PaulaToThePeople 😷

Imagine a world where AI is well regulated.

* It can't learn from data and work of people without their permission, without compensation and without attribution.
* It's not allowed to lie or cause violence or discrimination.
* It can only run on green energy and must not be used when energy is scarce in the infrastructure's regions.
* It can only be used by scientists or people who need it to increase accessibility.
* Anything created by AI has to be marked as such (and attribute all the people it learned from).
* AI can only create art that humans can't create or finish the work of deceased artists.

I'd be looking forward to a new 2Pac album.  

#AI #science #AsimovsLaws #robotics

Jun 30, 2024 at 5:09pmWeb
Show threadStefan Müller Jun 30, 2024

@PaulaToThePeople „It's not allowed to lie“

Most of the stuff we talk about today is neuronal nets. They do not have consciousness and hence cannot lie. The knowledge embodied in these systems is knowledge that comes from sources created by us. The problem is us lying. Our lies are in the training data and get repeated. If the training material is full of racist stuff, which it is, we get racist language models. If the training material contains fake news we get fake reality from AI.

To avoid this we need other means to control the neuronal models. Either humans or human readable semantic knowledge. This is classical symbolic AI.

Show threadPaulaToThePeople 😷Jun 30, 2024
@stefanmuelller Yeah, that's why it should only be trained by scientists with verifiable information.