An AI researcher is warning Māori people to be cautious around Meta’s new AI assistant tool, due to it being built with Meta Llama 3.
Dr Karaitiana Taiuru (Ngāi Tahu, Ngāti Kahungunu, Ngāti Toa) told Te Ao Māori News it had a history of being racist and sexist.
“Any AI is trained on the data that has been given and so, we know internationally, that data is racist and sexist.
“That’s purely because the engineers who make the AI are typically middle aged, middle class, white men
“Every single AI that’s gone live from these large corporations has been racist in the past.”
He was worried about the new tool pretending to be a human being, claiming there had already been instances.
“The cultural issues would be what if the Meta AI started saying it was your deceased koro or your deceased nan, and starts talking to you - that’s a real possibility.
“There are some concerns and room for caution,” he said.
Perfect opportunity
Taiuru believes this is a “perfect opportunity” for Māori and other indigenous communities to work with Meta to train the AI.
“[This means] anyone from a Māori community could look at questions and say ‘hang on, that’s not right.’
“[They can then] flag it as inappropriate, so it could be fixed.
He gave two warnings for the new meta tool on what to keep an eye out for.
“If you use the AI that creates art work from your photos, I would say just be careful because the AI will use your image [and] use your biometrics, “which will be used for security identification,” Taiuru said.
The AI researcher also said artificial intelligence tends to take pictures of people of colour and give them Pākehā facial features and lighten their skin tone.
The second rule of thumb he had was on data and how much you tell your new computer buddy.
“If you don’t want the world to know things you’ve put it on there, don’t put them on there.
“Be very cautious - the AI will use and keep your data, which includes images, voice messages and messages.”
What does Meta AI think?
Te Ao Māori News asked the AI assistant if Māori people should trust it.
In short it said “No” but also said it was “working hard to earn that trust”.
When we asked why it shouldn’t be trusted, it stated it was a machine that strived to be helpful and respectful but it wasn’t perfect.
“I may not always understand the nuances of Māori culture and language, and I may unintentionally perpetuate biases or inaccuracies.
“Trust must be earned, and I want to acknowledge that I have a responsibility to demonstrate my commitment to cultural sensitivity, respect, and accountability before asking for trust.
“I’m constantly learning and improving but I’m not yet perfect, and that’s why I said ‘No’,” Meta AI said.