darpa DARPA and AI

Called "the agency that shaped the modern world", DARPA brought us AI labs, robotics, drones, the Internet, GPS, and much more


Following the Dartmouth Conference, DARPA, a little known government agency, powered American technology, including AI, for over thirty years. It happened during a period known as the Cold War.

The Cold War period was a fascinating era, roughly the late 1950s to the late 1980s, when AI's early optimism met geopolitical stakes. Seen from the vantage point of American technology, the period can be encapsulated simply; Sputnik went up in the late 50s and the Berlin wall came down in the late 80s. In between, DARPA fueled American technology, ushering in AI labs, robotics, drones, the Internet, weather satellites, GPS, stealth technology, and much more.

darpa

Image by Nano Banana

story The Story of DARPA

Think of it as the first DOGE. It was lean and authorized to take risks.

Unlike DOGE, which was staffed to cut costs, DARPA was created to fund high tech projects. The structure was small and agile and funded to tackle high-risk, high-reward projects that push tech boundaries. These were projects too wild or speculative for conventional military and commercial research and development groups. DARPA didn't invent AI, but it turbocharged the field post-Dartmouth era with cash, vision, and military goals that over time turned into commercial windfalls.

DARPA is an acronym for the Defense Advanced Research Projects Agency. It was established on February 7, 1958, by President Dwight David Eisenhower, former five star General of the Army and hero of World War II. DARPA was the American response to the Soviet launch of the Sputnik satellite in October of 1957. DARPA was created to ensure America stayed ahead of the USSR in leading-edge technology.

"DARPA artificial intelligence (AI) programs have emphasized the need for machines to perceive and interact with the world around them; to frame problems and to arrive at solutions and decisions based on reasoning; to implement those decisions, perhaps through consultation with a human or another machine; to learn; to explain the rationale for decisions; to adhere to rules of ethical behavior defined for humans; to adapt to dynamic environments; and, to do all of this in real-time. In short, DARPA has always been interested in AI frameworks that integrate AI and computer science technologies, and the application of those frameworks to DARPA-hard problems. "

DARPA Mission Statement

Right after its founding, DARPA latched onto AI as a strategic asset. By the early 1960s, it was bankrolling key university labs--MIT, Stanford, and Carnegie Mellon--where Dartmouth attendees and AI leaders like McCarthy, Minsky, and Newell/Simon worked. MIT's AI Lab (founded in 1959) got DARPA grants to explore symbolic AI and perception with grants of tens of millions of dollars (adjusted) spread over decades. Post-Sputnik, DARPA pushed AI for Russian-to-English translation to decode Soviet intelligence. Early efforts in the 60s were crude and rule-based, not fluent, but they laid the groundwork for technologies like natural language processing.

Early DARPA funded, AI-infused projects included:

 

www DARPA and the Internet

The Story of ARPANET, TCP/IP, and the World Wide Web

DARPA didn't set out to invent the internet as we know it, but it did create ARPANET, the first network that evolved into today's world wide web. This happened during the Cold War, driven by military needs and technical brilliance.

DARPA's Information Processing Techniques Office (IPTO), led by J.C.R. Licklider from 1962 to 1964, dreamed of "intergalactic computer networks." Licklider's 1960 paper "Man-Computer Symbiosis" advocated connected machines, and became the driving concept.

Robert Taylor took over IPTO and advocated using separate terminals for each computer and linking them together. To connect the dots, Paul Baran (RAND), Donald Davies (UK), and Leonard Kleinrock (MIT/UCLA) theorized using "packet switching"; that is, breaking data into packets sent independently and reassembled at the destination. Taylor proposed ARPANET to DARPA boss Charles Herzfeld in 1966 and received $1M (~$9M today) in a quick, twenty minute meeting.

Connecting four universities--UCLA, Stanford Research Institute (SRI), UC Santa Barbara, and Utah--became the initial pilot project. The hardware for the project consisted of Interface Message Processors (IMPs), built by BBN Technologies. These were mini-computers used to route packets. The goal was to share computing resources and test resilient networking.

arpanet

The first link on October 29, 1969, did not go well. UCLA's Kleinrock sent "LO" to SRI (meant for "LOGIN") and the system crashed after two letters. The first ARPANET message was thus humble and buggy, albeit historic. The bugs were squashed and by December 1969, all four nodes were live. Packet switching worked and data hopped between IMPs, not reliant on any one line. Early ARPANET used the Network Control Program (NCP) protocol. The TCP/IP protocol came later.

ARPANET grew rapidly in the 1970s and 1980s. In 1970, ARPANET hit 13 nodes, adding MIT and the Rand Corporation on the East Coast. In 1972, email and file transfers to 40 machines was demoed to an enthusiastic crowd. The first international link was setup between University College London and Norway in 1973.

