For decades, the internet has steadily eroded the power of gatekeepers and democratised access to information. Attempts to resist that are seen as censorship or elitism, neither of which anyone wants to be accused of. But our new AI era may flip the script.
Social media gives anyone a voice, breaking the old model of one-to-many broadcasting from institutions, but no one is forced to listen to this cacophony. Large language models that extrude text at speed are now changing this, with institutions under a many-to-one assault by information they must attend to. This shows up in many forms – indeed, including in literal forms, as people use AI in their bureaucratic interactions with courts, schools, local officials and more.
These systems can’t deal with the volume of information and are clogging up. The implicit gatekeepers that kept such submissions under control – time, ability, access – have fallen to AI. The same thing is happening in science.
Mathematicians are calling for more mathematicians in the face of an AI onslaught. Meanwhile, the preprint server arXiv has seen submissions double in the past two years. Although preprints aren’t formally peer-reviewed, arXiv’s small team ensures minimum standards, and it is overwhelmed.
In response, it recently announced it would limit submissions to two per month – still remarkably high, with few scientists producing 24 papers a year of any quality. A kind of gatekeeping is the only way institutions can survive the AI era This kind of gatekeeping is the only way institutions can survive the AI era, unless we want to fully hand over the reins to AI and let it handle inputs as well as outputs. But there are good reasons not to.
As we report, tools for verifying mathematical conclusions are at risk from attack by AI, so keeping a human in the loop remains the only way for us to have a handle on the truth. That goes even more so for areas that are harder to verify than mathematics, which is essentially everything. So let’s cheer on the gatekeepers – they are going to have a lot of work to do.
Extract — continue reading at the source.
