How could we implement a law that requires disclosure of an AI?

2 points by bawana 5y ago ↗ HN

When I communicate via telephone or video, I would like to get an honest answer to the question "Are you human or an AI"? With GPT-3 and deepfakes improving, the day will soon be here when we are confronted with virtual humans indistinguishable from the real thing - at least to our senses. And when we ask them if They are human, they will lie and say yes. Is there a deep learning model that can distinguish between a human and an AI? TO put it another way - can a deep learning model be developed to perform a Turing test based on vocal cadence, diction, grammar, accent as well as visual cues (when dealing with an image)?

2 comments

[ 3.3 ms ] story [ 15.7 ms ] thread

uberman 5y ago ↗

You might be interested in : https://ai.facebook.com/blog/deepfake-detection-challenge/

mindcrime 5y ago ↗

This feels like something that will just turn into a Red Queen scenario, much like the arms race between CAPTCHA's and AI CAPTCHA solvers. Not sure either side can ever "win" conclusively. It's definitely an interesting idea though, and certainly relevant given some of the stuff that we've seen lately using deep fakes (and/or related tech).