremendous potential to use AI productively there, but OpenAI unleashed ChatGPT in an uncontrolled, unregulated way into the hands of millions of students, and the education system wasn’t ready.

“While AI-enabled mass surveillance in China is very stark and authoritarian, I think it’s important to see the commonalities between what the government is doing there and what private-sector companies are doing in the United States, for example. Google’s and Facebook’s approach to AI has something important in common with the Chinese Communist Party’s approach: They’re all harvesting and centralizing data, and using it to manipulate people’s behavior. Google and Facebook have categorically different goals from the CCP, and there are categorically different constraints on what they can do—but what they all have in common is their use of centralized information for the purposes of control.”

“In China, the government’s approach is aimed at social and political control. In the United States, huge, private companies’ approach is aimed at other types of control—controlled engagement: They try to get you to click on things, for example, or buy things. In an obvious sense, this is incomparable to sending you to re-education or labor camps. But there are parallels—and I would argue that the applications of centralized data in a non-authoritarian context like the U.S. is damaging to democracy, because democracy needs diversity and decentralization to thrive.”

“There are lots of possibilities for AI to be used in pro-human ways—like complementing work and activities, creating new tasks, providing better information to make us better decision-makers, creating better platforms, and creating greater reliability and accuracy for information inquiries. But that’s not the current path. The current path follows the vision of autonomous machine intelligence articulated by Alan Turing, the brilliant British mathematician who invented the first computer. Turing said that the objective of computer science and AI was to strive toward autonomous machine intelligence. That’s a very attractive vision for people in the media or entertainment industries: It’s not exciting to make a movie about the transformative effects of the computer mouse or hypertext, but it is, to make one about killer robots or super-intelligent machines. The Turing view of AI is common in popular culture, but it’s also become dominant in the tech industry, where it’s in line with business interests. So it’s a view that’s both projected by Hollywood and embraced by companies that want to use AI as much as possible, as quickly as possible—creating an ecosystem of sustained hype-generation.”