That’s one way to talk about other human beings.
As writer Elizabeth Weil notes in a new profile of OpenAI CEO Sam Altman in New York Magazine, the powerful AI executive has a disconcerting penchant for using the term “median human,” a phrase that seemingly equates to a robotic tech bro version of “Average Joe.”
Altman’s hope is that artificial general intelligence (AGI) will have roughly the same intelligence as a “median human that you could hire as a co-worker.”
It’s a disconcerting assertion, considering that it really sounds like Altman is looking to replace the work of normal people with a not-yet-realized AGI.
And according to Insider, it’s not even the first time he’s said as much. In a 2022 interview on the Lex Fridman podcast, Altman explained that this theoretical AI would be able to “do anything that you’d be happy with a remote coworker doing just behind a computer, which includes learning how to go be a doctor, learning how to go be a very competent coder.”
In other words, if you happen to live a “median” life, you could soon be out of a job — or, at least, that’s one way to interpret his comments. Cheers to our AI future.
As Insider and New York Mag both note, Altman isn’t the only person in the world of AI who uses the unsettling terminology. The phrase is present across an incredible number of AI blogs, and was even featured in a CNBC article titled “How to talk about AI like an insider.”
“Eventually, when we develop an AGI in earnest,” reads a blog post from an AI startup dubbed Snippet, which seems to fall very much in line with Altman’s median theory, “it would display the capabilities of the median human, but retain the potential to become an expert in the field, something we now consider reserved for the narrow AI.”
To use the word “median” specifically also feels like a distinct — and telling — choice. It’s a squishy term that could leave plenty up to interpretation. How Altman, or anyone else, could possibly go about determining a holistic definition for this statistical average is unclear. Regardless, such a quantification of the human experience feels in many ways dehumanizing and incomplete.
“Comparing AI to even the idea of median or average humans is a bit offensive,” Brent Mittelstadt, director of research at the Oxford Internet Institute, told Insider. “I see the comparison as being concerning and see the terminology as being concerning too.”
Adding that there’s yet to be a “concrete measurable comparison of human intelligence” within AI research, Mittelstadt also noted that the concept of a median person seems like “an intentionally vague concept as compared to having a very specific grounded meaning.”
Besides, the concept of performance and the much-less-tangible notion of human intelligence are two very different things — and as Middlestadt told Insider, equating them doesn’t quite add up.
“That is a hugely problematic leap to make,” said the Oxford researcher, “because all of a sudden you’re assigning agency, comprehension, cognition, or reasoning to these mechanistic models.”
Middlestadt isn’t alone in his critique.
“One thing that current AI architectures and models have shown is that they can achieve basically typical human-level performance. That’s not problematic in itself,” Henry Shevlin, an AI ethicist and professor at the University of Cambridge, told Insider. “I feel when we get into things like intelligence people are more touchy, and there are some good reasons for that.”