With enough data, you can model user behavior pretty accurately. But you can also use the same data to train models that generate valid user behavior patterns.
The ad industry is very profitable, but it seems to hang in the balance. Ads are effective as long as data you use to segment users are accurate. What happens if data becomes corrupted? Something like "This Person Does Not Exist" [0], but for user data. Your user behavior just remarkably similar to your actual behavior, but you can filter out the sensitive parts.
I'd love to see someone come up with a program or little gadget you can solder to the "smart"-things, and actual usage telemetry will train a locally deployed model that generates user behavior data that is indistinguishable from real user data, only it doesn't say anything about you.
The ad industry is very profitable, but it seems to hang in the balance. Ads are effective as long as data you use to segment users are accurate. What happens if data becomes corrupted? Something like "This Person Does Not Exist" [0], but for user data. Your user behavior just remarkably similar to your actual behavior, but you can filter out the sensitive parts.
I'd love to see someone come up with a program or little gadget you can solder to the "smart"-things, and actual usage telemetry will train a locally deployed model that generates user behavior data that is indistinguishable from real user data, only it doesn't say anything about you.
[0] https://www.thispersondoesnotexist.com/