The Humane Ai Pin: A Terrible Solution to a Problem That Doesn’t Exist
A GPT-4 wrapper, camera, microphone, and projector in a trench coat
I won’t bore you with the details here, and I encourage you to watch Humane’s full announcement video for all the information and demos, but here’s the rundown: Humane, a company founded by ex-Apple executives Imran Chaudhri and Bethany Bongiorno in 2018, has been hyping up a product announcement for a couple years, after raising hundreds of millions of dollars in funding, including from OpenAI’s chief executive, Sam Altman. On Thursday, it unveiled its first product to the world, the Humane Ai Pin (side note: There is no correct way to capitalize the name, as Humane has variations of “AI,” “Ai,” and “ai” on its website). The announcement has generated a lot of buzz within the technology community for a couple of reasons: Humane has hyped up the announcement for years, touting it as the next frontier of computing, and has said that the product is the first physical device made for artificial intelligence (what it means is transformer models, but we’ll roll with it).
The Ai Pin is a lapel pin that comes in two parts, a battery booster and the pin itself, the former of which goes at the back of your shirt or jacket, and the latter of which magnetically attaches to the battery, hugging the fabric of the shirt. The Ai Pin itself has a laser projector, camera, microphone, “Trust Light,” and touch interface that all allow you to control it. It does not have a screen; all of the interactions come through the speaker, microphone, and laser projector, all of which you activate by tapping the touchpad on the front of the device. The speaker faces upwards, pointing toward you, though others will most certainly be able to hear audio coming from the pin. The projector displays an operating system called “Cosmos” onto the palm of your hand when you extend it in front of the pin. You move through Cosmos with gestures like pinching your index finger and thumb, as you would with Apple Vision Pro, clasping and turning your hand to select controls.
You can also tap and hold the touchpad to activate the voice assistant, which Humane implied is the primary/preferred method of interaction with the Ai Pin. The voice assistant is powered by version four of OpenAI’s generative pre-trained transformer, which is the same transformer-style large language model used to power ChatGPT, the incredibly popular AI chatbot. The assistant, dubbed “AI Mic,” runs on-device, with the help of an unnamed Qualcomm Snapdragon processor, presumably one of the newer ones that supports the ability to run transformer models on-chip. The microphones aren’t always on and listening, unlike smart home assistants like the Nest Home devices from Google and Echo devices from Amazon; you have to explicitly tap and hold the touchpad on the device for it to begin listening.
AI Mic is a wrapper for GPT-4, as in it’s customized for whatever Humane wants it to do. It has multimodal capabilities, including searching the internet and viewing images taken from the Ai Pin’s on-device camera — just like ChatGPT — and can interact with many internet services, like TIDAL, the music streaming app. It accesses the internet via Humane’s own T-Mobile-powered mobile virtual network operator, a cellular network operated by Humane exclusively for its devices. Accessing that cellular network requires a mandatory $24 monthly subscription in addition to the pin’s initial purchase price of $700. You cannot use the Ai Pin without the monthly subscription, which also means that if Humane ever goes out of business, your Ai Pin will be rendered useless. Humane touts the Ai Pin as being able to answer basic questions, customize music playlists, take photographs and videos, and do other things a smartphone would, including making phone calls and sending text messages. Each Ai Pin gets its own phone number, provided by T-Mobile.
That’s the rundown on the Humane Ai Pin. Humane says it’s made to be a smartphone replacement, not a smartphone accessory, unlike the Apple Watch and other wearable gadgets. It’s apparent in the number of features it has, but I think this is fundamentally the wrong strategy to take with a product like this.
AI Mic
The main selling point of the Ai Pin is the AI assistant, AI Mic. The product essentially relies on the AI Mic to do everything. While Cosmos is the graphical user interface, the brain is the AI Mic powered by GPT-4. Chaudhri demoed the assistant and asked it to send a text message, which it was able to do. However, he also showed off the power of having GPT-4 running on-device and asked it to “make [him] sound more excited.” This is something only a large language model can really pull off, and it did it without a hitch. However, the hitch came when Chaudhri asked the assistant, “When is the next solar eclipse, and where is the best place to see it?” The assistant replied in a robotic-sounding voice with an incorrect answer, saying that the best place to see the April 2024 solar eclipse was in Australia. The eclipse is not visible in Australia, it’s only visible in North America, according to the National Aeronautics and Space Administration.
