In the wake of the demonetization fiasco on YouTube and the erosion of due process due to unrestrained “wrong-think” detection algorithms, one has to wonder whether these algorithms are being exploited by malicious users intent on destroying their competition.
Google supposedly is credited with effective algorithms preventing click fraud, but if false positives are necessary collateral damage in their attempts to minimize fraud, can one mimic click fraud to frame their competitors?
For example, if a botnet is created to hammer a search term, clicking through to a specific site as to appear like a fraudulent attempt to increase its search ranking, then Google’s algorithms are likely to detect this abnormal pattern and consequently downrank the page as punishment. However, if the website owner was not the perpetrator, how does he go about proving his innocence? In fact, he will have no idea that his site was targeted in that fashion, apart from the mountain of hits from analytics that makes it look like a DDOS attack, and then the subsequent reduction of search clickthroughs after the downrank.
You see this phenomenon panning out in the social media arena between radical identitarian groups and their “opponents”. As an alternative to doxxing, they will sometimes impersonate to get accounts suspended, demonetized, etc. And in the current state of over-reliance on AI to make hasty judgments, this appears to be an effective strategy as the falsely accused victims have a tough time defending themselves, having to suffer in both the short and long run. The victims are left spending time and resources defending their innocence to a “committee” under the presumption of guilt.
Obviously this is a complicated problem that does not have a readily programmatic solution, but it does serve as a warning when tech companies treat false positives too lightly, using algorithms to make guilty verdicts without due process.
* * *
For more tech commentary, be sure to subscribe to the RSS feed. Don’t forget to leave a comment, share and bookmark.