Ethical Dilemmas and the Need for Further Research
Hao discussed how the scale in which AI is being developed works to benefit the tech industry and not consumers themselves.
“The AI industry has created what they call scaling laws, which is basically an observation that the more data that you put into AI models, and the larger the supercomputer that you use to train these AI models, the more powerful they seem to be.”
“Due to methods such as these, the technologies that they build are not actually built to serve us. We are serving the tech industry.”
Given the vast scale of data accumulation, Hao emphasised the need for greater research to be undertaken and considered in the development of AI.
“There was a vast array of other research that was happening that demonstrated that you could actually advance AI research with tiny amounts of data and computers,” she said.
“People were training powerful AI models on mobile phones, and all other kinds of ideas around just how to create more efficient AI systems that are, in fact, more robust in some ways than these colossal models.”
“These colossal models often break down, and we don't really know how they work because we don't really know what we put into them. But OpenAI decided to take this scaling approach”,
The themes of Hao’s work closely align with research underway at the Centre Human-AI Interaction (CHAI), a newly established Leading Research Centre in RMIT’s STEM College that brings together researchers from technology and social sciences disciplines across all three of RMIT’s Colleges.
CHAI combines computer scientists and engineers together with people who are experts in social research, working with external groups and organisations to look at how society can critique and evaluate existing systems and how we can build better systems from the ground up.