These inaccuracies are common for large language models because the singular job of a language model is to predict the next word in a sentence. The transformer understands the prompt, splits it up into tokens, processes it, and then tries to predict what will come after that prompt. Nowhere does it look actual information up to make sure what it said was true. This is fine for AI chatbots, where a disclaimer below the message box is common (ChatGPT’s disclaimer reads, “ChatGPT can make mistakes. Consider checking important information.”), but it’s not acceptable for a product where its sole method of interaction is a chatbot assistant. The Humane Ai Pin, at least to my knowledge, doesn’t have a web browser. When you use the Ai Pin as a smartphone replacement as Humane wants you to, there’s a good chance of receiving incorrect information. Hilariously, this inaccuracy was not spotted by some reporter who was given the ability to report on a pre-release product — there were no press review units at all, in fact — but rather, it was in Humane’s highlight 10-minute demonstration posted on its website. You can view the inaccuracy in all its glory even now; it has not been corrected.
None of this is to drag on Humane. I’m sure throwing another company’s assistant into your product and calling it your own is difficult work. But it highlights the problem with this product: you can’t depend on it. Humane wants the Ai Pin to be a replacement for the smartphone. Chaudhri demoed a text messaging search feature — which I admit is very neat — a real-time translation feature powered by OpenAI’s Whisper text-to-speech model, and a personalized music DJ feature among many other things. However, if you’ve noticed, you can do every single one of these things on your smartphone you have in your pocket already. Real-time translation is just an app. So is the ChatGPT voice assistant. If you have an iPhone 15 Pro, you can set your Action Button to enable ChatGPT with voice, which uses the same GPT-4 model as the Ai Pin. None of this is innovation — it’s just blatantly copying another immensely successful product and putting it in a hardware box.
I have to imagine how the world would react if OpenAI made a smartwatch or smart home speaker. I reckon it would fly off store shelves. Heck, I’d buy one.
Never Bet Against the Smartphone
As Marco Arment, a well-known podcaster and developer of the Overcast podcast application, says, don’t bet against the smartphone. People love their smartphones. The smartphone is a marvel of engineering, fitting a full-blown desktop computer into your pocket. But more importantly, it does everything the Ai Pin can do, and so much more. Tech companies like Google and Apple have noticed this, which is why they also build smartwatches, further miniaturized computers that are a bit less helpful than smartphones, but that are also immensely more versatile and portable. Humane is trying to make AI portable — I see the vision. But this isn’t the way to do it.
Humans are visual creatures. We want to see things and interact with them. For thousands of years, we humans have written with ink and paper, communicating our thoughts and ideas via a common language inscribed on a physical object. The natural progression of ink and paper was the typewriter, which made it easier for humans to write long paragraphs of text on paper, then the computer, which converted those visual characters to series’s of numbers that could be manipulated in any way. These inventions that shaped humanity and society are visual inventions because humans are best at visually expressing themselves. Text is the most dense medium we know of to convey information — period.
Humane blatantly ignores this fact. I assume Humane’s original vision for this product was for it to be audio-only, but a bunch of engineers realized that it was a terrible idea and built a projector and user interface. Humane’s founders, husband-and-wife Chaudhri and Bongiorno, both share the idea that smartphones have somehow distracted us from reality, and were set on building a device that blends into your life. Aside from the fact that I find incredible irony in the creator of the iPhone’s touch interface (Chaudhri) saying phones have distracted us from reality, Humane’s founders ignore the fact that phones aren’t the ones that distract us — apps do. Nobody except for a small baby is mesmerized by a phone’s Home Screen. It’s not the phone, it’s the apps. If the Ai Pin had social media, (a) nobody would use it because nobody wants to view TikTok on a projector, and (b) the Ai Pin would become addictive. The idea that something needs to replace the smartphone because the smartphone is “addictive” is laughable.
I’d argue that the Humane Ai Pin does a terrible job of embracing reality. Nothing is more jarring than hearing someone talk to Siri in public, and it’s extremely unnatural to talk to yourself. It feels awkward to do it in public. Humane shouldn’t have imagined a world where everyone will talk to themselves — it’s just not happening. There is a reason phones are great: You can lean against a wall, unlock your phone, and instantly find yourself immersed in your own world with the ability to do absolutely anything you want with a full-blown computer in your hands. This promise is why phones took over as many people’s primary computers. A smartphone is a computer, but anywhere. The Humane Ai Pin is simultaneously trying to prove to the world that smartphones are addictive and trying to replace the smartphone as we know it. You can’t have it both ways. People love their smartphones for a reason: they’re fantastic little devices. The Ai Pin is not a fantastic device because it requires people to do something uncomfortable and awkward.
