ASIC tests AI usage to read public submissions
ASIC chair Joe Longo has told a Senate select committee that it has been running two pilot AI programs, including one to read public submissions.
Appearing before the Senate Select Committee on Adopting Artificial Intelligence today (21 May), the panel discussed how the corporate regulator is exploring AI and its potential regulation.
Asked if the regulator utilises AI in its own operations, Longo said it has been running two pilot programs.
“We do use AI, the first is in a pilot Commonwealth program on the use of Copilot and the second is using AI technology to read submissions. In ASIC’s world we are often having to read and absorb submissions because we consult heavily with the market so we ran a pilot to see if the AI could ‘read’ all those submissions and come up with accurate analysis to save hundreds of hours of human time.”
Graham Jefferson, digital and legal transformation lead at ASIC, added: “We’ve already been using machine learning for longer than AI has been topical and more recently we have been running these pilot programs.”
The Microsoft Copilot program involved 150 ASIC employees using it for a month, then providing feedback on how it worked for them. Meanwhile, the large language model pilot tested AI summaries of submissions against human counterparts, with submissions coming from those made to a parliamentary joint inquiry into consulting companies. This involved the use of the two US programs: Amazon Web Services and Llama2 which is run by Meta.
Jefferson said: “The pilot the chair mentioned was a standalone large language model to test whether generative AI could produce accurate summaries of submissions. The pilot was a success, but it is not something that we’ll be using going forward.
“We did a comparison of what the large language model generated and what our staff generated in a blind comparison and the results were not sufficiently good for us to use that technique in that particular way at this stage.”
Expanding on this, Longo added: “The AI gave a bland summary of the submissions, it wasn’t necessarily misleading, but it didn’t capture what the submissions were saying while the human was able to capture and pull out nuances.”
“We were particularly interested in any mentions of ASIC in the submissions to the inquiry, but the AI summaries were quite generic and the nuance wasn’t there compared to when the summarising work was carried out by an ASIC employee," concluded Jefferson.
AUTHOR
Laura Dew
Recommended for you
A financial advice firm has been penalised $11 million in the Federal Court for providing ‘cookie cutter advice’ to its clients and breaching conflicted remuneration rules.
Insignia Financial has experienced total quarterly net outflows of $1.8 billion as a result of client rebalancing, while its multi-asset flows halved from the prior quarter.
Prime Financial is looking to shed its “sleeping giant” reputation with larger M&A transactions going forward, having agreed to acquire research firm Lincoln Indicators.
An affiliate of Pinnacle Investment Management has expanded its reach with a London office as the fund manager seeks to grow its overseas distribution into the UK and Europe.
Needed to add - "consult heavily with the market " ... but don't publish or listen to submissions if they differ from our pre-determined viewpoint.