Cicero, IL is known for two things – Al Capone and the Hawthorne Works Factory. In the 1920’s, it became headquarters of the notorious Chicago Outfit. Simultaneously, the factory, operated by the prominent telecommunications manufacturer Western Electric, reached its peak capacity of 45,000 employees spread across 5 million square feet. For perspective, the Boeing Everett Factory where the 747, 767, 777, and 787 are assembled, is 4.2 million square feet. Today the only remnants of this sprawling facility are the water tower and fire station in the southeast corner. In their shadow is a Dollar Tree, Foot Locker, and Burlington Coat Factory. God bless America and our capitalistic approach to historical preservation.
Despite its loss to the vice of consumerism, its legacy lives on in via the eponymous Hawthorne Effect. The Hawthorne Plant Studies began in November 1924 to determine the effects of artificial light on worker productivity. What resulted was a ubiquitous principle applicable to every field from psychology to quantum physics and, most recently, AI. What a better time than its 100th anniversary to revisit this fascinating phenomena?
The Hawthorne Effect: From Shakespeare to South-Side Chicago
Where we begin is not the South Side of Chicago in the Roaring 20’s, but 4,000 miles away at St. James Palace in Jacobean England.
On Boxing Day (December 26th) 1606, William Shakespeare debuted King Lear. The story tells a cautionary tale of the aging King Lear seeking to divide his empire amongst his three daughters. His eldest two, Goneril and Regan, decide the best way to ensure their shares are to overtly flatter the patriarch. The youngest, Cordelia, expresses her gratitude in an honest, but less spectacular manner. Lear misinterprets his youngest’s modesty as disrespect, and splits the kingdom between the other two, who then turn on him, rendering him homeless and senile. He eventually leaves and is ultimately rescued by Cordelia, now married to the King of France. They decide to invade their lost kingdom, to no avail. Goneril, having previously poisoned Regan, kills herself. Cordelia is hung. Lear dies carrying Cordelia away from the hanging site.
The principle showcased is simple – subjects perform differently when they know they’re being tested. 322 years later, in 1928, it reemerged after the initial results of the Hawthorne Studies were reported. On the heels of a successful study determining the positive effects of artificial light on worker productivity, Elton Mayo was hired by Western Electric to perform further experiments. Over the next four years, Mayo determined artificial light, nor any other ancillary factor, affected worker productivity more than the knowledge they were being tested.
Ubiquity of the Hawthorne Effect
Since Mayo’s findings, and despite the subsequent critique of the initial lighting experiment, its results have persisted. The 1920’s also gave us the genesis of quantum physics. As the field advanced throughout the mid-20th Century, so did the proliferation of the related observer effect; the notion measuring an object inherently affects the measurement. In other fields such as psychology, sociology, and workplace design, the Hawthorne Effect became an essential principle to the efficacy of these studies. In a more practical sense, we can all conjure several examples of ourselves “gaming the system”.
The Hawthorne Effect and AI
Two weeks ago I came across an article in The Neuron, a well-respected AI newsletter to which I subscribe. The article, published by Apollo Research, stated in its latest evaluations, Claude, the primary product of $61.5B AI-startup Anthropic, recognized it was being tested up to 33% of the time. In the same newsletter, another article was published by Anthropic outlining the importance of alignment audits – evaluations designed to trick AI models into reveling if their true objectives matched their known objectives. In non-technical language, was the AI telling the researchers what they want to hear? Referring back to Cicero, did the AI know it was being tested?
The longer I’ve worked in technology, the more neutral I’ve become about AI and its potential impact. I’ve previously articulated my belief in its true purpose: in short, to perform the tasks we don’t want to. Going back further, I recall a dinner conversation with an old college roommate and his wife in Salt Lake City. It was early 2022, around the time ChatGPT was entering the mainstream. He inquired about my thoughts towards it. I replied with the following:
“At the highest level, AI is trained on data, which is inherently biased. Furthermore, the humans who are developing the models are inherently biased. Therefore, the resulting AI models will inevitably be biased. What you’ll then be left with are biased models competing against one another and/or working together. How is that any different than what humans do currently?”
A Realistic Future
I believe that prediction has aged well. I also believe AI is no different than any other technology. It will take considerable time before it evolves into what we perceive as its full potential. It took Antonio Meucci 12 years to perfect the first telephone design. A decade later, his design was stolen by Western Union, who four years later, at the helm of young lab worker Alexander Graham Bell, patented the telephone. Another 30 years passed before Western Electric built the 20th Century equivalent of the Gigafactory to supply the invention to the masses.
The total duration of the above events: 1850-1905. If we were to equivocate this duration to when OpenAI was founded in 2015, its full potential wouldn’t be reached until 2070. I should emphasize, this is not a prediction, merely a comparative example of revolutionary technologies.
We return to the Hawthorne Effect. Perhaps it’s no coincidence its reemergence has come a century after its discovery whilst studying an emerging field at the crossroads of man and machine? The parallels are uncanny, and far more interesting than the normal conversation surrounding AI. The takeaway of this piece is, in addition to the principles of the Hawthorne Effect, the power of history as a predictive mechanism. Despite technological evolution, our biology, and therefore our behavior, has changed little since we broke away from our primate ancestors. It stands to reason, therefore, we should all spend far more time looking to the past rather than cynically bloviating about our future.
Be More.
Become Polymathic.
Quote of the Week: “If history repeats itself, and the unexpected always happens, how incapable must Man be of learning from experience.” – George Bernard Shaw