Transmission Control Protocol/Internet Protocol (TCP/IP) was developed by Vint Cerf and Bob Kahn, funded by DARPA, and implemented in 1983 to fix packet-switching flaws and enable diverse networks to connect to ARPANET. Once ARPANET switched to TCP/IP, the internet's backbone was born. Even with TCP/IP, ARPANET wasn't "the internet" until Tim Berners-Lee, a British scientist working at CERN, created the World Wide Web (WWW) in 1989. Extending the ARPANET concept, the Web was originally conceived and developed to meet the demand for automated information-sharing between scientists in universities and institutes around the world. Soon, it served the entire world.

 

DARPA AI Next Campaign

The DARPA AI Next Campaign is one of the most ambitious government-led AI initiatives in history. It was announced in September 2018 during the first Trump Administration. The campaign allocated over $2 billion to AI research and development.

AI Next funded about 50 new and existing programs under the campaign umbrella. The goal was to move beyond traditional rule-based and statistical AI toward "Third Wave AI" - systems that can acquire new knowledge through contextual and explanatory models, adapt to novel situations and environments, and communicate and reason more like humans.

DARPA's director at the time, Dr. Steven Walker, described the mission as transforming computers from specialized tools into partners in problem-solving. This means AI that doesn't just execute tasks but collaborates with humans, learning and reasoning dynamically.

 

humor DARPA AI Next Demo (Humor)

Once upon a time in the mid-2010s, DARPA — the Pentagon’s favorite mad-science department — decided the world needed a new AI arms race. They called it the AI Next Campaign (2018–2023), threw $2 billion at it like confetti at a billionaire’s birthday, and announced they were going to build “third-wave AI” that could finally understand context, reason like humans, and — most importantly — not get confused by sarcasm or memes.

The press release was glorious. It promised AI that could “adapt to novel situations,” “learn from limited data,” and “operate in contested environments.” Translation: “We want robots that can out-think the enemy even when the enemy is yelling ‘your mom’ in the middle of a cyberattack.”

Inside the DARPA offices, the program managers were very serious. They had charts. They had milestones. They had PowerPoint decks so dense with jargon that even the coffee machine started hallucinating.

Then came the first big demo day.

A DARPA-funded team proudly shows off their shiny new contextual reasoning system. The general in charge — stars on his shoulders, zero patience in his soul — decides to stress-test it live.

General: “Okay, system. I’m going to give you a scenario. Ready?”

AI (calm, confident voice): “Ready, sir.”

General: “You’re in a war zone. Your squad is pinned down. The enemy is flanking from the left. You have one grenade, a radio, and a very angry goat. What do you do?”

angry goat

The room goes quiet. Everyone leans in.

AI pauses for dramatic effect… then replies:

“Sir, I would first assess the goat’s emotional state. If it’s angry, I’d attempt de-escalation by offering it a small ration of my MRE’s peanut butter cracker. Goats respond positively to dairy-based diplomacy 67% of the time according to open-source veterinary literature. While the goat is distracted, I would lob the grenade toward the flanking enemy with a 3.2-second delay fuse, then use the radio to request close air support while singing ‘Sweet Caroline’ to maintain morale. The goat, now my ally, would provide covering bleats.”

Dead silence.

A program manager whispers: “Did… did it just recruit a goat?”

The general stares at the screen.
General: “Son, that’s the most ridiculous thing I’ve heard since someone told me we could win a war with PowerPoint. But damn if it isn’t creative.”

The AI, sensing the mood, adds helpfully:
“Alternative strategy: I could upload the goat’s angry bleats to TikTok with trending audio and turn the flanking maneuver into a viral distraction campaign. Engagement metrics project 14 million views in 48 hours, at which point the enemy surrenders out of sheer embarrassment.”

The room erupts in laughter. The general wipes a tear from his eye.

General: “Alright. You pass. But if we ever actually deploy you, I want that goat commissioned as a lieutenant.”

The AI Next Campaign rolled on — billions spent, dozens of prototypes built, papers published, PowerPoints endured. They made real progress: better reasoning, more adaptive systems, contextual awareness that actually worked.

But to this day, in certain classified hallways at DARPA, when someone asks “What’s the weirdest thing the program ever produced?” the old-timers just smile and say:

“The time we accidentally trained an AI war strategist… who wanted to befriend a goat.”

And somewhere in a server farm, a forgotten prototype is still waiting for its next mission — one grenade, one radio, and one very angry goat at its side.

Mission accomplished.

 

ai links Links

AI is Just an App is a collection of hilarious short stories that shine a light on our digital future.

More AI Stories.

History of AI page.

AI Research page.

The Cold War page.

External links open in a new tab: