Commonwealth Bank treads carefully on ChatGPT bias

The Commonwealth Bank is actively assessing the risk of experimenting with or using conversational AI platforms such as ChatGPT, including to what extent its outputs are reliable.

Chief data and analytics officer Andrew McMullan told an Infosys and Trans-Tasman Business Circle event that he - and the bank more broadly - is highly “cognisant” of the challenge of understanding what generative AI is drawing from to produce answers to questions.

 "You do need to be cognisant of what AI is learning from and the data that it uses to come up with the answers," McMullan said.

“Everything that's ever been digitised and all the 'ChatGPTs', they're learning on data that could be potentially biased in terms of when it was written, and the narrative that was around at that time.

“From an organisational point of view at CBA, we're very focused on making sure that we understand completely the use cases. We have our own policies [for experimentation and use]."

McMullan said one of his roles within CBA is to act as an AI risk steward to “make sure that the principles and the policies and every single use case of AI is well considered before we would do anything with it.”

“I always believe that the use of AI or any advancing science or technology is a representation of the culture of an organisation," McMullan said.

Earlier this month, CBA told iTnews its AI Labs research team is "actively engaging around recent advancements in generative AI, including ChatGPT and similar technologies that are emerging.”

The bank is exploring use cases that could augment its customer service offerings, an area it has invested in for some time, notably with its customer engagement engine (CEE).

“For us, the use cases that we have are very focused on our purpose of trying to support Australia and Australians to have a brighter future, and the areas that we're really focused on [are] service," McMullan said.

“If you think about every time you call a call centre or you go into a branch, how much easier can we make it for frontline staff to serve customers and get the answer to the question that customers have?”

Outside of customer service, McMullan said AI had also proven useful for monitoring IT sysatems and reducing duplication of data.

He also said the bank continues to invest in academia and skills to work with AI and data.

“As we see the shift in the skills that are needed to be able to serve our customers brilliantly every day, we're training people who used to work in frontline jobs, to be able to come and engineer with us or become analytics professionals, or even get involved in machine learning and AI if they've got the desire to kind of reskill in that area," McMullan said.

He said he is “excited about what we can deliver over the next 10 years using AI,  the advancements in machine learning” and as “society becomes more digitised, I can see great capability being available to support the experience of our customers on Australia as a whole.”