Artificial Intelligence - moral robots?

A fascinating article in The Economist 1843 magazine about teaching robots right from wrong and how AI is learning morality.

Now I’ve been a sci-fi nut from way back, and this is a well-visited topic – from Isaac Asimov’s Laws of Robotics, to 2001: A Space Odyssey, Blade Runner and more recently the thought-provoking film Ex Machina.

But, reading the article, I was struck by the thought that, in current AI development, Right versus Wrong seems no different from Correct versus Incorrect. Let me explain:

When creating RPA software bots to automate repetitive tasks, we instruct the bot to look for certain conditions – essentially IF this set of values exist, THEN perform this task. Meanwhile, Mark Riedl, a director of AI development at Georgia Tech, seeks to educate robots on how to behave in social scenarios, and his preferred tool is stories – crowd sourced from Amazon’s Mechanical Turk – describing a typical story on a given social interaction like going to a restaurant. His natural language processing algorithms look for similar sentences from different stories and then begin to draw conclusions about social rules and norms.

Not a million miles then from an accounting RPA bot which learns that, at financial year end, it allocates more Cloud resource for specific tasks, and maybe only runs some tasks at that time. For social rules and norms, read business rules and norms.

Of course, in neither case is there yet AI sentient self-awareness, or even a sense of remorse or guilt if things go wrong. And go wrong they always will – as the author of the article acknowledges, we’ll never have accident-free autonomous driving technology, so our goal is just better than humans, but not perfect.

Purely personally, I look forward to the day when an AI system tells me that it’s “seen things you people wouldn’t believe”. But “I’m sorry John. I’m afraid can’t do that”?… not so much.