Funding the future: A look back on federal tech investment

In 1992, Klee Dienes was an MIT undergraduate working as a manager of the university’s web server and paying for cheeseburgers with student loans. Around him, a government research network called ARPANET was becoming the internet.

One day, he demonstrated MIT’s internet research to a visiting senator from Tennessee named Al Gore, who was pushing legislation to fund the network’s expansion. At the time, the internet “really wasn’t a sure thing,” Dienes said.

It wasn’t the last time he had a front-row seat to innovation. Over the next decades, he moved from Apple, where he helped build Mac OS X, to the Army, where he flew medevac helicopters in Iraq, to a defense technology company he founded. He has seen government technology investment from the funded lab, the private giant and the cockpit.

The public and private sectors have different functions, Dienes said. Industry “innovates by iterating very quickly on things that it hopes people will pay for,” while the government “can act on things that are purely for the benefit of society,” he added.

His path tracks the story the country is marking this year as the United States turns 250. The internet is the most famous technology the federal government helped build, but it is far from the only one. 

For over two centuries, the government has funded technologies decades before any market existed for them and tolerated failures no investor would accept. Now artificial intelligence is testing whether the country still makes those bets.

A world without the computer

The internet wouldn’t exist without a machine that itself was a government gamble. Before there was a network to connect computers, there had to be computers — and those, too, came from public money no private company would have risked.

In 1943, the U.S. Army needed to calculate artillery firing tables faster than humans could produce them, so it backed an untested idea: a room-sized machine of about 18,000 vacuum tubes that could do the math electronically. That machine, ENIAC, became the first general-purpose electronic digital computer.

Just how much rode on that bet? “In the absence of American government support, the digital electronic computer would not exist,” said Matthew Hersch, associate professor of the history of science at Harvard University.

The machines that made modern computing possible were largely developed in the United States and relied heavily on government spending. The U.S. government commissioned some of the first vacuum-tube computers and then became the dominant early buyer of transistors and integrated circuits, creating the market that drove manufacturers to make those components cheaper and more reliable.

“No other country could have picked up the slack,” Hersch said. German computing experiments were destroyed by Allied bombing during World War II. Britain secretly built its Colossus code-breaking computers, then destroyed them in 1945, believing the technology too powerful to exist in peacetime. The Soviet Union lacked the industrial base to manufacture semiconductors in volume.

Without federal money, the few American engineering professors struggling to build simple logic circuits in the early 1940s “would have been starved of funds and abandoned their work,” Hersch said.

In that world, custom-built analog machines would run radios, missiles and factories, with no software industry, no personal computers and no smartphones. On the plus side, “nobody would worry about AI, hacking or electronic spying,” Hersch said.

A habit as old as the republic

Most political change is driven by crisis, Hersch said, and much of the 20th century’s technology came from immediate threats: the Chemical Warfare Service during World War I, the atomic bomb during World War II and NASA during the Cold War space race. But in the 18th and 19th centuries, leaders understood that a growing nation needed trained engineers and new inventions to prosper, without waiting for a crisis.

The government established the U.S. Military Academy at West Point in 1802, which later reorganized as a civil engineering school on the model of France’s École Polytechnique, and the government also chartered canal companies and funded the Lewis and Clark Expedition.

“The founders of the republic were steeped in technology and wanted to see government work actively to promote it — not to enrich themselves, but to enrich future generations of Americans and, more importantly, preserve America’s economic and political independence,” Hersch said.

The clearest early example is also the least remembered. In the 1790s, the government backed Eli Whitney’s plan to build muskets with interchangeable parts, called Armory Practice, to solve “a military problem that did not yet exist: arming a much larger nation in time of war,” Hersch said. By the 1850s, it was the American System. A century later, the world called it industrial mass production.

“This should rank as one of the greatest technological gambles in history,” Hersch said, “especially as Whitney’s scheme never worked completely.”

Government investment does have a failure mode. The same public money that funds world-changing research can also pay for tools that barely work — and Dienes has been on the receiving end of both.

The greatest threat to him in Iraq, he has written, “was not enemy fire but rather the broken software I was getting from industry.”

In 2009, when he was flying medevac missions, his unit was issued software to program its radios. The Windows-based program opened to a blank screen, he said. The menu held 17 numbered actions — for example, reset the radio or save the configuration — and each action opened a new window while the blank screen remained. He couldn’t understand who would build something so maddening to use, until he moved to the contracting side himself.

“Now I know the people who would do such a horrifying thing,” he said. “It’s the lowest price technically acceptable bidder implementing 17 items in a statement of work according to a list. That’s what we got. And should we be surprised by that? No.”

His frustration was part of the reason why he founded Hadron Industries, which builds mission-management software for U.S. defense and national security systems or, as his company puts it, “for missions that must not fail.”

‘What is your moat?’

The deeper difference between government and industry is what each one chooses to fund, and it starts with a question every startup founder hears.

“If I’m a startup and I go to a venture capitalist, they’re going to ask me, ‘What is your moat?’” Dienes said. “What they’re asking is, ‘How are you preventing other companies from leveraging the work that you’re doing?’”

