I’m currently reading a chapter in Nick Bostrom’s book “Superintelligence: Paths, Dangers, Strategies” about giving AI values. While on the surface this may seem straightforward. It is anything but. Aside from the obvious questions like “who’s values?”.. where do you even start when it comes to programming them? One of Bostrom’s favorite illustrations about the dangers of AI is that if you instruct an AI to “make people happy” it will very recognize that the source of happiness is a chemical process in your brain and you’ll have wires sticking out of your skull and a silly grin on your face. What is happiness anyway? Love? Joy? If poets and authors can’t fully grasp these things how can an programmer possibly hope to build AI that can “maximize” these things. This is why more often then not AI is seen as a threat. How could it possibly be expected to understand and accept humanity? Bostrom poses some interesting possible solutions. One jumped out at me the other day when I read about Elon Musk’s belief that we most likely live in a simulation. One of the possible ways of instilling values in AI is through simulation. Basically you make millions, if not billions of versions of the AI and through a selection process.. or evolutionary process.. pick AI that exhibit traits that you want to keep.. and toss the rest.
The obvious question to Elon Musk about the simulation we are living in would be “Why?”. Who is running the simulation and what is the purpose? Well, if you view people as simulated bits of software.. perhaps those with desired traits are “harvested” while the others are tossed. Strangely or not so strangely this falls pretty close a Christian perspective that there are beings outside of our reality/simulation and when our software/hardware ends here.. it continues on either in a “better” reality or.. worse.
These thoughts apparently aren’t just mine as these seem to be echoed in this book: Your Digital Afterlives: Computational Theories of Life after Death (Palgrave Frontiers in Philosophy of Religion)
1 reply on “Giving values to AI”
Geoff, seems there is a real need for serious theological reflection on the issues you raise on this blog. I could see a series of essays by some youngish Christian (and Jewish) thinkers on this.