Side effects

From Tournesol
Jump to navigation Jump to search

ElmhamdiHoang-19FR argues that AI risks mostly revolve around unforeseen, unwanted and large-scale side effects. In other words, their main concern is that, as an algorithm optimizes for a seemingly safe goal, it will develop strategies with dangerous externalities. This is akin to an oil company whose profit maximization indirectly causes biodiversity loss, resource depletion and climate change.

In particular, leading side effects of attention maximization include misinformation, radicalization, manipulation, cyberbullying, hate, addiction, shortened attention span, loneliness, depression, suicides, discrimination, privacy loss and increased existential risks.


One major challenge of mitigating side effects is their unpredictability, especially as algorithms interact with highly complex environments like social medias.

As an example, in 2017, Facebook's CEO Mark Zuckerberg decided to increase the recommendations of Facebook groups' contents Forbes-17. However, JohnsonVJLG+20 reports that many such groups host an increasingly large amount of anti-vaccination propaganda.

BailaABBC+18 report an experiment where subtly increasing exposure to opposing views on Twitter led to increased polarization.