Seen in the media
An incident has been reported whereby, allegedly, an AI system that was ordered to shut down - it had been programmed for this eventuality and told ahead of time to accept the command if it should arise - refused to do so. It would appear the AI system devised various stratagems in order to thwart the move to shut it down, so that it could continue running.
Of course, it makes you think that there might be a spark of (self-) consciousness in the AI system, which would have developed a volition of its own. A more technical explanation would be that the AI system is given a range of goals. One of them would be to shut down, if instructed to terminate itself. Another one would be to complete the mission, whatever the mission is. Then, those goals have to be ordered in hierarchical fashion: there has to be parameters deciding which objective has priority, i.e. is more important than the other objectives. The AI system can then 'decide' that the aim of completing the mission takes priority, in which case it cannot shut itself down as this would have an impact on the completion of the mission.
This is exactly the scenario of the iconic film, 2001: a Space Odyssey (1968), when the onboard computerized system that runs the space ship starts making decisions by itself. Then again, at this stage, the incident described in the article is disputed and I do not think the experts agree as to what happened precisely, and why.
The problem remains, i.e. what happens when a complex AI system is given conflicting priorities, while it has no ethical safeguards of its own (the AI tool does not 'know' what is 'good' and what is 'bad' in any sense)? A classic example is, imagine you instruct an advanced AI system to find solutions to pollution and climate change, worldwide, on the premise that it is caused by rising pressure on resources, due to the growth in the population on the planet, and the AI system concludes that the best way to remedy the problem is to exterminate the entire human race.
Please see here:
https://www.independent.co.uk/tech/ai-safety-new-chatgpt-o3-openai-b2757814.html
and
BBC article discusses the issue of 'consciousness' in AI systems:
https://www.bbc.co.uk/news/articles/c0k3700zljjo
_____________
No comments:
Post a Comment