Audrey Woods, MIT CSAIL Alliances | July 27,2026
In his recent conversation on Building 32, MIT CSAIL Senior Research Scientist David Clark tells a story from 1995 when he got into a conversation with an economist and friend of his. “He said, ‘David, you have to understand the internet is about routing money, and routing packets is a side effect. And you screwed up the money routing protocols.’ And I said, ‘I didn't design any money routing protocols.’ And he said, ‘that's what I said.’”
This “negative epiphany” is one of the many lessons Clark learned from building the Internet and watching it shape society. Now those experiences are shaping his thoughts on the technological tidal wave of AI, inspiring his current efforts toward understanding what comes next.
“I initially thought, ‘I'm just going to leave AI to the next generation. I'm this old internet guy.’ Boy, did that not work.”
THE MONEY ROUTING PROBLEM: DESIGN DECISIONS FROM THE START
While Clark will “say without apology, the benefit of the Internet has outweighed the bad parts,” he’s learned a lot about human nature and the way people shape systems in unforeseen ways. One example he cites is the money routing problem his economist friend was talking about. In the beginning, Clark and his colleagues designed the Internet as an intentionally decentralized system. But by not explicitly creating protocols for, as an example, money routing, big players like Facebook solved those problems and built gigantic empires of profit based on that. Now, as AI systems are built, Clark advises, “If you want to solve a decentralized system, which might be more socially controllable, solve the money routing protocol and the packets will come along.”
Similarly, identity on the Internet was an issue Clark says was never fully solved. “We sort of have an email address, but it's not very well protected. I can pretend to be you and fool people and things like that.” This has been a problem for years, but the emergence of AI has kicked this question into high gear. “It's become pretty obvious to people right now, we better figure out how AI agents are identified and what attributes they have… When you talk to them, what happens? Can they harm you? If they harm you, can you sue them?” The answers are “still up for grabs,” although Clark says AI has an advantage over the Internet in that it doesn’t look likely to be a “tippy” market. Where Facebook can build a walled garden where everyone must be on Facebook to use Facebook, “not everybody has to be on ChatGPT, or not everybody has to be on Claude. I can use Claude and you can use Gemini and somebody else can use chat GPT.” When it comes to building verification systems, ways to monetize, or other foundational parts of the new AI market, Clark says, “I don't think anybody's going to win. I don't think it's a tippy market. I think we can have contested spaces.”
REGULATION & GOVERNANCE: ACT EARLY
The government, of course, will have a say in what happens, arguably more so than it did with the Internet. In the early days of building what would eventually become the Internet, Clark and his colleagues were “just twelve dudes in a room.” The government was definitely involved, with NSA representation in that room and funding from the Department of Defense, but there wasn’t the public attention and focus on the technology as they built it that there is with AI today. “In the 1970s, who cared?”
AI, by contrast, is attracting much more attention, conversation, and concern. “The whole world is watching,” which means the government is acting much more quickly than it did with previous technologies. “It's not measured in decades here, which is what was happening with the Internet. It's measured in a small number of years. And I think the speed of regulation here is informed by people's realization that the regulators came too late to the party with the Internet and with not enough knowledge to know how to do it.”
OPTIMISTIC ABOUT TECHNOLOGY, NOT PEOPLE
Arguably one reason the government is acting more quickly this time around is the realization that technology, however exciting or well-intentioned, can be easily leveraged by malicious actors. “It wasn't until we began to see the emergence of the more centralized apps that we recognized that there was an issue we hadn't appreciated, which is the potential for manipulation by people with a lot of market power.” It wasn’t just cyber terrorism or hackers, which Clark predicted, but larger, corporate manipulation and the weaponization of human instincts to “serve their own goals as opposed to the goals of the user.” And it’s hard to defend against this kind of thing, since malicious actors have the kind of direction and motivation that individuals don’t and decentralized systems struggle to account for. “Most of us don't wake up every morning and say, ‘well, the first thing I'm going to do today is defend myself from disinformation.’”
As we step—or are thrust into—the AI age, Clark thinks regulators, users, builders, and individuals should be hopeful about the technology and skeptical about the ways human beings will use it. “It's okay to be a techno optimist, but you can't be an overly socially optimistic person… The problems we're dealing with here are not technical problems. They are problems of society.” AI is undoubtedly going to “be used by people in ways that are manipulative, harmful, disruptive,” but hopefully, like the Internet, the good will outweigh the bad. “We just have to be much more sensitive to the magnitude of the bad parts.”
Listen to the full conversation—including the TCP “bake-offs” of the 1970s, the day the internet's engineers revolted against their own leadership, and why Clark's colleagues were running something like Zoom in the mid-1980s—on Building 32, or learn more about David Clark on his CSAIL page.
