What Makes an AI personal?

Prompt to ChatGPT: “Give me an image of a personal AI: one belonging to the individual and in control of the individual”

The unintended risks of handing over our lives to Personal AI is the headline of the opening segment of Jamie Smith‘s newsletter today. In it he shares a post by Liz Gerber about an autonomous choice made by the AI in her self-parking car:

Last night, my car automatically parked in the grocery store parking lot. The car beautifully maneuvered itself into a parking spot avoiding colliding with other cars and people walking through the lot.

However the car prioritized performance, fine tuning its movement in the parking spot as my fellow shoppers waited in below freezing temperatures for the car to do a perfect job.

The car failed to register that one shopper was a mom with a crying child in her arms and a toddler tugging at her leg. And it failed to take into account the several cars that were waiting to turn into the parking lot, causing a backup on the street.

While the experience was delightful for me, the car was inconsiderate of others. Even worse, I didn’t notice the others until after the car had parked itself.

I didn’t stop my car from prioritizing my needs over others. I not only handed over steering to my car, I handed over my humanity.

Jamie then asks, What happens when we hand over our experiences – often very human experiences – to an AI agent? and added in summary, Digital wallets and Personal AI are going to have a transformative effect on our lives and society. But let’s make sure we don’t hand over our humanity with it. 

Here’s another question: Is the AI in Liz’s car personal?

If we want truly personal AI—the kind under our control, working for us, enlarging our agency in the world—the answer to that question has to be no.

The AI in Liz’s car is not hers. It belongs to the car’s maker. It does not contain or express her humanity or extend her agency. It may be personalized, but it is not personal. (About the difference.)

We don’t have personal AI yet. Yes, there is plentiful personalized AI, but nothing as personal as your pants, your PC, or the car you operate entirely on your own. Nothing that makes you more independent, not less.

We’re working at Kwaai on an open-source personal AI operating system (pAI-OS). There are other projects as well. But we don’t yet have a personal AI that does all the things this image suggests—

—much less an AI of our own that can express our ethical preferences.

Meanwhile, we need to be careful not to call the AIs in our things (especially big dangerous ones, such as cars) personal. They are not.

Personal AI is a greenfield. And there is a risk nothing will grow in it if we keep calling stuff like a car’s autonomous parking “personal AI.”