Today I have some very special archival material to share.
It’s an original ideonomy experiment that comes from the MIT Distinctive Collections archive started by Professor Betsey Dyer, and it was one of the first communications the two of them had after their meeting at the famous Philamorphs forum on the Harvard University campus in 1984.
Gunkel, living on Homer Avenue in Cambridge, was halfway through the first year of his five-year grant to create the science of ideonomy.
The grant was being paid out through the Hudson Institute, which is a fact I discovered in my original research.
For the purposes of the experiment, Dyer—then a graduate student at the lab of Lynn Margulis at Boston University—provided Gunkel with three questions related to the field of biology. They were:
Do centriolar structures have an origin as independent genetic entities?
How do organisms control their own variability?
In what way do bacteria control the cycling of gold in the environment?
In response to these questions, Gunkel had his computer generate 22 sentences using the principle of idea chemistry (discussed in the last issue of the newsletter), almost like fill-in-the-blank madlibs, that were intended to inspire potential answers about these topics based on an existing list of possible answers.
The technique Gunkel is demonstrating here is known as “automated hypothesis generation.”
The concept of automated hypothesis generation was not new when Gunkel began his Ideonomy Project.
But I’ve done some basic research, and at a first glance it doesn’t seem like this particular method of automated fill-in-the-blank, modular hypothesis generation has ever been documented in the literature.
For example, around the same time Gunkel was doing this, other computer scientists at MIT were doing a different variety of automated hypothesis generation on data sets and bibliographic references in literature. That’s not what Gunkel is doing here.
Rather, Gunkel is populating template sentences with Socratic-style insights about the ideas being considered.
For example, here’s one of the answers Gunkel had his computer generate in response to the question about bacteria and gold:
Re: the question as to in what way bacteria control the cycling of gold in the environment: consider the genus of ignorance of/re best models of/re (the) things.
Take the best model of the bacteria-gold cycle, in other words, and investigate what parts of that cycle you are ignorant about.
Pretty straight forward, right?
Now imagine this repeated dozens, or even hundreds, of times, so that such prompts for further investigation can be generated at scale.
Now imagine automating the process of answer generation and analysis using artificial intelligence.
A number of such answers might, at the very least, have the potential to lead to a new research paper; even better, one such answer might provide the inspiration needed for a breakthrough.
Although Gunkel seems to have developed the process over 40 years ago, I don’t think anyone knows about or uses this version of automated hypothesis generation today. (This is yet one more area of research I am not able to do at the moment, however.)
What the Experiment Shows
Beyond the essential viability of the central principle, the experiment and the letters written between Gunkel and Dyer put a few other interesting things on display.
First, note that although Gunkel did not generate answers systematically for each of the three questions, these answers nevertheless exist in the “latent space” of this experiment, meaning that the same answer about the bacteria-gold relationship (“consider the genius of ignorance of/re best models of/re (the) things”) could be provided for the other two questions about centriolar structures and the variability of organisms. Therefore, while Gunkel provided 22 possible answers for Dyer’s 3 questions, in reality the set of possible answers is much higher if each possible answer is systematically matched against each of the questions.
In aggregate, this demonstrates that patterns of ideas underlying thought can be used to predict other possible thoughts. This, I argue, combined with the use of computerized answer generation, is doing science. The only thing that’s missing—the key link from my perspective—is an acknowledgement of the theoretical environment we are operating with in, i.e., that we are striving toward developing a “theory of ideas” that helps explain these patterns. This is what Gunkel was trying to get at when he said ideonomy was about “the laws of ideas,” except the problem was that he never actually tried to show what those laws were.
Reading through the results of Gunkel’s experiment, they’re going to at first glance seem a lot like gibberish.
But we need to pause a moment here and push through what social psychologist Daniel Kahneman has referred to as System I—that lazy, heuristic-loving part of our brain that isn’t used to receiving new kinds or formats of information and simply tosses it aside.
We need to instead engage with System II—the process of “thinking slow” that Kahneman popularized, in a book that continues to remain on the best-seller charts after many years—and spend some time thinking about what Gunkel is showing us.
