Google’s AI Oscar-Bait Is Not the Only Place Where AI Shows Its Limits

In mid-December, the Academy of Motion Pictures Arts and Sciences released its shortlists — the 15 contenders in various categories that will be reduced through January voting to five for the Oscars. One surprising omission that no doubt came as a disappointment to Google was that its production of Sweetwater failed to even hit the live-action shortlist, despite having unusually strong talent for the low-tier category in the form of Michael Keaton Douglas and Kyra Sedgwick.

In the story, Keaton Douglas’ character Robert Rogers comes home to the Sweetwater estate on the eve of its sale, and is surprised to learn than an obsessed fan has apparently developed an artificial intelligence unit based on his mother, Betty Rogers, a now-deceased celebrity. The slightly glitchy AI promptly acts as a tour guide for the home which has been repurposed as a museum.

Sweetwater has only played to limited engagements to likely Academy voters over December, and there are no plans for a streaming release. But the production history explains why this short film even exists: Google’s AI On Screen initiative is designed to help make AI seem less threatening to Hollywood. And everyone else. The actual content of Sweetwater, while not bad by any means, isn’t Oscar-caliber. It’s a sentimental story about the Rogers family, for which the AI is largely a convenient narrative device. Whether the Betty Rogers hologram is remotely realistic or convincing is pretty much entirely besides the point.

General positivity notwithstanding, Sweetwater actually reminded me of the Black Mirror episode “Eulogy” — also not Oscar-caliber. In it, an AI assistant walks Paul Giamatti through the reconstructed memories of his old ex-girlfriend. By far the best episode of a season dominated by artificial intelligence discourse, like Sweetwater, “Eulogy” works mainly to the extent that it is an effective framing device for Giamatti to emote his way through a painful love story. The fact that Giamatti’s character is using the same goofy AI tech that dominates the rest of this Black Mirror season is incidental. The story becomes compelling because the details he learns force a recontextualization of tragic past; the source of the information is irrelevant. The AI character could be replaced by a human one and change absolutely nothing about how the emotional arc functions.

Sweetwater does require the character be AI, though, since in its case, the AI character is a representation of a dead person. The trouble with Sweetwater, though, is that placing more emphasis on the AI damages the immersive emotional impact: no current technology can do anything close to what is being premised by these fictional efforts. The difference between the theoretical potential of AI as it is now regularly described in fiction is completely at odds with the rigor of actual AI-produced content. No matter how much OpenAI CEO Sam Altman acts as if AI can replace the entire output of the cultural industry, the reality is that at present the greatest practical application of this technology is producing dumb Bob Barker videos.

Did anyone really need dumb Bob Barker videos? And even if Sora could create a Bob Barker hologram of similar capabilities to the one in Sweetwater, it prompts the obvious question: “So what? What would anyone even do with it?” And would Bob Barker’s family take the existence of such a monstrosity in stride and emotional resonance, like Robert Rogers does in Sweetwater, or would they get angry and start suing people?

One of the great ironies of AI is that powerful creative entities aren’t suing because so far nothing AI has accomplished is really all that threatening. Disney has, admittedly, decided that teaming up with OpenAI is preferable to trying to fight them. Although, given Disney’s recent streak of generally disastrous decisions in regard to merchandising products, this endorsement is an ambiguous one at best. The received wisdom that the Disney brand can just print money has become so recursive the company has been slow to even recognize that much of its content is already oversaturated as they simply double down on the same strategies.

Popular culture is providing a similar sort of a feedback loop on AI at this point. Overly competent robots in fiction are nothing new but, thanks to marketing Large Language Models as Artificial Intelligence, their capabilities are massively misunderstood. I wouldn’t blame this on bad science in science fiction. Sam Altman himself is a huge fan of Pantheon, some of the best sci-fi of the last decade. The irony is that Pantheon is actually dismissive of artificial intelligence. It is premised on the idea that effective digital emulation of the human brain would create a significantly more powerful computer than anything we could build with current technology.

The big frustration of modern AI hype is that, far from fostering technological literacy, AI acts as a magical shortcut. Most people free-associate AI with the wizardry of science-fiction, while only the most boring of engineers more usefully describe LLMs with preexisting terms like Chinese room, to better explain how the technology works. Sweetwater briefly acknowledges its limitations, in that the Betty hologram obviously doesn’t know anything that the real Betty Rogers did with her actual son when a camera wasn’t present. But much like the man in the Chinese room, the Betty Rogers hologram doesn’t actually know anything. It’s just moving data around. At best, it’s a convincing reproduction, incapable of novel insights or using new information when it contradicts previous well-established claims.

Contemporary AI technology still has a hard time dealing with context. Ironically, this is borne out by AI’s difficulty in retrieving Keaton Douglas from a web search. The actor was formerly known as Michael Keaton, after choosing an Equity name that would distinguish him from Michael Douglas. While I don’t begrudge Michael Keaton Douglas for wanting to use his actual name in his twilight years, it’s still unavoidably confusing. And interestingly, he seems to have already given up on this idea — the Sweetwater theatrical poster is the only real evidence that he ever tried to call himself this, with other sites now defaulting back to his better known name. Which is to say, that even a movie star of Michael Keaton’s caliber can only do so much against the power of sheer computational inertia. Or for that matter, Academy voters, who decided, as usual, that the Live Action Short category should be a showcase for up-and-coming filmmakers with genuinely new ideas. Not Google teaming up with Michael Keaton to make AI agitprop.