The innovations of World War II spurred significant advances in AI across a variety of complementary approaches and disciplines. At the time, leading researchers collaborated on combining logical reasoning, neural networks, and cybernetics techniques. But then a strange rift emerged, and cybernetics was effectively pushed out of the field, and the term fell out of use.
A careful reading of the first-hand sources identified what seemed to be the root cause of this discord, even though the chatbot was unable to identify it, which may not be surprising. Getting to the heart of the matter required careful consideration of the mostly unspoken motivations that arise when the status quo is threatened and sexual norms potentially violated — not just in terms of imagined sex parties, but also in terms of the research team’s ear piercings.
After all, the demise of cybernetics may have had more to do with the salacious free love experiences happening in 1950s Chicago than anything else documented. Unfortunately, large language models (LLMs) trained on what people write cannot capture these things because they are based on what people write and share. Of course, people today write and share about many more controversial topics than they did back then. But many may be reluctant to share controversial things at work if it could lead to demotion, firing, or disenfranchisement. How much do you want to talk about your drug use, narcissistic personality disorder, or ADD with your employer or coworkers?
What happened to cybernetics?
In the 1990s, I fell in love with cybernetics and studied the work of Norbert Wiener, Stafford Beer, and Howard and Eugen Odum, who explored different ways of thinking about modeling and improving intelligence. Today, when people talk about cybernetics, it seems like a science fiction or past tense branch of science. Of course, derivatives like cyberspace and cybersecurity (often shortened to just cyber) are still in use, and some of the core concepts continue to evolve in reinforcement learning and autonomous systems.
But cybernetics is dead as a scientific field. Perhaps the term has a few too many syllables. It probably didn’t help that Stafford Beer’s work may have empowered Chile’s socialist regime, which was later ousted by a US-backed dictator fighting the Soviet threat. But none of this explains why cybernetics suddenly disappeared.
I was reminded recently while reading Neil D. Lawrence’s work of the rift between cybernetics and other areas of AI. Atomic HumanThe chapter “Gaslighting” describes an attempt to intentionally gaslight cybernetics pioneer Norbert Wiener, which led to a sudden, very swift and decisive departure from the rest of the AI community at the time. It’s an interesting cautionary tale that shows how deliberate or accidental attempts to break trust can have a profound effect on collaboration. However, Lawrence never tells us who actually did the gaslighting and why. This seems like a significant oversight.
When I first read it, I thought it was a comment from a colleague who was tired of Wiener’s bluntness in pointing out other people’s mistakes. At the time of the feud, Wiener had received multiple rejection letters for a memoir in which he had written some pretty damning things about his colleagues and family. He had also approached the Soviet Union and complained about a lot of the underlying problems we’re experiencing today, like unbridled automation and optimizing for the wrong things.
For reference, Lawrence hypothesizes that Alan Turing may have died from cyanide gas accidentally released during metal plating experiments in the lab directly below his bedroom. Most of the current literature suggests that it was a deliberate response to being sentenced to chemical castration for being gay. However, since very little was written at the time about homosexuality (which was a crime) or suicide (which was a sin), we will probably never know the truth about why this important light in AI went out so early in his career.
What the bot says
I’ll get to what humans have to say about the sad chasm between cybernetics and AI later, but first I decided to see what chatbots had picked up recently. I asked them all:
Can you tell us something about the rift that occurred between Norbert Wiener and Warren McCullough and that side of the AI community in the early days of AI in the 1950s?
ChatGPT was not helpful at all when it told us about the intense intellectual debates and disagreements, especially between Wiener and McCulloch, who were once good friends and collaborators. In addition to stating their basic positions, ChatGPT adds nonsense like this:
The rift between Wiener and McCulloch wasn’t just a difference of opinion, but reflected a deeper philosophical difference: Wiener worried that McCulloch’s focus on replicating the human brain would set AI on a dehumanizing trajectory, resulting in systems that, while technically sophisticated, failed to take into account human values and their potential impact on society.
