Study: ChatGPT Holds 'Systemic' Left-Wing Bias

Improve the News Foundation -

Facts

  • A study by researchers at the UK's University of Anglia has found that ChatGPT contains a "significant and systemic" bias towards left-wing ideology.1
  • Researchers asked the artificial intelligence (AI) chatbot to take part in a survey based on politics in the UK, US, and Brazil from both an unprompted perspective and from a left-wing perspective.2
  • The paper concluded that, within the surveys, the AI showed favor towards Democrats in the US, President Lula in Brazil, and the Labour Party in the UK.3
  • Each question was asked over 100 times, with answers then placed within a 1000-repetition "bootstrap" data sampling method to improve reliability. Dr. Fabio Motoki, the lead author of the study, claimed that answers "reinforce[d] concerns" that AI could "replicate, or even amplify" pre-existing biases online.4
  • In December 2022, Elon Musk — who co-founded OpenAI, ChatGPT's creator — tweeted that the "danger" of training AI to be "woke" was "deadly." Musk left OpenAI in 2018.5
  • The news comes after the University of Washington's own study also concluded ChatGPT and other GPT-4 models to be the most left-leaning artificial intelligence chatbots, while also finding Meta's LLaMa to be the most right-leaning, and authoritarian, model.6

Sources: 1Sky News, 2The Washington Post, 3Nairametrics, 4Forbes, 5Tvpworld, and 6DIGIT.

Narratives

  • Narrative A, as provided by Ars Technica. AI can certainly contain bias, despite not being alive. This has the potential to become extremely problematic in the future, holding the possibility of portraying societal opinions as truth on a mass scale. Luckily, there is still time to control and amend the problem. Programmers and their vigilance hold the key to mitigating this and making sure that AI becomes both an efficient and objective tool for all to use.
  • Narrative B, as provided by Spectator (UK). The ramifications of developing AI technology are far too vast to even try to predict. But the positives are just as numerous as the negatives. AI could act to tackle loneliness, solve climate change, and address income inequality. Attempts to stifle the technology will not work, and versions of AI will regardless soon permeate the lives of billions. AI is likely to evolve and improve over time.

Predictions