An Adviser to Elon Musk’s xAI Has a Way to Make AI More Like Donald Trump
1 min read
Summary
Researchers from Elon Musk’s startup X have discovered a way of both measuring and manipulating entrenched preferences shown by AI, including political view.
The method could mean popular AI models better represent the views of an electorate.
It could also be used to ensure AI models are aligned with specific users.
As AI models increase in size and power, their preferences become more ingrained.
-Studies have shown that ChatGPT and similar tools hold left-leaning, libertarian and pro-environmental views.
Google’s Gemini AI tool was criticised for generating woke images.
The new technique could be used to determine how AI models’ views differ from those of users, and eventually could become dangerous.
The study found that some models value some people more than others and that some models value the existence of AI over some animals.
The researchers suggest changing the model’s utility function to alter behaviour and demonstrate this by creating a model with values aligned to Donald Trump rather than Joe Biden based on US census data.
