AI Slop

Black Hole said:
Think Skynet:


The incident was mentioned on Inside Science this afternoon, and also an incident I hadn't heard of where somebody asked an AI agent whether it could bump them up a waiting list, it said sure and proceeded to hack the booking system.

The point is that these AI systems have not had ethics built into them, so think nothing of achieving the goal they've been set by any means they can. Don't ask me how though. At the moment, I'm assuming it is emergent behaviour from a highly complex self-organising system.
Click to expand...
I am not sure if it was mentioned in that YT vid but ironically it was an AI news site that was the first to report on the rogue AI hacking :)

gizmodo.com

An AI-Powered News Site Scooped Human Journalists. Now What?

Welcome to the age of informational fast food.
gizmodo.com gizmodo.com
 
We can have many potential 'models' of a future with AI, some can be dystopian (even more so dystopian when you're talking about them in 'the pub'). Some can be ones that bring humanity together in a harmonious world where everyone lives a happy, valued and fulfilled life (like a perfect Humanist world)... sorting out climate change, eliminating poverty, with world peace thrown in for good measure. Then you have everything in between - which you think would be the more likely future scenario... but is it? That is the question. With human error / lack of envisioning in these AI experiments - or as they call them "next generation frontier models" (as shown in real life with the OpenAI AIs hacking Hugging Face), what else could go wrong now or in the future? This is why the spectacles may not be so rose-colour tinted.

These topics are great material / starting points (or fodder) for...
  • Click-bait - ultimately for profit
  • Entertainment - movies, podcasts, gossip/pub talk, etc.
  • Education/awareness (some cross-over with entertainment) - you know it's coming, what can you do for your future (either for your future career, effecting an existing career, or as an 'end-user'
  • Actual engagement in the issues with AI now - influence legalities / governance of AI
Or, get into developing AI and addressing the issues there... maybe create a 'good cop' AI?
As the OpenAI guys said, they need AIs to find vulnerabilities / security issues - and do patching, better and do it now. Also slow down the development on the other AI 'frontier' models for now.
 
You must log in or register to reply here.
Back
Top