AI’s algorithmic mimicry cannot match craft of human actors

So one looks at Tilly Norwood’s sole onscreen outing, a comedy sketch called AI Commissioner, and scoffs at its wooden, stilted dialogue.

- 

But this fails to see the pattern of what is to come.

Consider that in 2019, ChatGPT’s GPT-2 model could barely count to five or string together coherent sentences. By 2023, GPT-4 was outperforming 90 per cent of human test-takers on medical licensing and bar exams.

Leading researchers at frontier AI companies now believe we will achieve systems matching or exceeding human cognitive capabilities across virtually all domains before 2030.

Given this pace of change, on what rational basis can anyone remain blase about AI actors? We should make no mistake that the gap between “embarrassingly bad” AI output in show business and “good enough” AI output will shrink at breathtaking speed this decade.

The second source of nonchalance is perhaps more dangerous: the faith that everyone involved, corporate players included, will act reasonably — respecting the art of DiCaprio and his peers, caring about intellectual property, managing widespread job displacement. This thinking is, to be charitable, delusional.

It is practically axiomatic that major AI companies operate on a “move fast and break things” philosophy — act first, apologise later, if at all.

OpenAI, not for the first time, put this on display with Sora 2, first launching with a policy letting users create videos featuring copyrighted characters unless rights-holders explicitly opted out, before retreating to an opt-in model after immediate outcry.