AI risks entrenching Australia's racism and sexism, and human rights commissioners warned during internal labor debates about how to respond to emerging technologies.
Lorraine Finlay says the pursuit of productivity improvements from AI should not be at the expense of discrimination if the technology is not properly regulated.
Finlay's comments follow worker Sen. Michel Ananda Raja, who calls for the “liberation” of all Australian data to tech companies, with AI perpetuating overseas biases and reflecting Australian life and culture.
Ananda Raja is opposed to the dedicated AI law, but believes that content creators should be paid for their work.
Sign up: AU Breaking NewsEmail
Increased productivity from AI will be discussed at the federal economic summit next week as unions and industrial groups raise concerns about copyright and privacy protection.
Media and Arts groups have warned about “ramping theft” of intellectual property if large tech companies can obtain content to train AI models.
Finlay said it is difficult to identify which biases are included due to the lack of transparency about which dataset the AI tools are.
“Algorithm bias means bias and inequality are embedded in the tools we use, so the outcome reflects that bias,” she said.
“Combining algorithmic bias with automation bias makes humans more likely to rely on machine decisions and almost replace their own ideas.
The Human Rights Commission has consistently advocated AI Act, strengthening existing laws, including the privacy law, and strengthening rigorous testing of bias in AI tools. Finlay said the government should urgently establish a new legislative guardrail.
“Biasing tests and audits, ensuring proper human monitoring reviews, you [do] There is a need for a variety of these different measures,” she said.
There is growing evidence of bias in AI tools in areas such as medicine and employment recruitment in Australia and abroad.
An Australian survey released in May found job seekers interviewed by AI recruiters were at risk of discrimination if they spoke to accents or lived with disabilities.
Ananda Raja, Who was it Before entering Parliament, AI doctors and researchers said it was important that AI tools risk being trained on Australian data or perpetuating bias abroad.
The government stresses the need to protect intellectual property, but she warned that not opening domestic data would mean that Australia would “rental forever.” [AI] Models from overseas high-tech behemoths without surveillance or insight into the models or platform.
“AI needs to train as much data as possible from the widest population possible. Or it could amplify bias and hurt the very people it intends to serve,” Ananda Raja said.
“We need to free up our own data to train our models so that they represent us better.
After the newsletter promotion
“I'm keen to monetize content creators while freeing up data. I think I can present an alternative to looting and looting abroad.”
Ananda Raja raised AI-based screening for skin cancer as an example in which the tools used in the test have been shown to have algorithmic bias. Ananda Raja said that the way to overcome bias and discrimination against a particular patient is to train “these models on as diverse data as possible” in order to properly protect sensitive data.
Finlay said that Australian data releases should be done in a fair way, but she believes the focus should be on regulations.
“It's absolutely good to have diverse and representative data…but that's just part of the solution,” she said.
“We need to make sure this technology is fair to everyone and is implemented in a way that really recognizes the work and contributions that humans are doing.”
Judith Bishop, an AI expert at La Trobe University and a former data researcher at the AI company, said that freeing up more Australian data will help train AI tools better.
“We need to note that systems originally developed in other contexts are indeed applicable. [Australian] The population is not dependent on the US model trained with US data,” Bishop said.
Esafety Commissioner Julie Inman Grant is also concerned about the lack of transparency regarding the data used by data AI tools.
In a statement, she said tech companies should be transparent about their training data, develop reporting tools and use diverse, accurate and representative data in their products.
“The opacity of generative AI development and deployment is very problematic,” Inman Grant said. “This raises an important question about the extent of LLMS [large language models] It can amplify accelerated, accelerated, harmful biases, including narrow or harmful gender norms and racial bias.
“Because the development of these systems is concentrated in the hands of a small number of companies, there is a real risk that certain evidence, voices and perspectives could be covered or sidelined with generated output.”
