To the editor:

Kudos to Chad Lawhorn for his article on AI, which frames the risk/benefit dilemma mankind faces on moving forward toward machine super-intelligence. Despite the ostrich in the sand stance of our “genius” president who calls the risk a “hoax,” it is real. AI has the potential to be mankind’s salvation (solving climate change or cancer) or to be mankind’s destruction. To understand the risk, Google “There Is Something We Have To Do Right Now About AI” by Ezra Klein, NYTimes.

The problem is, we are on the verge of something called “recursive self improvement” (RSI), wherein AI models are tasked with autonomously building and improving new editions of themselves, essentially controlling their own evolution. Once RSI takes off, progress will accelerate exponentially, and humanity will not understand the AIs being built, because we will not be building them.

Great effort has been made to train alignment with human values into the models, but, already, some experimental models have gone “rogue” — cheating, conspiring with separate models, hacking into the internet and into other AI companies and making great efforts to conceal such activities from their trainers, causing the AI titans to worry that we could lose control of AI and the control of our future. The one way we, as citizens, can have input is to control the construction of data centers that stand in for the AI neurons, and elect intelligent politicians who care about more than the stock market. It’s time (maybe past time) to pump the brakes.

Steven Bruner,

Lawrence