New research suggests that large language models are snobs, and we puny humans are the things they are snobbish about.
Spotted by Futurism, initially published in the Proceedings of the National Academy of Sciences, suggests that LLMs like GPT-4, GPT-3.5, and Meta’s Llama 3.1 have some serious and concerning antihuman bias. When asked to judge any content, models overwhelmingly pick the one written by another AI over one written by a human.
When AI models are shown a human-written summary and an AI-generated one, they tend to favor their kind. In tests, GPT-4 was the most notable offender, consistently favoring AI-generated output over human-authored work.
And while humans in the study also showed a slight tilt toward AI content, the AIs demonstrated a much stronger, unmistakable favoritism toward themselves.
Your Chatbot Bestie Is Biased Against You
The real-world implications are concerning. Researchers note that companies are already using AI to screen résumés, and their findings suggest that human applicants may soon or already be at a disadvantage, especially if they’re not utilizing AI to enhance their résumés.
That’s a whole new kind of systemic discrimination, one not based on race or gender or religion, but on whether you’re human or an algorithm.
The researchers warn that this AI-centric favoritism could widen the divide between the haves and have-nots, favoring people with the financial means and expertise to access advanced LLMs, essentially creating a class system where humans not utilizing AI are shut out of opportunities. If you’re applying for a job, you’re not only up against other people, but you’re up against a machine that prefers machine-made stuff.
Study co-author Jan Kulveit offers a solution: if you suspect an AI is judging your work, feed it through an AI first. Make the AI like it. Just don’t lose your humanity in the process. The robots will one day come for your job, but in the meantime, they’re the hiring managers determining if you’re robotic enough even to get the job in the first place.
The post AIs Are Playing Favorites, and Humans Are Losing appeared first on VICE.