The Humane AI Pin. What, what, what?
I recently said that the Humane Ai Pin reminded me of the Microsoft KIN phone.
First, let’s be clear, the Ai Pin is a smartphone. It has a phone number. It makes calls. It has a data plan. It interacts with the Internet. It is made of smartphone components. TLDR: It’s a smartphone. It does not disrupt or replace anything.
The issue I am trying to wrap my head around is that it’s objectively bad at both: Being a phone and being smart.
It's a bad phone because it's obscure in usage. Hands-free interaction has never worked well. "Call Bethany" is only ever good if you know exactly one person named Bethany. Sending dictated messages without seeing them is always a trust issue, no matter how neat an LLM model can write them for you. And it's bad because the default mode of interaction and calling is a loudspeaker. No matter how megapersonifisonic the speaker is, it's public.
Of course, if you use headphones, all of this is moot since it now is exactly like your existing smartphone.
And it's a bad Internet device because it ignores the web, apps and indeed any way to efficiently consume content. Being talked to in 1x is an aggravatingly slow way of consuming information. Even if it claims to be able to summarize everything for you.
What problem?
The smartphone is a black hole of integration. Just like a Swiss army knife, it subsumes all use cases and tools you can possibly do with this form factor.
Look at how the first iPhone was announced: "It is a music player, a web browser and a phone!" Since then, convergence didn't stop and it's now a newspaper, a camera, a TV, a radio and many more things.
Every now and then a company declares that less is more and creates a not-smart phone. It's true that smartphones might not be the best idea in a lot of ways, but the opportunity cost of not having all the tools and use cases available is simply too high. Dumb phones don't sell.
But what Humane wants to solve is not the smartphone, it's all the little problems with the computer itself: How to turn it on? How to add some configuration or data? Why doesn't it synchronize my contacts? How do I share this data with a group of friends? Why can’t I find my stuff? How to do an update? How to get my device connected to the obscure network in this conference center?
You know, computer things.
Humane wants everything digital to "just work", which is an insanely hard problem in a messy, unorganized, and unstructured world.
Like the book example. “This is $28 online!” At which store? New or second hand? And what do you mean “Buy it”? Do you have an account at that specific store? Which payment option do you use? And which shipment address? Where does the invoice go? When will it arrive? Don’t tell me you have solved what Amazon couldn’t: Seamless voice-based shopping.
What solution?
It seems that Humane wants to solve all this messiness with OpenAI and an army of GPTs.
Well, there is this tiny issue that ChatGPT, you know, lies. Sorry, it “hallucinates” (1).
(1) Lying would imply intent, which ChatGPT does not have. But OpenAI is a company, and they happily use a cute sounding semantic excuse like "hallucinating" to consciously hype a product that has no concept of truth. It might not be lying, but in my opinion it's equally dishonest.
Even in the demo the Ai Pin happily hallucinated incorrect information:
- The next solar eclipse will be visible in North America, not in Australia or East Timor
- Those fourteen almonds should have on average about 3–4g of protein in total, not 15g
- That book is “Spacesuit: Fashioning Apollo” by Nicholas de Monchaux, and the paperback has a list price of $44,95 and retails for around $30-something on Amazon, not $28
It just does it in a way you don't notice. Even Humane didn't fact-check the answers when they green-lit and published the video. And that’s scary. This is a literal black box that cannot be trusted with anything it tells you.
But even if it did tell the truth, then it would just be a personal assistant device. An Alexa or Google Assistant you can wear. And that doesn't solve computing because the promise of "It just works" does not start with the information layer, it starts with the IT layer. Unfortunately, ChatGPT cannot help you with that in any way.
What now?
One interpretation is that Humane created a dumb phone with bad user experience and that cannot be trusted.
Another interpretation is bringing me back to the KIN phone. In many ways the KIN platform was ahead of its time. An ambitious team identified a use case (phones as the center of social digital life) and doubled down on it. But it was early, the execution was weird, and it was expensive.
The same could be said about the Humane Ai Pin. Humane has been working on the concept since 2018. Back then OpenAI, ChatGPT and LLMs weren't a thing. Likely, an ambitious team at Humane wanted to solve computing by creating something more, well, humane.
My guess is that they fumbled with the reality of solving all this IT mess for five years, but it wasn't really going anywhere. Then OpenAI came along and offered a solution at least for the informational layer. It might not have been what they wanted to achieve, but it seemed like the best way this concept can be simulated right now. So, they took it and doubled down.
The thing is, though: The KIN was a spectacular failure. Being early, weird, and expensive wasn’t really a good combination. I somehow doubt that being inconvenient, expensive, and a liar is a better one.
Best of luck, Humane.