In this scenario, the 2028 elections have already passed, but the discussion isn’t about whether the Democrats, Republicans, or another party won.
Rather, the question is whether that administration is using AI to develop certain policies in the economic, scientific, technological, transportation, military, tax, health, or other sectors. Would you accept that administration using these tools, or would you prefer that humans make those policies?
this is kind of a weird one, because I think LLMs generally would natively recomend above average policies, but you know for a fact that the people currently in charge would train this specific LLM and feed it data and configure it’s memory/context/prompt are going to massage the inputs such that it’s essentially replacing “all the rich white guys decided to give Israel 999 billion dollars to Israel, sorry buddy, that’s democracy” to “sorry buddy, computer says, we gotta give 999 billion dollars to Israel”.
It would be a waste of resources to try to switch to an LLM who would be in agreement with that administration’s policies.
I feel it would be more accurate to say they’d use the model for advice, then apply their human touch-up. But it would be quite ironic if the successful policies were the ones proposed by the AI and left unchanged, while the modified ones were a complete failure.
I feel like People In High Places having AI psychosis (Thiel famously) opens the possibility of LLM companies giving specific individuals pre-determined answers. Allowing them to get more no-bid contracts, make extra on Kalshi, and exert undue influence without having to go through traditional methods of corruption/patronage networks in the U.S. system.
Hegseth’s DoW using ChatGPT for target identification in Iran shows how military operations could be easily influenced, for instance.
As others have said, it also allows blame to be thrown elsewhere. Funny how AI got so pushed right after Epstein. Any photographic evidence is suspect now.
The humans are making those decisions whether you use AI or not. There’s lots of bias in training any form of model, and many people doing the training dont likely prioritize. Yes there is actual training bias like leakage and sample distribution issues that everyone knows, but the implicit biases are far more insidious. Look at all the facial recognition software or the countless prison ml/ai applications that are ridiculously biased to minorities. AI is no better and arguably worse. Are you going to ask the AI how it got to its answer? Because it can’t tell you about its little weights.
“AI” boosters are overwhelmingly techno fascist arch reactionaries, especially in “the West,” so no, I would not like this
People are already asking chatbots how to vote, so we’ll get whatever govt the bot vendors want to install. The bots will reinforce their priors.
It depends on which model they’re using. If it’s a GPT-4o or Claude 3, I’d say abort the mission before you cause a major disaster. But if they’re more powerful models than the Claude Fable/Mythos 5, GPT-5.6 Sol Ultra, or Pro, I’d be curious to see.
