Patrick Gunkel Proposed a "Manhattan Project in AI" Way Before It Was Cool

These days it’s very common to hear someone say loudly, in public, that we need a “Manhattan Project in AI” as a national security effort.

In a June 2024 manuscript, former OpenAI employee (and now disgraced investor) Leopold Aschenbrenner predicted that the government would take over AGI development by 2026-2027 by launching a Manhattan Project style initiative and merging leading labs under government oversight.

Palantir CEO Alex Karp made a similar call in early 2025, writing in a new book that the U.S. and its allies should “without delay commit to launching a new Manhattan Project” for AI-driven military technology.

In November 2025 commentary for the Center for Strategic and International Studies, a high-class D.C. think-tank, Jake Kown and Benjamin Jensen directly advised the U.S. Army to take on the project and urged the U.S. government to “start now.”

What nobody knows—and what I hope historians will one day discover—is that American futurist Patrick Gunkel was writing and planning for this very moment way back in 1983, even using the phrase “Manhattan Project for AI.”

The below list is original archival material included as part of a July 23, 1983, memo Gunkel sent to Ed Fredkin and Marvin Denicoff, two well-known and well-connected computer science researchers affiliated with MIT. ¹

The context was Gunkel’s attempt to plan a conference on artificial intelligence at MIT.

Timing wise, this list was drawn up several years after Gunkel had coined the term “ideonomy” and began talking about the science of ideas, and about six months before his critical grant from the Richard Lounsbery Foundation came through.

I find the memo date—July 23, 1983—meaningful from a circumstantial evidence standpoint.

Gunkel’s great long-time benefactor, Herman Kahn of the Hudson Institute, had died less than three weeks previously, on July 7.

Gunkel was likely on the ropes, desperate to find some new angel investor for his intellect and probably churning out a bunch of project proposals like this one.

Unfortunately for Gunkel, the plan for an AI conference at MIT failed for reasons that are clear from the transmittal memo preceding this list.

I will share the transmittal memo at a later date.

For now, I hope this release is enough to eventually get serious people and even historians of AI to pay some attention to Gunkel.

Among several other goals, I’m hoping to create more accountability for the woman who privately holds this memo and numerous other important documents and items she took ownership of after Gunkel passed away.

Unfortunately, in 2024 this woman cut off communications with me for unknown reasons despite what seemed to begin as a promising collaboration.

I apologize this document is missing page 9, items #19-21.

I don’t know whether I failed to photograph it or whether it had been removed from the document.

Gunkel was getting into DARPA and military-industrial-intelligence type stuff by #19, so perhaps Fredkin had him remove the page. Or maybe I was in so much of a hurry that I just forgot to photograph it.

I hope the woman I’m referring to has a change of heart and has not already sold her Gunkel materials to a private collector, which is what I fear.

I hope that she delivers, or arranges for them to be delivered, to the MIT Distinctive Collections archive as Gunkel himself wanted, so that future historians will be able to know for sure.

In 1982, Gunkel was incredibly motivated by Japan’s surprise launch of its 5th Generation Computer Systems project.

According to Wikipedia, Japan’s project “aimed to create an ‘epoch-making computer’ with supercomputer-like performance and to establish a platform for future advancements in artificial intelligence.”

Gunkel thought an announcement like that should have been the United States’ to make.

Although the U.S. did respond to Japan’s project later in 1982 by setting up the Microelectronics and Computer Technology Corporation (MCC), placing it in control of a former Admiral and military-intelligence spook named Bobby Ray Inman, and although Gunkel even rubbed shoulders with some of the MCC’s employees in Austin in the late 1980s, Gunkel clearly thought the “Manhattan Project” framing was important. Either he was unsatisfied or unaware of the MCC effort and felt the need to keep pounding the table about the importance of a crash program to build AI.

Historians of this time period may point out that Gunkel’s ambitions were premature.

In retrospect, it might be argued, numerous other technologies “needed” to be invented in order for us to get where we are today with large language models (LLMs).

I feel like this criticism fails on two-and-a-half points.

First, it ignores the importance of framing.

We like the JFK-inspired “moonshot” analogy these days, and Gunkel perhaps could have used that same analogy here. (Granted, a “Manhattan Project” analogy may have seemed a bit, er, undiplomatic, as a response to Japan). Whatever the case, Gunkel identified the need to catalyze support for such a massive project at the same time it evoked past American accomplishments. The United States did ultimately respond to Japan’s 5th Generation Computing Project, but not in any “national mission” type way.

Second, criticizing Gunkel for making an “unrealistic” proposal in 1983 creates a post-hoc fallacy.

Not only does it fail to recognize what might have materialized as a result of a “Manhattan Project in AI” in 1983, but it presupposes the present path we’ve taken to LLMs is the only possible path to artificial intelligence, which probably isn’t true.

A sub-criticism in this vein is that it presupposes we even understand how LLMs work enough to know how we got here.

Many machine-learning researchers were surprised the transformer architecture breakthrough of 2017 ended up being so successful and there are still a number of open questions about how the technology works.

In light of everything we still don’t understand about LLMs, we may have very well taken a different route to AI and found ourselves equally confused—or perhaps, having arrived there earlier, would have sorted out the mechanism behind the breakthrough by now.

I’m sadly not well-versed enough in the history of artificial intelligence to provide enough context to explain this document in any detail.

And as always, there’s the problem of my very limited resources and time.

What I can say is that—in trying to prepare for a conference to be held at MIT and thus give himself a funding lifeline—Gunkel was putting an enormous amount of thought into artificial intelligence.

The document below provides 24 points and a number of additional sub-points—in the form of a list, of course—that runs the gamut from good reasons to go all-in on a “Manhattan Project for AI” to how the project should be structured to what the project itself should be called.

One interesting subtext here is that Gunkel was friendly with a number of Manhattan Project alumni. He even refers to physicist Philip Morrison in one of the list items; Morrison was a key Manhattan Project alumni who studied under J. Robert Oppenheimer and was crucial to the atom bomb’s development in several areas.

So the concept of a “Manhattan Project” may have come together in Gunkel’s mind both a way of solving an urgent financial problem as well as an opportunity to elevate himself to the stature of the brilliant men he often tried to rub elbows with.

Like the U.S. government posting captured Iraqi documents related to its nuclear program in March 2006, hoping to leverage this new idea called “crowdsourcing” in order to find smoking guns, I’m just going to put this out here and hope that someone else is able to find something useful.

However, unlike the “Operation Iraqi Freedom Document Portal,” which shut down nine months after going online because it was determined that the trove basically contained a road map to building an atomic bomb (never saw that one coming, right?), I’m relatively certain there isn’t anything here that will put the world at even more risk than it already is.

Note: I have the right to disseminate the below material. It may not be copied, stored, reproduced, or disseminated without express written permission. However, excerpts can and should be used for scholarly purposes.

And the below was an unnumbered addendum to the memo:

1) At one point in the memo, Gunkel refers to a “4 July 1984” Newsweek article. This is a typo. I verified that the mentioned article appeared on July 4, 1983.