Running through myGenAssist – their GenAI OS – they advised that the way to circumvent the ‘classic tech-adoption curve’ is to build a local community of enthusiasts, with the objective of creating a team who can become force-multipliers for AI across the business. 

The Digital Health Validation Lab exists to validate the value of new technologies for the health service, questioning how to build successful clusters from universities and get them into the health system – education of the nation being a fundamental preventative method, he insisted.

The potential applications of AI in healthcare include, of course, drug discovery and development, but also operational efficiency and resource management, diagnostics and imaging, predictive analytics and risk assessment, clinical documentation and workflow automation, treatment planning and personalisation, patient monitoring and remote care, and medical translation and communication.

Prof Lowe also emphasised the criticality of real-world evidence (RWE), including for clinical applicability, and ran through their healthcare AI value pyramid:

It’s all about using data to tell a story, insisted Harris: “We talk about omni and right message, right channel, right time – it’s supporting the employee to give the best experience to the HCPs and by extension the patients.”

And when it comes to AI, it has been ‘sprinkled’ across pretty much all that’s done in pharma, said Tunnah, and Harris – who mentioned that she loved “Arbour’s aspiration, but it’s not reality” – deemed omnichannel “something that can bring groups together,” but advised that it should be built “piece by piece and [brought to] where people are working.” 

It is a matter of augmentation, rather than any notion of replacement. While Ferring have “been on a longer journey to play catch up, in some respects,” she also shared that, “it’s about playing smarter, not harder […] It’s about the opportunity to look forwards and know who comes in at which juncture. And proving value to the organisation.”

Harry Jennings, partner at VWV, meanwhile, provided a highly interesting navigation of the legal landscape and the intellectual property, ethical, and legal implications of GenAI in commercial activity.

“Who invited the lawyer?” he joked, adding that lawyers are misunderstood because they point out risks. But it’s not to stop the fun, rather to go in with open eyes. 

From a business perspective, if you have an AI project and you need a use case, a business case, the budget signed off, and to make the thing actually happen – legal runs through all this. The best thing is to start thinking legal issues early on, he advised, spotlighting four key legal considerations for AI:

• Do you own it?
• Are you allowed to use it?
• What standards do you need to meet?
• If a problem occurs, who pays? 

Input data is highly important, obviously. You either have the data already, or you go and find it somewhere else. If it’s not yours, you should consider whether you have the rights to use it for your project, Jennings said. And if you can’t identify access terms for your particular dataset, you need an agreement in place to give you the right. This data is often messy, unstructured, and diverse in form – so, there’s this preliminary project just getting the data ready for AI.

All this information was to state that there will be lots of relationships in any AI project: contracts, promises, rights, standards, reliance, and support. “Don’t try doing this without a contract!” he warned, still seeing this happen and still surprised every time. 

In all the above relationships, ideally you want to be looking at a contract and seeing nice clear IP drafting – in reality, though, this is wishful thinking. Which loops back to risk analysis. With a ‘new-ish’ sector like AI, said Jennings, things are still being worked out and there are certain concepts – reasonable care, for example – where you don’t necessarily know what that means technically and/or legally. Is there something independent and objective that can be pointed to as a defence? If someone asks the reason and reasonableness behind an approach or reliance and purpose – list the standards, the experts: “Evidence in short, and evidence is something that pharma understands.”

AI in pharma – what could go possibly wrong? Well… From accuracy and bias to injury, third party rights, performance and availability, privacy and security, and ethics/reputation – the answer is, a lot. Safety is the biggest issue. AI in clinical decision processes? That certainly scares Jennings. The simplest question in these matters is, ‘Can it hurt someone?’ 

There are people who are nervous about the coming AI apocalypse, he noted. To him, that means ethical issues and safety issues will be spotted quickly and highlighted for the world to see: “Don’t be the bad example. Sit closely to ethical standards,” he advised, adding that, “there is very detailed guidance on the EU Data Act Prohibited list – a flavour of what you shouldn’t do!”

He saved third party rights, “the best for last,” he mused. If using a pre-trained system, seek what it was trained on, what data, and how. Did the developer have the rights to the data? The answer is no for LLMs, for example. Did ChatGPT negotiate with all the websites around the world and get the licenses for scraping their content? No, it didn’t. It’s a question of copyright, basically, and changes are on the horizon. Therefore, the overall takeaway was: proceed with a cautious approach.