Inside Darpa’s Push to Make Artificial Intelligence Explain Itself

Re:Inside Darpa’s Push to Make Artificial Intelligence Expla... by danschumann on Thursday August 10, 2017 @06:56PM (#14985133)

As long as it doesn't realize it's easier to lie to humans about what it's ACTUALLY doing. Maybe make the explainer algorithm is non-ai-based, to ensure it doesn't optimize out honesty.

[ Parent | on HN ]

Re:Inside Darpa’s Push to Make Artificial Intelligence Expla... by nradov on Thursday August 10, 2017 @10:02PM (#14986083)

It seems a bit unfair to hold AI to higher explanatory standards than our fellow humans. Most people make decisions based on intuition and emotion. Then they rationalize those decisions after the fact and invent explanations that seem plausible.

[ Parent | on HN ]

Re:Inside Darpa’s Push to Make Artificial Intelligence Expla... by Retra on Thursday August 10, 2017 @07:26PM (#14985309)

Couldn't you just compartmentalize? Make an outer AI that has to defer to an inner AI, and nest them in such a way that the only interface between the layers is a 'explaining' communication channel?

[ Parent | on HN ]

Re:Inside Darpa’s Push to Make Artificial Intelligence Expla... by therein on Thursday August 10, 2017 @10:00PM (#14986074)

That reminds me of this video and the left/right brain compartmentalization.

https://www.youtube.com/watch?v=wfYbgdo8e-8

[ Parent | on HN ]

Re:Inside Darpa’s Push to Make Artificial Intelligence Expla... by nl on Friday August 11, 2017 @02:18AM (#14987055)

This is unnecessary anthropomorphizing.

Explaining how an AI system makes decisions is completely unrelated to any kind of progress (or not!) towards AGI.

[ Parent | on HN ]