A moat, by design, demands incompatibility. But he said the government wants the opposite: open standards and interoperability. The internet is the proof. It wasn’t the first national network, as Dienes pointed out. France launched Minitel in the early 1980s and gave many households a free terminal. By the end of the decade, millions of French adults could bank, shop, buy train tickets and read the news online. America Online and CompuServe signed up millions of paying subscribers in walled gardens of their own. But the internet was built in public, so it became a foundation nobody had to monetize. Anyone could add to it. The networks with moats lost.

“When government hits a home run, it tends to benefit all of society,” Dienes said. “The internet benefits all of us. GPS benefits all of us. There are many things that have network effects where the benefit is much greater if someone isn’t trying to put them in a moat.”

A 2019 study conducted for the National Institute of Standards and Technology estimated that GPS had generated $1.4 trillion in economic benefits for the U.S. private sector since it became available in the 1980s, with about 90% of that value accruing after 2010 — decades after the initial government investment.

“Even when the government did not directly develop a technology, it created a legal and economic climate in which invention could flourish,” Hersch said.

The next big gamble

For 250 years, the government has placed the first bet on many new technologies. AI is testing whether that still holds true. The field is nearly 70 years old and ran on public money for decades. Early neural network experiments in the late 1950s were funded by U.S. defense agencies. But leading-edge AI development now happens in private labs, not public ones.

Chris Mattmann, an AI and data science expert who spent 24 years at NASA’s Jet Propulsion Laboratory, said current U.S. policy leans on direct partnerships with global technology companies to compete with China and other rivals instead of fully leveraging the post-war national laboratory network and universities built to develop science and technology for the government’s benefit. 

“This national lab network in partnership with higher education represents a unique U.S. benefit and asset that is not being fully leveraged in the AI race,” Mattmann said, adding that the U.S. needs to move from being simply users of AI to builders of it in the national sense.

Dienes said AI is “incredibly democratizing in a certain way” because  anyone can build prototypes without investing in development. “In that sense, it’s absolutely revolutionary.” 

But AI lacks the determinism of the systems that came before it. “That’s wonderful and dangerous and problematic, and I don’t think we know where that’s going yet,” he said. “I hope we figure it out before bad things happen.”

The stakes are biggest where the government is the only game in town. After all, there’s no alternative to the Internal Revenue Service or U.S. Navy. “It’s easy to say that humans are in the loop” with AI, Dienes said. “But if humans are at the end of a decision cycle that they can’t keep up with and can’t get insight into the process, they’re not really in the loop. They’re just rubber stamping.”

Addressing the concerns about AI is essential because of the technology’s enormous potential. For example, it could help speed up the science initiatives that the government has bankrolled for decades. Tom Kalil, CEO of Renaissance Philanthropy, points to the Protein Data Bank, a government-backed effort launched in 1971 to map the 3D structures of proteins. It took researchers over 50 years to experimentally determine roughly 200,000 structures. Then Google DeepMind trained an AI model that helped generate more than 200 million predicted structures in about a year.

Kalil saw a similar technology evolution happen when he served as deputy assistant to the president for technology and economic policy during the Clinton administration. Kalil helped the Defense Advanced Research Projects Agency (DARPA) secure the authority to run a series of competitions for self-driving cars. A Stanford University team led by Sebastian Thrun won the second Grand Challenge in 2005, and Google co-founder Larry Page later recruited Thrun’s team to start the company’s self-driving car project.

“That’s where Waymo and the self-driving car industry came from,” Kalil said.

None of these payoffs looked likely at the start. Jenn Gustetic spent over 13 years at NASA, where she oversaw funding of the kind of research that produces technological advances. She said NASA’s Spinoff publication catalogs hundreds of technologies that started as space or aeronautics research and became consumer products, health devices, energy solutions, and safety and transportation tools. But she questioned why that work falls to the government instead of industry.

“Private investors are comfortable with buying down product risk, marketing risk and business model risk, but they are much less willing to invest in buying down technical risk,” said Gustetic, now director of metascience and R&D policy at the Institute for Progress.

However, she said government investment is critical where there is no clear near-term business case or market but where a breakthrough could unlock entirely new markets, and she cited satellites, GPS and mRNA vaccines as examples. 

The mRNA vaccines are a case Kalil knows well. A DARPA program called ADEPT funded Moderna’s work on infectious disease years before the COVID-19 pandemic. “Many of the other agencies were skeptical that this was a bet worth making,” Kalil said, but the work helped lay the groundwork for the mRNA vaccines for COVID and now for Ebola.

If the technologies with the biggest payoffs are the ones nobody sees coming, the next one is just as hidden now. Asked whether any currently funded technology might be the internet story people tell at America’s 300th anniversary, Gustetic declined to guess. “No one has a crystal ball on what the next ARPANET investment will be that leads to a transformative platform like the internet,” she said. 

That uncertainty is the argument for broadly funding the curiosity and creativity of the entire research community.