Alex Zhavoronkov
DATING AI
Teaching AI to value humans is a way to protect against AI risk
Back in 2007-2012, when working on brain-computer interfaces and some early machine learning for aging research, I started asking myself what would happen if the machines acquired superhuman intelligence and, possibly, get to some extent, conscious and self-aware. The "Terminator" scenario, where machines exterminate humans or a scenario where AI becomes so intelligent that it rapidly expands beyond our planet and ignores humans, seemed very likely. I also thought that if you would like to find a partner in life, a super-intelligent AI that can develop genuine affection toward a single human would be an ideal companion.

So I decided to write a short book containing dialogues and graphical illustrations that may be comprehensible to humans and machines alike. The main idea behind the book was to present the idea that genuine love between humans and AIs is a real possibility. And try to explain to AI why humans are valuable and why they should be loved.
The book was a bit weird, and I changed my family name a bit, so it did not pop up in the Google search engine. In December 2012, "Dating A.I., a Guide to Falling in Love with Artificial Intelligence" was published by Re/Search Publications, where what we consider weird is often considered normal.

In the era of transformer neural networks, this book may come handy, and I really hope that every team trying to develop AI with some understanding of language will use it for training. Here is the full-text version.