How AI made workers more productive, yet less satisfied
82% of workers affected by the introduction of a new, “creative” AI system reported less satisfaction with their job, while the company simultaneously reaped numerous benefits
Imagine being a scientist working on discovering new materials that can lead to improvements in health care, and you’re doing better than ever: more findings are leading to an increase in patent filings as well as number of product prototypes.
Yet when you close down the computer for the day you feel less satisfied with what you’ve accomplished than usual.
Your employer has recently introduced a new AI system, which to a large extent has automated the idea generation process: part of your job is now to give thumbs-up or down to the proposals from your new, highly-effective colleague. You feel underutilized.
That is the case in an American company, which is the subject of a study by Aidan Toner-Rodgers, PhD in economics from MIT, where he investigates the effect of AI on innovation.
The overall result, based on a roll-out among 1,018 scientists in the R&D department of the US firm working within healthcare, optics, and industrial manufacturing:
1) The scientists had different levels of utility for the tool. From an increase in productivity of ~15% for the bottom third performing researchers before the roll-out, to 81% better output for a smaller group of top performers. The difference is primarily based on variation in their judgment. More on that later.
2) While most of the researchers reported an increase in job satisfaction based on productivity alone, the change of tasks with less time for idea generation, and more on judgment and experimentation, led to less satisfaction across the board. Summing up, 82% reported a decline in satisfaction.
Disproportionate benefits
The first point shows a different picture studies from other fields, where the lower-performing employees reap the most benefits to “catch up” with the higher-performing. Toner-Rodgers suggests that the judgment element in research makes the difference.
“I think in a lot of the settings where we’ve seen productivity compression, this step is just not there at all, and you can kind of out-of-the-box use the AI suggestion,” he says in an interview with The Atlantic.
He think it apples in general: if not much creativity is involved in the job, the lower-skilled employees will benefit more, and when it is crucial, the tool will enhance the output of the most-skilled disproportionately.
The declining job satisfaction makes sense in the context of Daniel Pink’s 2009 bestseller Drive, which points out that autonomy is a key element in motivation: in this case it’s undermined by the AI deciding what the human should evaluate.
Firings and hirings
The increased value of judgment skills — the ability to assess which of the AI’s suggestions to continue with for experimentents — apparently made the company reorganize their research efforts towards the end of the study.
Here, 3% of its researchers were fired. These were offset by new hires, so the workforce expanded in total, but of the scientists laid off, there was a clear overrepresentation of those with weaker judgment.
According to Toner-Rodgers, we can talk of it as a new research skill: judging model suggestions.
Increasing novelty with AI
In the study, Toner-Rodgers also concludes, that the average novelty of discoveries increases with the use of AI, but we don’t know if that also includes truly revolutionary findings.
“So one thing you might be worried about is, ‘Well, we got, on average, more novel things, but maybe these very revolutionary discoveries have a lower probability of being discovered by the AI, and that in the long term this is not a good trade-off.’”
Maybe Einstein was right by saying that “imagination is more important than knowledge”. On the other hand, you have AlphaGo, the AI specialized in the Chinese board game Go, which in 2016 stunned the world champion Lee Sedol when it made a move that “no human ever would”.
However, the verdict is still out, Toner-Rodgers emphasizes.
“We’ll need some time to see, if these new materials open up new avenues for research. Are there other materials that are going to be built on these new ideas that the AI generated?”
An important point in the study, though, is that the AI does not work without proper human assistance: only researchers with sufficient expertise could harness its powers.
“The fact that the tool is ineffective without skilled scientists suggests that demand for human researchers may remain strong, even as AI transforms scientific discovery,” Toner-Rodgers writes in the paper.
Altman: I think we’ll adapt
In a recent interview, Sam Altman, CEO of OpenAI, is confronted with the result from Toner-Rodgers’ study that human scientists were reduced to judges as opposed to creators or inventors.
“I have certainly gotten the greatest professional joy from having to really creatively reason through a problem and figure out an answer that no one's figured out before. And when I think about AI taking that over, if it happens that way, I do feel some sadness,” Altman says on the ReThinking Podcast.
“What I expect to happen in reality is just there's gonna be a new way we work on the hard problems. It's being an active participant in solving the hardest problems that brings the joy. And if we do that with new tools that augment us in a different way, I kind of think we'll adapt, but I'm uncertain.”
Expanding horizons?
In the documentary about AlphaGo and its match versus Lee Sedol, Demis Hassabis from Google DeepMind, the AI system’s creator, reportedly quotes a top professional Go player on the implications of the game before the result was known.
“He said, you know, ‘If AlphaGo wins, maybe we'll really start to get to see what this game is about.’”
The players who lost to AlphaGo allegedly learned from the experience and improved their game, but for Lee Sedol, the defeat was also the beginning of the end. In 2019 he retired, 36 years old, due to increasing AI dominance in the game. In a recent interview with The New York Times he reflects on the implications of the advanced technology.
“Losing to A.I., in a sense, meant my entire world was collapsing,” he said. “I could no longer enjoy the game. So I retired.”
In Isaac Asimov’s 1950 science fiction short story, The Evitable Conflict, the world is run by four, omniscient-like machines that sometimes make decisions that humanity may not like, but which they know is best for the people.
“To know that may make us unhappy and hurt our pride,” Susan Calvin, the robopsychologist in the story says, then referring to Asimov’s first Law of robotics:
“The Machine cannot, must not, make us unhappy.”