BUSINESS NEWS e opportunities and challenges presented by tools such as ChatGPT. Ian Taylor reports
Productivity gains ‘will still rely on human evaluation’
It appears self-evident that generative AI should bring improvements in productivity, but that could depend on how it is used because, despite all the digital developments of the last 20 years, official data suggests UK productivity has stagnated. Deloitte’s Ellena Ronca-
Thompson insisted: “We can have productivity gains if we apply generative AI with discipline.”
Ellena Ronca- Thompson
productively, you start to realise productivity gains.” Rather than eliminate jobs,
Ronca-Thompson suggested generative AI-use would require new skills and new jobs. She noted: “The calculator was introduced decades ago. It still needs a skilled operator. “It’s important having a
She suggested AI tools such as
ChatGPT could “radically reduce the time required to generate reports” and said: “In my team, we use that extra time on research and development. When you give people time back and use it more
human in the loop when these models are generating tons of content. You can’t just push it into production and forget about it. “A skill set needs to come in
now around critical thinking, knowing the right questions to ask, understanding and being sceptical about an answer, being able to evaluate whatever answer you get.”
‘Incredibly powerful tool must be used responsibly’
There is a need for caution among businesses developing uses for generative AI tools such as ChatGPT. That is according to Google’s
Jay Chauhan, who told the conference: “This is an incredibly powerful technology that we need to roll out and scale responsibly. “We need to take a cautious
approach to generative AI. When we launch something, we need to get it right. We need to do it responsibly.” He suggested that when Google
launches something “it makes a huge difference”, adding: “We want to make sure it’s safe, [that] we’re observing consumers’ privacy, [that] we’re getting rid of any bias. It’s not a race to be first, it’s a race to get it right.” Deloitte’s Ellena Ronca-Thompson
noted: “We’ve developed a safety framework [at Deloitte]. We’re evaluating use cases and, for every application we decide to put into
travelweekly.co.uk
production, we’re clear on the limitations at the outset and what could be some of the implications.” For example, she said: “What
if a hacker tries to interact with a chatbot to access private company information? How do you put safeguards in at the beginning to make sure you don’t have to claw back a product or suffer reputational risk?” Simon Powell, chief executive of
travel technology firm Inspiretec, dismissed a suggestion that launching ChatGPT was irresponsible, insisting: “It’s hard to control or to stop this type of technology. There will be debates forever and a day about whether it is a good thing. [But] there was no way to stop this coming out. It was always going to happen.” However, Powell acknowledged
the need for human ‘curation’ of ChatGPT results, arguing the content produced requires “a human
Jay Chauhan
Simon Powell
to critique it”. He said: “Text needs to be looked at to make sure it’s relevant, that it’s correct and you’re happy to send it to a client.” Ronca-Thompson argued: “Just
because you can, doesn’t mean you should. My car can be driven faster than 70 miles an hour, but that doesn’t mean I should. I can add bells and whistles to an email, but how is that going to deliver anything to the bottom line?”
Deloitte director warns ‘models need to be trained’
The need for caution in using generative AI stems from the underlying structure. Deloitte’s Ellena Ronca-
Thompson explained: “You have the computer power which has made it possible, the data underlying it and the large language models consuming the data. Issues come up where the data underlying the model is inadequate or the model generates content that doesn’t exist. “What is not well understood
is that these models are probabilistic” – meaning they incorporate randomness. She said: “If I ask, ‘What was
my revenue yesterday?’ and ask the same question again in two minutes, the answer should be the same. But when generating content in a more commercial style, it can be different every time. We’re still having to train the models to get more deterministic answers” – meaning answers determined by the parameters and values set by those operating the model. Ronca-Thompson argued:
“Analytics teams are no longer a bottleneck. But how do we know the technology is going to be used as intended? You want to make sure you’re not getting model creep [or] data creep and you’re not introducing bias.”
Moderator Ian Taylor
2 NOVEMBER 2023 55
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64