Humane seems to have forgotten a world before smartphones, because if its engineers and executives remembered that cruel time, they wouldn’t have built a product with such limited functionality. AI assistants have their purpose, but they aren’t meant to replace an operating system. The Ai Pin is not a computer; it’s a speaker with GPT-4 that mediocrely compares with actual smart speakers. Physics prevents the Ai Pin’s onboard speaker from sounding any good, GPT-4 consistently gets facts incorrect, and speakers can’t convey information with the density and precision that displays can.
People are very accustomed to the smartphone. There are at least three generations on Earth right now that have not known a world without computers. Humane is founded by minimalists who don’t understand the practicality of visual devices. I would’ve been able to give Humane a pass if it did its job well, but from the demos, AI Mic is bad at being an assistant, and the Ai Pin is bad at being a computer. Some small fraction of nerds might find value or novelty in a minimalistic personal assistant, but the vast majority of people realize that their phones and smartwatches can do everything the Ai Pin can do, but better. People aren’t itching for a smartphone replacement — it’s not a problem that needs to be solved. And trust me, you really can’t win in the technology sector if your product doesn’t appeal to the masses — it’s simply impossible.
Cosmos and the Camera
The visual, projected interface of the Ai Pin, Cosmos, is largely impractical. In the demonstration video where Chaudhri switched music tracks via Cosmos, the projected interface was hard to read. I’m unsure if this is because of the angle at which the video was shot, or something else, but the visual interface seems like a sub-par, backup experience. Clearly, the Ai Pin is meant to be used with your voice, not with your eyes or hands. Cosmos has no app store, though Humane says an application programming interface is coming “later” — whatever that means. I don’t think it’ll be for third-party developers to make apps for the Ai Pin, however, as Humane’s pitch is that the Ai Pin doesn’t have apps. The entire operating system compliments the device in a way that’s quite incredible — something I’d expect from two ex-Apple engineers. There’s a now-playing screen (you can connect Bluetooth headphones to the device), a call interface, and some other screens with presumably many more that haven’t been demonstrated yet. None of these are apps; they’re just part of the system.
I assume this is in an attempt to “cure” smartphone addiction. The Ai Pin doesn’t provide any experiences — it’s purely utilitarian, which I find is a jarring product decision. Again, I cannot stress this enough: Humans are visual creatures, and we choose to look at things to understand them better. We talk to people, and we look at things. Cosmos is the only visual component of the Ai Pin, and it’s lackluster at best. Cosmos seems more like the Control Center for the Ai Pin rather than its primary software experience, which is the GPT-powered AI Mic. There’s no keyboard, obviously, and the only way to use it is by looking like a dork and extending your hand out to use your palm as the surface on which the projector projects the interface. The gestures seem fine, though also awkward at times; I’m unsure of the last time I twisted my palm while it was extended out in front of me. Sure, I guess Cosmos does cure smartphone/computer addiction, but not because it’s inventive. It’s just hostile to use. I don’t know why someone would use this instead of their smartphone.
The Ai Pin also has a camera, which I think is the only useful feature of the device. You can take pictures and videos by asking the assistant, and they save to your smartphone (because Humane itself realizes it’s not a smartphone killer) via an app/website — it’s unclear what this is — called Humane.center. It’s like Google Glass. Humane didn’t elaborate on exactly how good its camera is, but I’m guessing it’s pretty modest in resolution and fidelity — the Ai Pin is a lapel pin, after all.
The more interesting use case for the camera, however, is where GPT-4V (the “V” stands for “visual”) comes in. Humane’s founders both touted Cosmos’ health features, including a calorie tracking feature. The idea is simple: hold up some food, ask Humane AI how many calories are in the food, and tell it that you’re going to eat the food. Simple; it logs all of the information in Humane.center so you can view it later, and you can even ask it how many calories you ate later. This doesn’t apply to just calories, but also sugars, carbs, proteins, etc.
The problem is that this doesn’t work even nearly as well as it sounds like it will. Chaudhri held up some almonds and asked the Ai Pin how many calories were in them, and it answered with a number three to four times higher than the correct answer.
As many have pointed out, this idea of asking AI to determine the macronutrients of a food just by looking at it is nearly impossible to execute. Food is incredibly complex, with different ingredients, portion sizes, and additives. Nobody — not the most seasoned dietitians and chefs — can determine how many calories are in a food just by looking at it with such a degree of precision. While nifty in theory, it’s just not helpful in practice. And once again, it’s hilarious that these errors showed up during a pre-recorded world premiere demonstration of this product. Please, Humane, proof-watch your damn videos.