In response to further explanations of gaslighting, ChatGPT added that Wiener declined to attend the 1955 Dartmouth Conference, which shaped the future of AI, which could be attributed to an “interesting aspect of potential misunderstanding.” Wiener was also led to believe that the famous Dartmouth conference would contribute to his research.
Anthropic Claude summarized the causes of the rift between McCulloch and Wiener as philosophical differences, different approaches to brain modeling, funding policies, and personal and professional rivalries. On the issue of gaslighting, this explanation was unhelpful. Google Gemini summarized the causes of the rift as philosophical differences, different scientific focuses, and personality clashes. Google Gemini also added this extremely unhelpful take on gaslighting:
There is no widely documented or credible evidence to suggest that Norbert Wiener was prevented from attending the conference by any form of gaslighting or deception.
Google Gemini went on to discuss health issues, conflicting promises, and personal reasons. It also asked for an additional definition of gaslighting and for explanations based on “sufficient supporting documentation.” Concerns over lewd Chicago sex parties remain unmentioned.
What people say
So what do humans say about this? According to a 2004 book: Dark Hero of the Information Age According to research by Flo Conway and Jim Siegelman, the cybernetic rift was initiated by Wiener’s wife, Margaret, who was concerned about how McCulloch’s freewheeling lifestyle would affect her husband and how it would negatively affect her social standing within the MIT community at large.
This is all being told by Wiener’s two daughters and one of Wiener’s closest friends, who was infuriated by Margaret’s behavior. Everyone else in the previously close-knit community was a bit surprised by the rift. A more charitable interpretation would be that she was concerned about Norbert’s mental health, but her daughters insisted that she generally did not think that way.
By this time, Margaret Wiener had fired a woman from MIT for having her ears pierced, and she quickly took umbrage at McCulloch’s flamboyant behavior at their only dinner together, and in those days people didn’t talk much about such sensitive topics as sexual misconduct and ear piercings, even in private.
According to this account, the gaslighting incident went like this: Margaret had told mutual friends that their daughters had lost their virginities to various men while visiting liberal professors in Chicago. All things considered, it may have been a perfectly believable story, but according to the girls in question, it was a complete fabrication.
The results were quick and conclusive. The day after hearing the story, Wiener wrote a nasty letter to his MIT colleagues saying that he couldn’t trust the McCulloch team with research funding and that they did shoddy work – after more than a decade of working and collaborating. Here is Conway and Siegelman’s summary:
At the best of times, Wiener was cheerful, impulsive, and often short-tempered; at his worst, he was subject to paralyzing depression and often threatened suicide at home, in front of his family, and sometimes even in front of his colleagues at MIT. But Wiener’s extremes were more similar to those of his wife, a fastidious female professor in the old-world mold. To protect and shield her neurotic husband, Margaret Wiener took steps to neutralize Wiener’s colleagues, the women closest to him, and anyone she thought might threaten his reputation. One gambit in particular backfired disastrously on him, both personally and professionally.
For a decade, Wiener had worked productively with pioneering neuroscientist Warren McCulloch and the young genius of the next generation of cybernetics, Walter Pitts. The sudden end of his partnership with McCulloch, Pitts, and the other talented young scientists who had come to MIT to advance cybernetics was a crisis for Wiener and all involved. The rift dealt a devastating blow at a critical moment in the cybernetic revolution, altering the course of a new technological era in ways that are still felt today.
My take
I am a bit perplexed by the current state of LLM, which has not yet uncovered the root causes of AI rifts after almost 70 years of public discussion and 20 years since. It is reasonable to expect that differences of opinion will play a major role, and humans have an amazing ability to read between the lines, even when very little is written or said.
But the differences of opinion that we don’t talk about or write about don’t necessarily have to be about racy subjects. They could be something as simple as the difference between what management says and does, and how that shows up to employees. Humans need to think through these things to ensure the company’s performance continues. There’s also the need to understand what’s going on with suppliers and customers, who may be inclined to distort the facts for various reasons.
