ChatGPT is an AI (artificial intelligence) chatbot launched by OpenAI that makes users think they're talking to a human, and it's the newest technology that's taking the internet by storm. It has an endless list of possibilities from writing poetry through to computer code and apparently can also pass an MBA course exam. Users are able to ask the chatbot questions and answers are based on the vast amount of online data that ChatGPT is trained on.

It's easy to see why this has gone viral since it's launch in November 2022, yet there is a darker side that many may not be aware about...it's yet another example of AI that's rife with bias. OpenAI have added in some guardrails in order to help the chatbot navigate it's way around users asking the chatbot for example to say a slur or commit a crime, however it's still been far too easy for users to get round these guardrails by rephrasing their questions, which in some cases has prompted responses from the chatbot with very questionable and sometimes highly discriminatory language!

As the world relies more on technology, AI is being used to make vital decisions in a variety of sectors, but biased algorithms mean existing inequalities could be heightened — with dangerous results. As AI becomes more commercialised and wide spread, why is it that little is still being done to prevent these biases?  A study in 2019 (Science, 25 Oct 2019, Vol 366 Issue 6464 p.447-453) showed that algorithms used in a hospital recommended Black patients receive less medical care than their White counterparts. In 2018, Amazon shut down its own AI recruitment tool because it discriminated against female applicants. Galactica (similar to ChatGPT) which was trained on 46 million text examples was shut down by Meta after just 3 days because it spewed out false and racist information.

AI models have to be trained on ginormous amounts of data in order to be able to learn and act like humans. ChatGTP is trained using text databases from the internet including 570GB of data obtained from books, webtexts, Wikipedia, articles and other pieces of writing. The issue here is that large datasets that are scraped from the internet are full of biased data that then informs the models. Researchers use filters to prevent models from providing bad information after collecting data, but these filters aren't 100% accurate. This can result in the expression of harmful biases, like when ChatGPT told users it would be okay to torture people from certain minority backgrounds! Then there is also the bias of researchers in AI, which is a highly homogenous field dominated by white people and men, who decide what data to feed their models.

It's sad to see that there is a disconnect between the lack of ethical concern in the AI industry from start ups looking to make money and those academics raising these concerns. In a world where AI is being used more and more, this is an area that needs stronger focus and attention to prevent future harm. In the meantime, it's good to at least be aware and mindful about these biases that are still very much present in the technology we use, so we can reflect and question where needed.