It’s because of the way the conjecture is structured. It proposes that we believe a multi billion dollar corporation would risk their profit margins on the tastes of one engineer in the shadows.
Occam's razor would say that such a large corporation does not have the direct controls to monitor every single actor. Is it so hard to find actors acting against the interest of their employer? I've even personally experience cases where the actors are rewarded for such actions due to flaws in the incentive structure.
Saying AI does funky things from time-to-time is hardly an extraordinary claim.
Meanwhile, an individual injecting bias without the hack's true purpose being obvious to co-workers would be difficult. The more subtle and circumstantial the hack, the more likely it will be broken/ignored by someone else's changes.
A hack is still possible, but I'd want evidence of intentional bias before entertaining the claim.
Because most people don't work in silos that allow them to operate with impunity. I'd assume at a company the size of google any feature is going to have minimum of 3-4 different sets of eyes on it between Engineering, QA, UX, and Product.
