,
7 tweets,
2 min read
Read on Twitter
YouTube’s automated fact-checking tool probably made things worse by misclassifying the #NotreDame tragedy as a 9/11 conspiracy. Humans may have built cathedrals but they haven’t yet built machines that can understand when cathedrals are burning down. washingtonpost.com/technology/201…
We struggle to imagine how machine learning machines think in latent space, high dimensions. But this incident shows us a glimpse into an AI. The machine found a pattern to predict there would be an irrational, emotional association between the Notre Dame fire and 9/11 WTC.
The YouTube fact checker algorithm recognized signals in high-dimensions to flag the topic as a 9/11 conspiracy theory video flagged for a corrective. But there’s a kernel to the mis-classification that splits informational truth to emotional consensus—it feels like 9/11 somehow.
We don’t know exactly what triggered the misclassification but it’s likely that toxic comments contributed to the automated fact check, which in turn may have suggested and validated the conspiracism to wider audiences as fact check labels breaking news. A weird backfire effect.
YouTube’s algo is way too immature for prime-time and this botch job shows it. But also shows us that we need to get ready for machines that seem to understand the uncanny, irrational, emotional space of the human condition. We can’t mistake that for fact just because it’s tech.
Facts may well prove out that initial reports are confirmed: Fire was accidental. A better auto-response would’ve probably been a message urging viewers to wait for officials to report verified news and resist temptation to make and share conspiracy theories about breaking news.
The fact that this dumpster fire stuff happens means YouTube apparently is not spending its largesse on employees who are paid to go on their site to find broken stuff streaming into the cerebral cortex of billions. Reporters find this stuff so easily. Shooting fish in a barrel.