Humane’s founders, during the demonstration, also asked the pin to search the price of a book online. Again, another great idea — imagine being in a bookstore, holding up a book, and wondering if you can buy it online for cheaper — but in the demonstration, the Humane Ai Pin abysmally failed once again. Many have pointed out that the book from the unveiling is more expensive online than the Pin had said, which again, isn’t very good. Chaudhri then asked the Ai Pin to “buy it,” but the pin never asked for confirmation or authentication, and didn’t even say which retailer it bought the book from. These integrations with third-party services aren’t found on Humane’s website. It’s simply an abysmal failure in both advertising and product engineering.
Again, I’m not trying to drag on Humane. This is a new product that uses generative artificial intelligence — something Humane didn’t develop itself. But it’s proof that generative AI, or more precisely, transformer model-based chatbots, aren’t ready to be relied upon. The Humane Ai Pin heavily relies on technologies that are in their infancy, which makes me wonder, what’s the future of Humane, and what happens to it if its partners bail out on this strategy?
The Future of Humane
OpenAI, the founder and chief executive of which, Sam Altman, is a major investor in Humane, only announced the GPT-4 API Humane uses to power the Ai Pin a couple of months ago, back in April. As I said, Humane was founded in 2018 by Chaudhri and Bongiorno, and according to The New York Times, chose to build the Ai Pin instead of a women’s healthcare product after a Buddhist monk led them to an angel investor turned CEO who persuaded them on the idea. This is a true story; I am not making this up. In 2018, transformer model-powered chatbots were in their infancy — GPT-1 was announced in June 2018, just months before Humane’s founding — so how could Humane develop its product around a technology that had just freshly been announced to the world? Even though I find Chaudhri and Bongiorno’s decision-making not fully sound, you have to be a real idiot to quit your high-paying job at Apple to work on a product that might have no future at all.
No, what I think happened is much easier to understand: Humane was founded in 2018 to develop a device — maybe a pin, maybe some other wearable device — that aimed to do much more than the Ai Pin does now. Maybe it had a camera, a screen, or something else — I don’t know. But whatever it was, it wasn’t this. That product was somehow impossible to create, so sometime around November 2022, Humane pivoted to the Ai Pin as we know it now, as AI chatbots emerged and it realized it could use them to its advantage. Now, it has made a mediocre product using the technology the investor community is going gaga over — pre-trained transformers. In our current society, if you want to raise a lot of money in seed funding, just make a GPT-4 wrapper. It sounds almost too good to be true, but it’s correct. That sounds like exactly what Humane did, though just a couple of months early.
Now, it has shown the product off more to get more funding, and of course, it’s built around “AI.” It’s a sound business strategy. If I said this about any other company, I’d feel a bit bad because it’s quite defamatory and offensive. But if you watch the 10-minute demonstration from Humane’s founders, it lacks excitement. Both of them are trying to be the Steve Jobs of AI technology, but the strategy is failing miserably. The event, if you’ll call it that, was dull, boring, and eye-roll-worthy. It was not a product demonstration; anyone who’s debuted a product — no matter how stupid that product is — has been more excited about showing that product to the world than Chaudhri and Bongiorno were. They’re terrible on camera.
The whole thing was set up like an investor pitch rather than a serious product presentation. There was no excitement, no joy, and no enthusiasm for the service. Just a dull, boring pitch for a product that nobody is really interested in. I speak like this not because I’m a pessimist about technology, but because Humane’s marketing team hyped the product up like it was the next iPhone. Earlier this year, when Chaudhri first teased the product at a TED Talk, my expectations died down slightly, but I was still enthusiastic about what the founders had to bring to the table. There’s room for another gadget, trust me. I even think there’s room for one powered by GPT-4. But this isn’t it.
I have major concerns about how Humane will be able to keep this product up and running far into the future. What happens if OpenAI sunsets its API? What happens if Humane goes out of business? What if an Ai Pin breaks; what does the service procedure look like? What if Humane retires its MVNO? These are all concerns anyone who buys a product from a scrappy Silicon Valley start-up shares. But this time, it’s different. The Humane Ai Pin is not cheap; it’s $700. For $700, Humane has to have its stuff figured out.
The Humane Ai Pin is Humane management’s admission of failure. It’s very clearly not what they wanted to build nor what they meant by a “smartphone replacement.” I hate to be the downer here. I was excited about this product, and I still think Humane can build something impressive. After all, the people behind Humane are very smart. I’m rooting for them. This isn’t just some average Silicon Valley start-up — it’s by Apple executives. But the Humane Ai Pin lives in a world where consumers are all minimalists, just like its inventors. It fails to understand reality.
I’d love to be proven wrong, but the Humane Ai Pin looks like an abysmal failure just waiting to happen. The problem isn’t just the $700 initial price and $24 monthly subscription, but the entire product’s philosophy altogether.
