O ne of the joys of writing about business is that rare moment when you realise conventions are shifting in front of you. It brings a shiver down the spine. Vaingloriously, you start scribbling down every detail of your surroundings, as if you are drafting the opening lines of a bestseller. It happened to your columnist recently in San Francisco, sitting in the pristine offices of Anthropic, a darling of the artificial-intelligence ( AI ) scene. When Jack Clark, one of Anthropic’s co-founders, drew an analogy between the Baruch Plan, a (failed) effort in 1946 to put the world’s atomic weapons under UN control, and the need for global co-ordination to prevent the proliferation of harmful AI , there was that old familiar tingle. When entrepreneurs compare their creations, even tangentially, to nuclear bombs, it feels like a turning point.
Listen to this story. Enjoy more audio and podcasts on iOS or Android Your browser does not support the
Since Chat GPT burst onto the scene late last year there has been no shortage of angst about the existential risks posed by AI. But this is different. Listen to some of the field’s pioneers and they are less worried about a dystopian future when machines outthink humans, and more about the dangers lurking within the stuff they are making now. Chat GPT is an example of “generative” ai , which creates humanlike content based on its analysis of texts, images and sounds on the internet. Sam Altman, CEO of Open AI , the startup that built it, told a congressional hearing this month that regulatory intervention is critical to manage the risks of the increasingly powerful “large language models” ( LLM s) behind the bots.
In the absence of rules, some of his counterparts in San Francisco say they have already set up back channels with government officials in Washington, DC , to discuss the potential harms discovered while examining their chatbots. These include toxic material, such as racism, and dangerous capabilities, like child-grooming or bomb-making. Mustafa Suleyman, co-founder of Inflection AI (and board member of The Economist’s parent company), plans in coming weeks to offer generous bounties to hackers who can discover vulnerabilities in his firm’s digital talking companion, Pi.
Such caution makes this incipient tech boom look different from the past—at least on the surface. As usual, venture capital is rolling in. But unlike the “move fast and break things” approach of yesteryear, many of the startup pitches now are first and foremost about safety. The old Silicon Valley adage about regulation—that it is better to ask for forgiveness than permission—has been jettisoned. Startups such as Open AI , Anthropic and Inflection are so keen to convey the idea that they won’t sacrifice safety just to make money that they have put in place corporate structures that constrain profit-maximisation.
Another way in which this boom looks different is that the startups building their proprietary LLM s aren’t aiming to overturn the existing big-tech hierarchy. In fact they may help consolidate it. That is because their relationships with the tech giants leading in the race for generative AI are symbiotic. Open AI is joined at the hip to Microsoft, a big investor that uses the former’s technology to improve its software and search products. Alphabet’s Google has a sizeable stake in Anthropic; on May 23rd the startup announced its latest funding round of $450m, which included more investment from the tech giant. Making their business ties even tighter, the young firms rely on big tech’s cloud-computing platforms to train their models on oceans of data, which enable the chatbots to behave like human interlocutors.
Like the startups, Microsoft and Google are keen to show they take safety seriously—even as they battle each other fiercely in the chatbot race. They, too, argue that new rules are needed and that international co-operation on overseeing LLM s is essential. As Alphabet’s CEO , Sundar Pichai, put it, “ AI is too important not to regulate, and too important not to regulate well.”
Such overtures may be perfectly justified by the risks of misinformation, electoral manipulation, terrorism, job disruption and other potential hazards that increasingly powerful AI models may spawn. Yet it is worth bearing in mind that regulation will also bring benefits to the tech giants. That is because it tends to reinforce existing market structures, creating costs that incumbents find easiest to bear, and raising barriers to entry.
This is important. If big tech uses regulation to fortify its position at the commanding heights of generative AI , there is a trade-off. The giants are more likely to deploy the technology to make their existing products better than to replace them altogether. They will seek to protect their core businesses (enterprise software in Microsoft’s case and search in Google’s). Instead of ushering in an era of Schumpeterian creative destruction, it will serve as a reminder that large incumbents currently control the innovation process—what some call “creative accumulation”. The technology may end up being less revolutionary than it could be.
LLaMA on the loose
Such an outcome is not a foregone conclusion. One of the wild cards is open-source AI , which has proliferated since March when LL a Ma , the LLM developed by Meta, leaked online. Already the buzz in Silicon Valley is that open-source developers are able to build generative- AI models that are almost as good as the existing proprietary ones, and hundredths of the cost.
Anthropic’s Mr Clark describes open-source AI as a “very troubling concept”. Though it is a good way of speeding up innovation, it is also inherently hard to control, whether in the hands of a hostile state or a 17-year-old ransomware-maker. Such concerns will be thrashed out as the world’s regulatory bodies grapple with generative AI . Microsoft and Google—and, by extension, their startup charges—have much deeper pockets than open-source developers to handle whatever the regulators come up with. They also have more at stake in preserving the stability of the information-technology system that has turned them into titans. For once, the desire for safety and for profits may be aligned. ■
Read more from Schumpeter, our columnist on global business:
America’s culture wars threaten its single market (May 18th)
Writers on strike beware: Hollywood has changed for ever (May 10th)
America needs a jab in its corporate backside (May 3rd)
Also: If you want to write directly to Schumpeter, email him at [email protected]. And here is an explanation of how the Schumpeter column got its name.
GPT’s reaction to this article:
The article discusses the shift in the AI industry towards prioritizing safety and regulation over profit maximization. Startups are putting in place corporate structures to ensure safety and are even offering bounties to hackers to discover vulnerabilities in their AI technology. The article also highlights the symbiotic relationship between these startups and big tech companies like Microsoft and Google, who use their technology to improve their own products. The article raises concerns about the potential for regulation to reinforce existing market structures, limiting the revolutionary potential of AI. Open-source AI is also discussed as a wild card that may disrupt the industry. Overall, the article provides an interesting perspective on the changing landscape of the AI industry.