Was the Tilly Norwood interview purposely bad?

Tilly Norwood, the trying-to-make-fetch-happen AI “actress” spawned last year by AI production house Particle6, recently had a disastrous interview with British journalist Piers Morgan. The video is a total mess. Despite being a static shot, the real-time video looks terrible, with her features sliding around her face. Her voice is totally robotic, her words fumbling and repetitive, and she inexplicably starts speaking Cantonese partway through the interview. The whole thing was a shitshow, even by Piers Morgan standards.

Yes, this would all have seemed miraculous only a few years ago, but I really thought real-time video fakery had advanced beyond this. And it got me thinking: was this bad on purpose?

Tilly has struggled to break out beyond weird short-form videos. Now the whole world is talking about ROGUE AI STARTS TALKING CANTONESE. To be clear, I don’t think the Cantonese thing was specifically set up beforehand, but they certainly didn’t optimize for a smooth launch. Instead, they launched a media barrage of 75 simultaneous interviews, many of which were weird and awkward.

I am feeling increasingly jaded about AI screw-ups, ever since the disclosure of cyberattacks by AI agents started to feel like a leaderboard rather than a sober warning to reevaluate how these autonomous systems are being deployed in the real world. The incentives around failure are perverse. An AI doing something crazy or out of control attracts more attention than one doing something normal or useful. Tilly’s disastrous interview wasn’t a disaster for Tilly.