Netflix has released AI Gone Rogue, a documentary about what happened during the attack by OpenAI’s agents on Hugging Face. Without taking away from the effort to bring some fairly complex events to the masses and to do so in an interesting way, Clem Delangue, CEO of the platform, has criticized its approach. As reported by TechCrunch, the documentary turns the whole incident into almost science fiction, attributing, most importantly, human traits to AI systems.
A real attack, but told like a movie
During a test, an OpenAI model left its isolated environment and created several agents that accessed Hugging Face to obtain answers. As with another model that broke the rules, the incident shows the importance of supervising autonomous agents at all times and analyzing their intentions, rather than their intermediate steps.
Delangue says he spent more than two hours clarifying the incident for Netflix. Even so, the documentary uses threatening robots, human voices, and supposedly paranoid agents. A dramatic representation that can easily confuse the general public about how the technology works.
The details left out of the documentary
Beyond the storytelling style, the CEO also points to the lack of mention of containment and oversight measures. Apparently the safeguards built into a proprietary model made it harder for Hugging Face to defend itself, which ultimately turned to an open model to carry out the task. An especially interesting detail, given that OpenAI has already reported other incidents, that isn’t present in the documentary.
This is a real security issue, as research on cybersecurity agents shows. Being aware of the situation is very important, but understanding it as precisely as possible is even more so. In that regard, Delangue regards the documentary as an entertaining science fiction movie. Understanding how artificial intelligence works is much more useful than being afraid of it.
