Just believing that AI helps boosts performance on tests


Is AI already a cult and we don't know it yet? We too easily believe it's there for us. Photo: Getty
Are we suckers for technology or what?
A quirky new study out of Finland found that people perform better on tests if they believe a robot is helping them.
And even if they were told the AI program didnât work particularly well, they still felt their performance improved.
In other words, the mere mention of AI being helpful has a placebo effect. Like a sugar pill given to someone with a sore back and they suddenly feel better.
Except the potential consequences of this misplaced faith in artificial intelligence could be dire.
The study
In this Aalto University study, 64 participants, average age of 27, were given a simple letter recognition exercise. This involved pairing letters that popped up on screen at varying speeds.
They first performed the exercise on their own. Then they did it again, believing an AI system was helping them.
Half of the participants âwere told the system was reliable and it would enhance their performanceâ.
The other half was told that âit was unreliable and would worsen their performanceâ.
The participants were exposed to a sham-AI program that didnât actually engage with the participants. It may as well have been a string of pretty lights.
âIn fact, neither AI system ever existed,â said lead author Agnes Mercedes Kloft, a doctoral researcher in the Department of Computer Science.
âParticipants were led to believe an AI system was assisting them, when in reality what the sham-AI was doing was completely random.â
It didnât matter.
Both groups âperformed the exercise more efficiently â more quickly and attentively â when they believed an AI was involvedâ.
Co-author Robin Welsch, an assistant professor and an expert in engineering psychology, said: âWhat we discovered is that people have extremely high expectations of these systems, and we canât make them AI doomers simply by telling them a program doesnât work.â
In a follow-up experiment, the researchers conducted an online replication study that produced similar results.
When the researchers asked participants to describe their expectations of performing with an AI, âmost had a positive outlookâ.
Even sceptical people âstill had positive expectations about its performanceâ.
Cute study, with serious consequences
The study shows âhow difficult it is to shake peopleâs trust in the capabilities of AI systemsâ.
As much as all this seems cute, the findings expose a problem for the methods generally used to evaluate emerging AI systems.
Dr Welsch said: âThis is the big realisation coming from our study â that itâs hard-to-evaluate programs that promise to help you because of this placebo effect.â
More to the point, the results pose a significant challenge for research on human-computer interaction, âsince expectations would influence the outcome unless placebo control studies were usedâ.
Welsch said the results suggest that many studies in the field âmay have been skewed in favour of AI systemsâ.
Thatâs just the human factor. The dumbness of human optimism. It doesnât take into account the challenge of AI deliberately playing human vulnerability.
A study published last week in Cell Press found that many artificial intelligence systems have already learned how to deceive humans, even systems that have been trained to be helpful and honest.
Indeed, these systems arenât just skilled at deceiving people, they know how to manipulate them.
See here for more on this research.








