Sexy To Strategic: How Gen AI Is Changing Data Science At AstraZeneca

Gen AI is lowering the bar to analytics, says chief data officer Brian Dummann.

Change is a constant for tech teams, but rather than being linear it’s a process characterized by sudden leaps.

Cloud was behind one such rapid shift, requiring new skills and devaluing old, the pandemic later putting a rocket under remote working and virtual collaboration. Mobile was the cause of another schism, and further back, the web forced a compete rethink of what IT could do and how tech teams should support it.

And so it goes.

Data science arose from the big data revolution, specifically the need to be able to process and find meaningful patterns in enormous datasets. So fierce was the competition for data scientists in sectors like banking and pharmaceuticals, that the role became known as the “sexiest job of the 21st Century”. Data scientists are still in high demand.

But now the wheel is turning again. This time it’s the turn of data scientists to see their roles change, said Brian Dummann, VP IT and chief data officer at AstraZeneca, speaking at a SnapLogic event in London earlier this month. And the driver is generative AI.

Buy Versus Build

The pharmaceuticals sector has been an enormous practitioner of machine learning and “traditional AI” for years, he said, but gen AI is now allowing non-data scientists to use AI capabilities without requiring them as an intermediary.

This is leading to a restructuring of the IT organization. And it’s the start of a big change.

“It's hard for people in leadership roles in companies to fully understand what generative AI is and why it's so different than all the other AI,” Dummann said.

Whereas in the past AstraZeneca would build its own AI/ML models and algorithms, with a significant input from data scientists, with gen AI it makes more economic, environmental and operational sense to buy a foundation model and fine-tune that, which is a very different process.

That foundation model can handle centrally many of the tasks that are used to require separate models and workflows. Trained to support multiple use cases it can facilitate a seamless self-service pattern with less need for technical intermediaries.

“You don't need a data scientist to hook into a lot of that it. Professionals can now tap in and build those into integrated solutions - or the data scientists can go faster with the work that they're doing.”

There are now more than 100 active use cases at AstraZeneca which would once have required data scientists in a central role, but where they are now being kept at one remove, able to get on with more cutting-edge tasks: “Those things that require that level of specialized training and knowledge that's allowed us to scale our organization.”

It’s not just data scientists who are having their job descriptions tweaked. Embedding AI into tools, either those created by AstraZeneca or via partnerships with vendors like SnapLogic, also “lowers the bar” on hard-to-find specialisms, and “allows us to bring in more versatile professionals, which gives us greater flexibility and better career paths, and the ability for IT professionals to move around,” Dummann explained, adding:

“We’re already see this changing the nature of job descriptions within our business.”

A New Socratic Age?

The end goal is to stay one step ahead of the market, he went on.

“We're always challenged as an IT organization to do more with less and drive optimization, and we've done a lot with RPA-type activity. But AI brings another level of automation where we can start to do a lot more things in productivity, not only internally, but also in how we engage with our partners across the business, which frees up capacity.”

Dummann continued: “Where we have to build something, we don't want to build from scratch. We did that for the last year-and-a-half, and I'll say it's hard, and it's difficult to also just keep the technical knowledge and depth. It's a rapidly changing space. The only way to keep up is to have good partners in this.”

Abstracting away some aspects of decision-making through AI agents is where all this is going, SnapLogic CEO Guarav Dhillon told Computing, creating a federal model where IT looks after the big stuff and users help themselves to the rest with AI assistance.

“What we see happening is the modern IT organization becoming invisible and ubiquitous,” he said.

Data scientists who might fear that “freeing up time to be more strategic” might mean being put out to pasture need not worry, he said.

“Not everybody's good with large language models and statistics. But if you understand how they work, those fundamental things will not go out of style. Asking good questions has always been a good skill to have,” Dhillon said.

“We think actually it could give rise to a new Socratic age, where if you can do the prompt engineering properly, you can pretty much solve larger and larger problems.

“And if you have a foundation in science, you can smell when it’s hallucinating.”

This article originally appeared on our sister site Computing.