Being smart, clever and fast, and manipulating the technology would elevate the service level in the business, McElligott continued.
The big AI risk was of ignoring compliance or failing to prepare for compliance, the MHC partner said.
This was the year for firms to get going on assessing high risk and prohibitive risk, he stated.
Crunching data
AI was particularly deployed in credit scoring and fraud detection, crunching a large amount of information and data to provide recommendations on giving or withholding finance, he added.
AI is very useful in tracking how credit cards and online payments are used.
Strong use in various customer-service settings sees AI prompt human agents to give answers to consumer queries on previously requested information.
Standardised responses were a massive benefit in these cases, McElligott said, and the AI also provided bullet summaries after the conversation concluded.
He warned, however, that, while AI tools were sophisticated and powerful, they also must be trained with a relatively simple overarching training architecture.
McElligott stated that AI was “not going to come and take over the world in the morning.”
Do not put your trust in AI outputs, McElligott added, because, despite their power, “they cannot do magic”.
Sometimes, in the absence of getting the right answer, AI gave any answer, he said,
Rebuked
He referred to the New York lawyers rebuked by a judge for submitting proceedings in which they copied and pasted AI output, citing a case law that never existed.
“None of us in our wildest dreams would cut and paste something out of Google, for use with internal or external clients,” he stated.
“Obviously, the AI should be treated in the same manner,” he said.
There are also known privacy issues with the technology, which should be scrutinised before using AI.
“There will be issues with regard to the liability for the outputs of these machines,” he commented.
He pointed to a chatbot on an Air Canada website which gave wrong information, directly contradicting the business website, on making a claim for a bereavement air fare.
This was successfully challenged, so it could not be presumed AI simply ingested what was already on a website and helped customers navigate that information, he said.
Unintended consequence
“Here you have an unforeseen, unintended consequence,” he said, and liability for mistakes must be considered before taking on the technology.
AI legislation provided for fines greater than those under GDPR, and this showed that the EU was serious about regulation, he stated.
Given the timeline of the act, firms should this year be gathering information on all the AI systems in-house and reporting the use to which each of them is put, because that was how the legislation worked, he continued.
“I would recommend putting in place responsible AI use policy – most people are doing that at this point and it's not a lot of work. Then, finally, track proposed use of AI as you bring it in and update existing tools,” Brian McElligott said.