Looking to curate a small-scale event to spark discourse that filters through the ecosystem, a cherry-picked vertical slice of Scottish tech and business took in The Data Lab’s AI for Good Event. 

Two speakers observed the use-cases for AI for both a business and public ‘good’ lens, punctuated by the organisation’s CEO, Brian Hills and MSP Daniel Johnson. 

The former stressed The Data Lab’s need to “educate people on using the right tools and the right technology to solve the problems they’re facing, as well as really understand those problems to begin with.” 

The latter, MSP Daniel Johnson, took an aspirational view on AI’s potential to “do things more efficiently, more effectively, enhancing the skills we have in our workforce, rather than replacing them”, particularly within the context of Scotland’s “three great economic challenges: technology disruption, climate change and changing demographics.”

Brian Hills, CEO of The Data Lab

These were essentially preambles to the evening’s two main speakers. The first being Liz Whitefield, executive director at Hippo Digital, who viewed AI through a refreshingly cautionary lens; followed by Dr. Elizabeth Fairley, chief operating & data officer of Talking Medicines, who conversely, provided a compelling case study of how LLMs can tangibly be used in a way that’s effective, while remaining cognisant of the confluence of regulation and modern data ethics.

Be honest, does your business actually need AI?

Liz Whitefield, executive director at Edinburgh-based digital services partner Hippo Digital, coming from a storied background in digital transformation, pushed back against the flood of businesses scrambling to throw AI at any data sets or business process they conceivably can.

While certainly not a new notion, Whitefield’s assessment is a pragmatic one in that we “shouldn’t start with data… start with the problem you’re trying to solve, what are our objectives? And what role does this play? That’s really fundamental, before you start getting caught up by the shiny and exciting stuff.”

In Whitefield’s view, data literacy and good data practices are fundamental to solving issues within your business using what you already have.

She says: “It’s about data being a fundamental part of how you solve problems within your business and is that data sitting in the most effective part of your business in terms of your delivery – it needs to be part of your HR strategy, your operational strategy and is it integrated as part of your technology strategy.”

Liz Wakefield, executive director, Hippo Digital

Asking these questions will often throw up roadblocks as to how you can effectively integrate something into LLMs.

If robust data is not already forming an integral part of your strategy at the key delivery layers of your business, then it would be a fallacy to expect to be able to leverage any potential benefits of AI in a way that makes it worth the cost of integration.

To qualify this point, Whitefield gives a salient example around NHS England’s online patient records – vaunting the incredible “data story” around the 50 million NHS account holders who can access records while allowing 100 different services to access their accounts, providing a “transformational” ability for the NHS to exact data governance in a way that delivers positive outcomes.

Conversely, however, Whitefield points out that for NHS England, there are 15 million incorrect registered addresses.

“That’s 15 million citizens not getting letters, not getting invites to screenings – by setting data standards, by strong data governance, we can cleanse that data.

“Beyond that, with the same data, if we apply machine learning, around how people use services, how long people have been waiting for services etc, our data scientists can model on demand.”

These points illustrate Whitefield’s stance that good data governance, driven by the need to solve an inherent and tangible problem, creates a solid platform in which machine learning and other technologies such as LLMs can have a positive impact on outcomes.

Using AI Ethically and on the Right Side of Regulation

Dr. Elizabeth Fairley, chief operating & data officer of Talking Medicines – the Glasgow-based company that offers dynamic healthcare intelligence, parsing data pulled from social media to better understand medical discourse actual patients are having – presented an alternative view. 

If Whitefield’s perspective was one more akin to a cautionary tale for companies trying to retroactively fit AI around an existing business model, Dr. Fairley presented a case for AI done as a ‘best case’ – it solves an inherent problem, has been through the rigour of regulatory process, and is keenly mindful of the messy data ethics that comes along with LLMs as we know them.

“From day one, we worked very closely with regulators,” says Dr Fairley.

Talking Medicines sought to understand how patients engage with medicines and how they can optimise messaging to patients. 

To achieve this and provide better outcomes for patients, there has to be an understanding of the kinds of conversations people are having about their healthcare – social media naturally being a rich vein of discourse in this regard.

“There’s a lot of data analysis that takes place in a clinical environment,” says Dr Fairley. “For us, we were keen to see the non-clinical, conversational aspect.”

Recommended

• OpenAI Changes Data Controller in Bid to Adhere to GDPR
• OpenAI Faces Lawsuit Over Potential ChatGPT GDPR Violations
• EU Creates ChatGPT Task Force to Keep Up with AI Developments

You don’t have to be a data scientist to know that there’s an astonishingly vast amount of unstructured data on social media, and this is where Dr. Fairley and her team observed the opportunity to leverage machine learning and AI technology.

On this, Dr Fairley says: “[We used machine learning and AI] technology to help us understand how we can begin to classify that huge amount of data and structure it.”

It’s important to note that while social media has long been abused as a source of data harvesting, Talking Medicine sought an ethical approach, with names and identifying information not forming any part of the data sets they create, just manually curated ‘traits’ that help uncover patterns of behaviour and discourse.

While undoubtedly no small undertaking (Dr Fairley was quick to stress that it’s taken years to build a model that can parse data in a way that actual insight can be derived), the effort has resulted in data that speaks a more candid perspective of patient experience – the kind of data that arguably couldn’t be harvested in a strictly clinical environment.

While taking years to build LLM models that can sort and structure this data, it’s the proliferation of generative AI that has helped Dr Fairley be able to derive intelligence. 

Namely, through the creation of Talking Medicines’ own GPT, aptly called DrugGPT.

She says: “[DrugGPT] works within our own curated data set, which is stored in our own data vaults, and we use that model to extract intelligence for our customers.”

Essentially, Dr Fairley espouses the benefits of AI and machine learning is expediting ideation – but, to coincide with Whitefield’s point, this is on the basis that you have quality data, handled in an ethical way.

While the AI for Good event certainly offered enough aspiration to generate some excitement around the technologies potential (specifically around our very strained public services), the business case is much more muted – remaining, broadly-speaking within the same rhetoric that any decent digital transformation consultant has been saying for the last 15 years: start with good data, identify tangible problems then consider using AI to solve that problem or improve processes.