Hi Chat-GPT, Who should I vote for?

How many people do you think might ask this question in the next few years?

Some estimates suggest AI now handles up to 250 billion queries a month. We know around 1 in every 5 uses of AI are for search so that’s 50 billion searches a month.

We can reasonably expect 100s of thousands (if not millions) of people to consult AI about which way to vote in elections.

This is a gamechanger - and it hands enormous political influence to the people who design AI tools.

So how do AI engineers and ultimately AI tools learn about politics?

I found the job advert below interesting. It’s a recruitment company looking for people who can help to ensure that an AI product properly understands politics. There are adverts for people from different parts of the political spectrum.

It looks like the company in question wants to find people who can independently evaluate the answers given by an AI chatbot for political accuracy and neutrality.

It’s good that they are thinking about this - but is it good that it happens like this? Up to $70 an hour to shape how millions of people engage with politics?

We know that many LLMs don’t tend to prioritise trusted information sources, instead drawing on sites like Trip Advisor, Reddit and YouTube when answering questions from their training data.

This means that if you wanted to influence how people see politics you could build a huge website (perhaps even using AI) and train other AI tools on that data.

This is what Elon Musk is doing with Grokipedia which, if it works, could end up feeding into the answers hundreds of millions of people get when they ask AI about politics.

I find the advert interesting because it shows what a low value these companies are putting on political neutrality. They pay their engineers millions of dollars a month but the people that get the politics right - they can work for a modest hourly fee.

Tell me again whether you think we need to take more steps to protect democracy?