New AI Experiments at CES

Every year's CES gathering is the single most important stage for vendors to show off their consumer products. This year the headline act was, of course, Nvidia CEO Jensen Huang's keynote — Jensen is the current top name in AI (arguably), and coverage and analysis of him are already piled sky-high, so what I'm more interested in are some smaller, more curious questions. Like: where else can AI actually be used?

Getting Specific — the Commercial Market Hunts for New Possibilities

The main question over the past year, as AI has gotten deeper and more entrenched, is that nobody's quite sure what's left beyond the use cases GPT and the big model makers have already staked out. So the vendors at CES are all trying to figure out their own corner of the AI story. PC applications are still the mainstream, of course, but this year's CES made clear that people are starting to get much more specific about where AI fits — and you could see plenty of attempts at exactly that.

CES has had AI assistant gadgets for a few years now, and that category hasn't performed well. The two best-known names, Humane AI Pin and Rabbit, are widely considered failures. The main reason is that the big model makers' own tools are already so convenient to use on a computer, and every so often the top platform makers refresh the human-computer interaction logic on PCs and phones anyway — so if you build a standalone product, the giants will periodically pull the rug out from under you, and you're never going to win that fight.

But there is one interesting, genuinely getting-specific case that's already found a commercial market over the past few months: what people are calling "hand-built AI toys." Since something like GPT can already hold a conversation, all you need is a small electronic device with input and output (concretely: an old smartphone will do), stuff it inside a stuffed animal's belly, tweak the model a little, and you can run it offline. That conversational back-and-forth lets it complete something like a real chat loop and produce a companionship effect. If Tom Hanks's volleyball had come with this feature, he could have literally chatted with Wilson on that desert island.

Some people are already selling these little gadgets on Twitter/X and Taobao, and business seems decent — a quick look shows plenty of them are being manufactured to actual tooled-mold quality, not just hacked together. Compared to the companion robots of science fiction, these are still crude — arguably two entirely different things — but it doesn't feel like the future of companion robots widely used for chatting and play is all that far off.

Another human-computer interaction case for AI that I think is genuinely more mature is glasses. Smart glasses have been a market ever since Google Glass, many years now, and last year saw real progress. Apple's Vision Pro wasn't exactly a hit, but Meta's collaboration with Ray-Ban has been a genuine success. The job requirements were kept deliberately small — record video, handle some simple AI interactions — and that was enough to prove the use case works.

AI Smart Glasses Look Like the Most Mature Application Yet

In that same vein, one of the more interesting glasses vendors at this year's CES was Halliday, which projects light directly into the eye — a green beam shone straight into the eye itself, rather than forming an image on the lens. The image forms on the eye. Naturally, that makes people a little nervous; I noticed reporters in the coverage were a bit worried about whether it could damage the eyes. The upside of this approach is that the glasses themselves can use ordinary lenses, and the lenses don't need to be oversized, since the projection goes straight into the eye — so the supporting hardware can stay small too, meaning it can fit into a normal, lightweight frame instead of some bulky custom design.

You can imagine glasses like this producing live captions that amount to real-time translation. Live captioning and real-time voice translation are arguably the single biggest AI use case out there right now, so this category of eyewear is positioned to walk straight into an already-mature market. I'd expect a fair number of similar products to launch this year.

As for the rest of CES, there's always the recurring question of why some products bother using AI at all. Plenty of commentators love to mock these as pointless, but honestly, I rather enjoy this category of product. Take the smart seasoning dispenser that can scan a recipe (that one deserves a perfect ten on the pointlessness scale), or — very reasonably, for the American backyard — a smart bird bath (with AI that can identify the species visiting it). Then there's gaming gear maker Razer's new Copilot, an AI that acts as an esports coach and analyzes your play. These are all defensible use cases, but they feel just slightly forced, like products built mainly so the company could say "yes, we have AI built in too."

I actually think the least exciting AI developments overall are, in fact, the most important ones. Take this year's biggest CES exhibitor category — automotive AI — which generated an enormous volume of coverage. Or the news that Panasonic partnered with Anthropic (the company behind the well-known Claude model) on an AI fitness coach — boring but genuinely solid news, because getting these large models into more everyday use cases requires the big appliance makers and the other companies that already shape our physical world to quietly get us used to living with AI everywhere.

These real, practical applications are sensible but not exactly eye-catching, and yet I still spent a lot of time at CES looking at all the slightly-forced AI-enabled gadgets — it felt a bit like throwing spaghetti at the wall to see what sticks. A little pointless, sure, but throwing spaghetti at a wall is fun in its own right!