In about 1988, James J Keene sent detailed information to DARPA (Defense Advanced Research Projects Agency) on his invention of the "intron", perhaps the first AI algorithm based on Neural Network Learning. The story developed like something from a spy novel leading to Nobel prizes awarded to others and billion-dollar companies.
Computer programs can automate tasks to increase productivity and exhibit a sort of "intelligence". Further improvement requires a human programmer to modify the program code. This automation goes back to the very first computer programs. Most computer scientists agree that a major departure from this computer automation era is systems that can teach themselves to master a task, which may be designated "artificial intelligence" (AI).
Intron Research Program
Over some five or more years in the 1980s, I developed the "intron" invention that could learn to perform a task successfully based on real-world feedback on its actions. As far as I knew at the time, this intron project was the first neural network displaying learning.
The objective was to trade commodity futures contracts profitably. The daily real-world data (W) was the open, high, low and close prices of a number of commodity futures. Each day, this state of the real-world was presented as stimuli (S) to an array of introns, one for each commodity contract. Thus, each intron was exposed to all the data W for all the commodities monitored (bonds, metals, currencies, grains, etc). Each intron then issued no reponse or a response (R) to buy or sell its particular contract at market open or close. The success of responses R (profit or loss) was then determined from subsequent data W to provide feedback (F) which updated knowledge (K) in the responding intron (Fig. 1).
The human operator had just two tasks. First, obtain from the Wall Street Journal the daily data W and type it in. Second, after the intron array achieved net profit performance, the operator would place the buy or sell orders R to a commodities broker before market open or close.
In this project, a "training period" was operation of the system until consistent net profits where achieved. Since real-world market dynamics could change over time, training continued after the success criterion of net profits was achieved.
Intron was short for "intellectron" and presented as a physical device. That is, the intron could be implemented with software as described above or as a hardware module mounted on a circuit board which interfaced the "intron chip" with the device bus sockets on PC motherboards.
Intron Invention Sent to DARPA
In about 1988, the author sent detailed information to DARPA on his invention of the "intron", perhaps the first AI algorithm based on Neural Network Learning. This information included:
1. Description of the intron invention.In short, the package mailed to DARPA was not just an idea, not just what might be. It was a complete disclosure of the intron invention with a successful demonstration that anyone could grab and run with it. Many of the key documents contained copyright notice.
2. Design diagrams such as Figs. 1 and 2.
3. Software code to implement the intron in real-world tasks such as commodity futures contract trading.
4. Documentation that the intron could learn to perform tasks successfully, namely buy and sell commodity futures with consistent net profit.
In retrospect, it may have been a bad idea to send such a complete package to DARPA.
Initial DARPA Response
Men in Black. Several weeks later, two gentlemen knocked on my door in the beach-front Villas del Mar apartments in Isla Verde, Puerto Rico. I invited them in. They had questions. I was thinking, "They saw the movie. Both wore black suits and had crew-cut hair style." We talked about the 16-bit PDP-11 computer in the living room. They wanted to know more about me, where I worked. Maybe they wanted to see if there were posters of Marx, Lenin or Chairman Mao on the walls. No such posters. These "men in black" were polite and friendly. They confirmed that DARPA had received my package.
Recorded Phone Calls. A month or so later, my phone rang. I picked up, "Hello." The caller said, "I have sent over the recordings of Dr. Keene's phone calls." Pause. I replied, "Excuse me, this is Dr. Keene." Silent pause. Hangup. In this creepy episode, the caller had mistakenly dialed my number instead of the person to whom he was sending the recordings.
In addition to a second confirmation that my intron package was received, the recording of my phone calls suggests that DARPA deemed that the content of my intron invention was important. Apparently DARPA wanted to know who I knew and what I talked about. Why? Perhaps they wanted to know who else knew about the intron project. However, at the time, I shrugged this off as a comic mistake. I did not ask why DARPA might want to make me a character in a spy novel.
Fast Forward to 2025
The "men in black" and the comic caller came to mind when the Nobel Prize in physics was awarded to two AI researchers. Commentators immediately noted an unusual classification of computer science as part of physics. Was it possible that my intron project and DARPA was part of this story?
Juergen Schmidhuber
The question of the possible role of the intron project in AI developments requires a scholarly literature review. In an amazing coincidence, a look at my x.com feed recently included a long post with exactly that by Juergen Schmidhuber (@SchmidhuberAI), a German computer scientist noted for his AI work, specifically artificial neural networks.
His web page says, "In the late 1980s Schmidhuber developed the first credit-conserving reinforcement learning system based on market principles, and also the first neural one." Do you believe in coincidences? The intron project was demonstrated to work in the financial markets. Schmidhuber claims to have developed a "learning system based on market principles".
Did Schmidhuber have any link with anybody at DARPA? In his CV, he was "Postdoc at University of Colorado at Boulder (CU): July 1991 - December 1992. (Declined postdoc offer by Caltech.)" The "men in black" story above suggests that DARPA operatives can show up anywhere. Did other proverbial men-in-black look at Schmidhuber's profile as a computer guy who might follow up on the intron project information they had?
The present relevance of Schmidhuber's recent post (Fig. 3) is literature citations in which the first true AI papers appeared in 1991, assuming I understand this material correctly.
Homework For Internet Sleuths
How might internet detectives trace who handled the intron project at DARPA and to whom information about it may have been transmitted? Quite a homework assignment. I can only dimly see what readers might find. For example, does DARPA appear in the CVs of any of the authors of papers in the 1991 time frame in Schmidhuber's bibliography (Fig. 3)? I already checked for Schmidhuber. DARPA is not in his CV. Maybe DARPA links are not routinely included in CVs.
Hopefully, this has been a fun story. Is there any message beyond the question of whether the intron project was a "founding participant" in AI? Yes. The big questions are integrity, honesty and justice in how people treat each other.
© 2025 James J Keene
No comments:
Post a Comment