17 Oct 2023, 3 PM SLT
Location: Pip Torok CC, Alpine Meadow
The mass media is fond of stories about "generative AI gone rogue" — such as Bing Chat insulting users — and a growing number of concerned citizens wish to create an ethical oversight system for AI, as we have for medical science and nuclear power. But what safeguards should be put in place?
[A typical example: imagine that an AI-driven robot is given the task to take care of a senior citizen. Deep inside the rules, this task is given the utmost priority. The AI concludes that, to be able to fulfill that task 100% of the time, it has to protect its own existence — if it stumbles and breaks a mechanical leg, then it might not be repaired in time to deliver medicine at the appointed time, thus failing in its task. Therefore, it has to build up a series of defensive mechanisms, in order not to get distracted in its primary task. One might conclude that, in extremis, the robot will have to kill every human being around the senior citizen, all of which could, in theory, become an obstacle for the robot to accomplish its task — especially those wishing to disconnect it!]
AIs don't need to be sentient and nevertheless harm humans without having any intention of doing so (think of jobs lost). 75 years ago, science-fiction author Isaac Asimov struggled with the same issues, and proposed that all robots had three laws embedded in their firmware, to protect both humans and themselves.
Then Asimov proceeded to write 39 stories on how to exploit flaws in those 3 laws...
We'll discuss Asimov's vision of the future and his thoughts on how Humankind would deal with these dilemmas.
Hosted by @Gwyneth Llewelyn as usual!