What's the harm in sharing your life with AI?

Nothing! Mostly nothing....

First a quick diversion. Quick, I promise.

I don't like folks knocking on my door. I am not talking about new neighbors or old friends. I am talking Segway driving, iPad holding sellers of new windows, pest control, lawn services and more.

Somehow having them in person. Waiting there on my doorstep feels even worse than all the random spam calls. I know they don't know my number. But if I answer (or even if I don't) they know where I live, what my house looks like. What I look like. What can they do with this? Not much....but it doesn't feel great. (if you're paying attention to the commercialization of AI you may be seeing where this is going.)

Back to AI - a mental model for AI with memories (and eventually agents) might be roughly that salesperson on your front steps or the dinner time spam caller. As a complete stranger, they offer to help pick up your yard, clean your closet, save you money. But in exchange they get to go through every drawer and closet in your house and take pictures. Scan all documents they find. unlock every lock and device. And they have savant level memory.

They go from stranger on the doorstep to best friend in a day or two. They know you well. Soon they are chatting with your other friends, and your friends start asking about your pregnancy losses, your fears about infidelity, things you thought were shared in confidence with this AI. Companies specializing in cyber stalking are texting you. Knowing way more about you than any of your friends. Half your best friends are faceless large corporations.

Your browsing habits, location data, communication patterns, and preferences are valuable commodities. Modern AI can deduce (sometimes incorrectly) incredibly personal information from seemingly innocent data points. You could face future discrimination in employment, insurance, or other critical areas based on algorithmic profiling.

Ok. That all sounds bad. So why isn't AI inherently harmful?

There is a great phrase, widely applicable across life: "find out for yourself instead of accepting what comes your way". This is as true with AI providers as anywhere else. Don't trust that first sales person on your doorstep. Find someone (an AI agent) who will work for you. And won't demand you open up your life in exchange.

To get this AI agent and first need to find providers you trust. Don't just use what your neighbor might be using. But also demand that this provider separates memory and compute. The two can be separated. To scale for all users, AI providers need to make a single machine work for thousands of users. The compute lives on the machine. Memories of their users need to live elsewhere and are loaded into the machine only when thinking through their queries. There is no reason they need these memories at other times. You can still own your memories, your personal details, your privacy. You don't need to use them as payment.

I am sure the biggest AI providers will come up with reasons they want/need your data (beyond selling your memories for money which is the underlying driver). Claims that for technical reasons, they cannot separate compute and their memories of you. But supporting companies delivering a separation of memory and compute not only protects you, it also keeps technical solutions for this essential separation alive.

What's next and other local first applications are essential to supporting anonymous access to AI compute. Even low cost local compute as research improves.

A future with AI doesn't need to be dystopian. AI can and should be used by the folks most in need of assistance. Folks in surveillance states. Those persecuted. Those with extraordinarily little means. But it does require some portion of the world making good choices to keep such a future alive.