AI now can fake empathy, blowing a way professional services giant KPMG
The wizards of KPMG’s local operations have given AI a brain, and now a heart, bolstering their courage as they plan to invest $US2bn during the next five years on harnessing the power of the much-hyped technology.
First there was artificial intelligence, now there is artificial empathy, with KPMG’s chief digital officer John Munnelly declaring the technology has “moved to another level”.
The consulting titan deployed its own customer ChatGPT-style tool six months ago, and the results have exceeded expectations, transforming the way the firm operates.
Its KymChat – which uses Microsoft’s AI platform – has drafted advice to clients, accelerated research, ensured greater compliance with environmental, social and governance compliance, and been able to answer day-to-day staff queries, taking pressure off the KPMG’s HR team.
And it was in one of those routine staff queries that left Mr Munnelly blown away.
“We’ve recently been loading our people policies (into KymChat) and the other day one of my guys is going away – his father was sick in Hong Kong, and he had to take some leave,” Mr Munnelly said.
“He asked KymChat, ‘I’ve had a death in the family, which leave do I take?’ KymChat came back to him and said ‘well this is the leave you take’ but first of all it started with ‘I’m sorry for your loss’.
“I went, ‘wow’. OK, we have now moved to another level in terms of technology.”
The response imitates a fundamental human response, revealing how quickly AI has advanced and what has fuelled its rampant spread across the economy. But technology that appears to empathise with workers stokes fears that it will assume more human roles, sparking the biggest disruption to jobs since the dotcom era.
Mr Munnelly says AI is indeed performing more of tasks that have been traditionally assigned to junior workers, such as preparing research for tax advice, and often meant working “long hours”. But he says this also creates opportunities.
“We’ve got to be very cognisant that learning by osmosis has traditionally been part of the professional services industry – juniors coming in, doing the sort of paper review, presenting the documents to the partner and the partner reviews it. If we’re automating loads of that stuff, they’re not going to learn that particular way.
“But it opens up slightly different roles. We’re talking about an AI librarian or AI curator, a new role in the business where you own the responsibility for curating the library of documents that it (AI) uses. They’re the sort of roles that AI will end up taking over.”
Generative AI, which can perform myriad tasks based on simple verbal prompts, is only as good as its input data. Models that are connected to the internet can be fraught, creating what is known as “hallucinations” or work that is worthless and error-prone.
“I keep saying to customers, don’t try to use it with the World Wide Web as the data source because you just don’t know what it’s going to pick up and reference,” Mr Munnelly says, highlighting the need for quality curated data.
This is despite the best efforts from Google, which unleashed its high-powered rival to ChatGPT last week but conceded it hadn’t eradicated hallucinations completely.
If the data is good, then Mr Munnelly says KymChat – which is based on Microsoft’s AI platform – can “do the first 80 per cent” of a draft.
“ESG (environmental, social and governance) is a good example or tax – we will be bringing out our Kymchat tax solution to our tax team in about February – and that’s using precedents database, so we could do a first draft.
“Now, it’s never going to do a complete document because you still need that human interaction. You still need someone to verify it. It’s not perfect.”
But it comes pretty close to perfection.
“We’re now starting to publish a KymChat confidence index. We’re testing it against a set of questions – like hundreds and hundreds and hundreds of questions – and it answered within 91.7 per cent confidence. So we’re giving people that confidence number on the tool so that they can understand it’s not 100 per cent. And it’s never going to be 100 per cent. But 91 is better than 60.”
Similarly, KPMG also produces a score when it bills clients, disclosing how much it used AI for a particular job.
“I want to make sure that we’re continuing to be transparent about it. That’s important, you have to say. And in fact, in some of our most recent documentation, we’re actually getting a percentage, how much of it was prepared using AI tool. You know, ‘54 per cent of this document was prepared using an AI tool’.
“So there’s a very explicit understanding. The next thing for us is … charge for that as a service because if it’s accelerating the time it takes to do these and perform these functions, we need to mix up our charging and our charging more needs to include technology as well.”
Globally, KPMG is investing $US2bn ($3.05bn) in AI and cloud services during the next five years as it expands its partnership with Microsoft. And it is expecting a high return on investment, with the technology expected to generate $12bn in revenue over the same period – annually that’s about 7 per cent of KPMG’s overall revenue, which totalled $US34.64bn in the year to September 30.
It comes as the firm appointed David Rowlands as its global head of AI last week and launched the “KPMG Trusted AI framework with the aim of ensuring a responsible and ethical approach to AI transformation”.
More Coverage