What we also need to acknowledge is that Gunkel wasn’t inventing a method for humans.
He was inventing a method for future artificial intelligence systems.
He was developing a process for future systems to systematically scan through “all possible answers” that can be given about a particular idea, searching for weak points where nobody had provided an answer yet. This would represent a point where a breakthrough could emerge.
Other Things of Note
Although Gunkel didn’t get anywhere close to providing all possible answers about Dyer’s three questions, his experiment did provide enough detail to grasp the methodology.
And, if my instincts are correct, this letter and Gunkel’s ideonomy work more broadly could help rewrite history by marking Gunkel as a pioneer in the field of automated hypothesis generation.
“It’s high intensity play. It’s right on the edge. It’s a little scary,” Dyer wrote to Gunkel in response to the answers he sent her. “I’m glad you’re doing this stuff, otherwise it might not get done for another thousand years.”
Dyer was not a computer scientist, but her remarks to Gunkel help contextualize how novel and important his work seemed to many well-educated people at the time—Dyer included.
Among other conclusions that can be drawn from this initial exchange between the world’s first ideonomist and a graduate student in biology, I’m also struck by the patience, care, and consideration Dyer gave to Gunkel’s list.
She went down all 22 questions and tried to answer them—every one.
Would you do the same?
We should also note the way Gunkel labeled his response to Dyer.
It’s a bit cut off in the image below, but in the upper-right corner of the photo we can see “1984 Aug 7 - applied ideonomy - ideocombinatorics - biology.”
This was Gunkel’s way of organizing his work with respect to the different sections that he imagined for his book.
And yet despite the designation, this material did not end up among the ideonomy papers posted to the MIT website.
This confirms visually the thing that I have sufficient evidence of by this point, which is that only a portion of Gunkel’s ideonomy materials were ever posted.
Who knows how many of Gunkel’s experiments and writings existed only as correspondence to other people—correspondence that is likely now gone.
Rewriting History?
On page 5 of Gunkel’s progress letter to Alan McHenry, President of the Lousnbery Foundation, which appears at the beginning of Gunkel’s Orange ideonomy Volume, Gunkel wrote that he “fell in love under inauspicious circumstances” at the beginning of the third year of his project.
This is a reference to his infatuation with Dyer.
Once this distraction was overcome, however, and Gunkel moved to Austin, Texas, the work on his magnum opus truly took off in earnest—half a nation away from “loathsome Cambridge.”
In subsequent years, Dyer ended up being an important support and motivational interest for Gunkel throughout the Ideonomy Project.
Some of the documents that worked their way into the ideonomy manuscript posted piecemeal on MIT’s ideonomy website were either inspired by, involved the assistance of, or made reference to Dyer.1
It probably goes without saying, but none of the documents I share in my newsletter—and none of Gunkel’s writings, generally—have ever been appropriately analyzed in an academic context.
One of my main ambitions with my research is to establish enough room for others to come in and do that work.
I already mentioned the potential to write the history of automated hypothesis generation and provide something very useful in the age of artificial intelligence.
But beyond this, could anything Gunkel and Dyer discussed in this correspondence rewrite the history of the relationship between gold and bacteria?
Are there ideas here that were unworkable at the time, but some new advance in the current context makes these ideas suddenly viable?
If any of Dyer’s questions are still relevant, would Gunkel’s methodology work within the context of modern computational techniques to continue to attack these problems?
A new discovery coming from old archival materials like these might not be probable, but it’s certainly possible.
So if this seems like a fun space to dive into, then please come on in… the water’s fine.
Gunkel’s Original Letter:
Dyer’s Response:
Note: I have the right to disseminate the above material. It may not be copied, stored, reproduced, or disseminated without express written permission. However, excerpts can and should be used for scholarly purposes.
(1) See, e.g., “Notes on a Wonderful Group of Interrelated Ideas Developed in Discussion with Betsey Dyer on 1985 May 4, Dealing Generally with Possibilities for Quasi-Biological Evolution and Sophistication in the Purely Chemical Realm,” on electronic page 216 of the Pastel Green document.