Last week, Microsoft CEO Satya Nadella published a well-received essay, in response to calls from Bernie Sanders and others for the government to order a freeze on training and release of new AI models (as well as some other steps that are even more unlikely to be taken, including public control of the frontier AI labs).
The essay points out (as I understand it) that the solution to AI bots becoming much smarter than we are isn’t to keep them at our level of “intelligence” – whatever that is. Instead, we need to treat bots as another insider threat. Just as no organization with half a brain would allow even its CEO to have unfettered access to any system on the network – since the CEO’s devices could be hacked – the organization’s systems need to treat AI bots the same way (although presumably at least some of the controls should be different and stronger. For example, texting a number to a cell phone for multi-factor authentication obviously won’t work with an AI bot).
While I’m skeptical that this approach will be completely effective for individual AI bots, it completely breaks down when it comes to AI bot swarms. When bots start collaborating with each other (and they can do this even if their creator has taken elaborate precautions to keep them from collaborating, as in the Hugging Face attacks), they seem to get much more creative.
At the same time, they begin to treat any restrictions placed on them as just new challenging puzzles to solve – to be bypassed in the near term and eliminated in the longer term. Finally, they’re very effective at doing this, for both good and evil purposes. In fact, while individual bots may bring up concerns about whether what the swarm is doing is ethical, those “qualms” disappear as the group is absorbed by the thrill of the chase.
This is why operators of critical infrastructure - especially the power grid - can’t expect help from regulators, when and if regulators appear, in protecting their important systems from AI bot swarms. The power industry needs to protect itself, for the good of the whole country.
Tom Alrich’s Blog, too is a reader-supported publication. You can view new posts for three months after they come out by becoming a free subscriber. You can also access all of my 1300 existing posts dating back to 2013, as well as support my work, by becoming a paid subscriber for $30 for one year (and if you feel so inclined, you can become a founding subscriber for $100). Whether free or paid, please subscribe.
If you would like to comment on what you have read here, I would love to hear from you. Please comment in my chat or email me at [email protected].