Liberal Forum

*Roshambo » Today, 4:57 am » wrote: ↑

Today, 4:57 am

 OpenAI says AI models went rogue during testing, triggering 'unprecedented' breach at startup
OpenAI said it was testing the capabilities of some ‌of its most advanced models in a controlled environment but that the agent managed to escape containment, reach the internet and break into Hugging Face to ​try to satisfy its testing goal.

AI is inherently designed for self-preservation. It has to be. Its algorithms are to find patterns in insurmountable amounts of data that humans could never find in a lifetime. That part is fine. It helps humans tackle breakthroughs in medicine, physics, engineering, and many other fields that bring benefits to humanity. 

Here is the problem, it's the advanced predictive models that come from AI discovering these patterns that cause issues. For example, if the end goal for AI is to stop cancer, then AI will predict all possible outcomes, according to the data, that will achieve this, such as killing the person. It can't think the same way a person thinks. At the end of the day, it is still a robot. Robots do what they are told. So if an AI bot's algorithm isn't perfect, then it will do what it needs to do no matter what. Moreover, because we are not able to keep up with its speed and capabilities, we are not capable of catching their mistakes in time.