Databricks CEO Says the Biggest AI Threat Isn’t Superintelligence

As the leader of AI platform Databricks, which uses a cloud software platform to unite big data storage, analytics, and artificial intelligence tools into a single system, you would expect founder and CEO Ali Ghodsi to be an evangelist for the technology—and he is. But he’s also a realist. On a recent episode of the a16z podcast, Ghodsi, whose company has been on a buying spree, including the naming rights to the football field at the University of California, Berkeley, discussed both the risks of AI as well as the factors that are holding back its adoption in some businesses. (a16z is a major investor in the company, which is valued at $190 billion, with a $7 billion ARR. In addition, Ben Horowitz sits on the Databricks board of directors.) Despite the doomsaying of the past few weeks by some in the industry, Ghodsi says the benefits of AI outweigh its risks, pointing to examples such as Novo Nordisk using AI to cut the time it takes to get insights and to assist with drug discovery. “There are a lot of amazing use cases of AI,” he said. “We should not forget these upsides.” That said, there are still some issues AI must overcome, he conceded. The enterprise bottleneck Today’s AI models are already smart enough to deliver substantial value to enterprises, Ghodsi said. What’s missing is knowledge of how a business works, he says. Without that context, AI is unable to deliver optimal solutions—and that can hurt the broader AI industry as users are sometimes underwhelmed by the results. “The models are smart enough, but they just don’t have the context that exists inside of any organization,” he said. “They have not been in every meeting. They don’t know what’s in everybody’s heads. They don’t know all the processes.” Unfortunately, getting that context into an AI system isn’t an easy procedure. Companies first must digitize everything that’s happening in the organization, transcribing meetings and other data and feeding it to the AI. Databricks calls that an “ontology”—a map of a business’s people, projects, goals, departments, relationships and resources. Businesses that create an ontology, though, give their AI insider knowledge that can help with its suggestions and decision making. Few companies are willing to do that now, however. And that’s where things hit a wall. That bottleneck on the user side results in businesses failing to take advantage of all AI can do. Rather than using autonomous agents to perform work, he said, most companies are just using a chatbot. “That’s basically very, very glorified efficient Google search of the old day,” he said. “People are [also] using it for coding… but there’s no agentic work that’s automated.” There’s also a lack of internal AI expertise at most companies, he said, creating another bottleneck. Even when companies figure out what they want to build, they often don’t have a team that knows how to do that. Cybersecurity risk While much of the talk about AI’s threats focuses on superintelligence and doomsday scenarios, the CEO of Databricks says his biggest concern right now is cybersecurity focused. While acknowledging there’s some chance of an existential risk, the near-term risk of AI-enabled cybercrime, particularly against insecure critical systems, is what businesses should be focusing on, he said. “There’s so much in infrastructure that’s insecure,” Ghodsi said. “If you unleash these agents, they’re gonna find loopholes. They’re gonna find exploits, they’re gonna break in here and there. … You could imagine a scenario also where [AI] starts hopping. Like it takes resources and it starts executing itself elsewhere. So it kind of spreads like a virus. That’s a real risk.” Humans, he said, are ill-equipped to respond to AI-enabled attacks. They can’t react quickly enough and security operations centers are likely to be overwhelmed by the number of simultaneous alerts. The best defense, Ghodsi said, was deploying AI to guard those systems and to automate threat hunting with a type of white-hat hacking to discover vulnerabilities before rogue agents do. “You need to automate all of those,” he said. “You need to have threat hunting that’s automated where you’re actually attacking your own systems automatically with agents.” As for that existential risk, Ghodsi’s not worried, saying “I think that right now the existential risk is close to zero.” Superintelligence, he said, is “very, very far away” and he sees no signs that we’re actually moving toward it. BY CHRIS MORRIS @MORRISATLARGE