The impact ChatGPT and large language models (LLMs) are having on the practice and profession of data science is something we discussed recently with data scientists from Unilever, BT, Deliveroo, and others. So, it was interesting to hear a perspective this week from Osama Rahman, director of the Data Science Campus at the UK Office for National Statistics.
Speaking Tuesday at an online event, Rahman mused on how LLMs brought both potential benefits and risks. For example, those who can already code can code more efficiently now with the help of LLM-powered tools. However, those same tools also allow non-coders to code – and inexpert use of tools and code presents risks. How do we guard against this, he was asked. “I don’t know,” came the response, “other than you have to observe and clampdown on it.”
Such problems are by no means new or unique to the post-ChatGPT era, of course. As someone with a background in economics, Rahman said he has, over the years, observed “inexpert uses of economics.” His advice was to “make sure experts are plugged in” – to teams, conversations, decision-making processes, etc. – “and are seen as the experts in the use of these tools.”
The discussion was wide-ranging, and also took in questions on whether data scientists have the right skills at this moment – “Skills evolve, it’s just a natural process… We need to keep a culture of curiosity…” – and whether enough is being done to address ethical issues – “My key issue is that ethical frameworks need a lot more discussion and debate than it takes to put out a new tool… I don’t have much to add, other than that there is a problem.”
However, two questions – and answers – jumped out at me as particularly interesting. Rahman was asked: Have we delivered on the promise of data science from 5 years ago? “No, but that’s because expectations were wrong,” he said. “Data science wasn’t going to completely and utterly transform government. But where it has delivered is in an evolving set of tools, people, and skills coming in and allowing us to do impactful stuff. It hasn’t delivered on the false promise that it would change the world, but it has delivered a lot.”
He was also asked: How will data science and AI have changed the world in 5–10 years? “I’m not sure it will,” he said. “It will do certain things. It will allow us to address certain analytical problems more efficiently.” Rahman then offered a salutary reminder. Email once made life more efficient; now, we’re all at risk of “death by email.”
We’ll be sure to update this post with a link to a video or other recording of the event, if/when it becomes available. For now, be sure to check out our two-part discussion on LLMs and data science:
- How is ChatGPT changing data science?
- Large language models: Do we need to understand the maths, or simply recognise the limitations?
- Copyright and licence
- © 2023 Royal Statistical Society
This article is licensed under a Creative Commons Attribution 4.0 (CC BY 4.0) International licence.
- How to cite
- Tarran, Brian. 2023. “What’s the future of data science and AI in an LLM world?” Real World Data Science, May 26, 2023. URL