Dr. Rebecca Pope, a Clinical Neuroscientist by background and Director of Data Science and Engineering at KPMG UK, is currently on voluntary secondment at Great Ormond Street Hospital’s DRIVE unit. Rebecca will be speaking at the upcoming AIMed Pediatric virtual conference on 10 and 11 November. But beforehand, she shared some of her insights on how to approach learning more about artificial intelligence and overcoming challenges in AI deployment, mitigating biases and much more.

  1. Understand what you would like to achieve with AI

“Personally, I think getting started in the field is really challenging mainly due to the hype and it’s a relatively new field applied to a medical context. My advice in thinking about the use of AI is what problems are you looking to solve? AI is absolutely not a magic bullet and understanding the question(s) is key. Then, start to up-skill in the theory and mathematics that sit behind AI approaches, such as machine learning.”

“My other piece of advice is to recognize, as clinicians and allied health professionals (AHPs), the great amount of knowledge you already have! This may sound condescending, and that’s in no way my intention, but I often speak with clinicians and AHPs who are too daunted by the field – my advice is that data science applied to medical problems is a bunt tool without this pre-requisite knowledge! Once you’ve overcome this mental hurdle, focus on what you want ‘to be’ in the field: Do you want to apply algorithms to clinical/ operational problems you’re seeing on the frontline? Or are you wanting to have a dialogue with developers and data scientists – here, the nitty gritty of the math is less important, but the interpretation of the results and guiding the build team is paramount.”

“If you do decide to pursue coding and building data pipelines and applying algorithmic approaches – then a key part of this success in my view is learning to appreciate failure. My team and I probably spend 50-70% of our time debugging code. That’s OK and I find that this is a hard lesson to accept for those early in their coding/data science career – but this is ‘part and parcel’ of being a data scientist – it’s not that the code fails – it’s that you need to quickly move on from that and look to understand why and improve and iterate quickly. Of course, with experience, your coding and training of algorithms will improve, but it’s the resilience you build over time to keep going after hours of code failures and debugging that is important – I wish someone had said that to me when I first started my career!”

Dr. Rebecca Pope, Director of Data Science and Engineering at KPMG UK

 2. Understand that AI is only a tool

“I think AI should be relabeled as ‘Augmented Intelligence’ because fundamentally, it is a tool, a (software) Swiss Army knife, selected by human being and to be evaluated by a human being. I am an absolute advocate of ‘human-in-the-loop’. Algorithmic selection and training, relevant to the problem you’re trying to solve is important here – taking the analogy further, you wouldn’t (I hope!) choose to use a corkscrew from the Swiss Army knife to cut a piece of paper. Often AI systems fail to deliver value or a true return on investment not because the technology is ‘wrong’ per se, but rather, the person behind the keyboard has chosen the wrong tool for the problem at hand.”

“When considering the impact AI will have on the workforce, our notion of what constitutes a ‘job’ comes into question. A job should be thought of as a set of interlinked tasks, on which AI will have a differential impact. For example, AI will significantly impact manual, repetitive tasks, but have a lesser effect on tasks which are non-routine with high cognitive load.”

“The skills gap AI will create can therefore be divided into two elements: i) Jobs where the component tasks are changed and modified by AI. The workforce will need to re-skill. ii) Jobs that will encompass new tasks due to augmented intelligence, ushered by AI. The workforce will need to learn to use and interpret AI systems. Notably, the summation of new and modified tasks will also drive job creation (e.g., Chief AI and Robotics Officer in the National Health Service; NHS). My main concern centres on that we do not have a working culture (on the whole) or education system that promotes or supports continuous learning and re-skilling.”

 3. Understand the unique healthcare challenges

“There’s a lot of hype in AI applied to healthcare. But actually if we take a step back and an objective lens, there are very few AI systems are in production within the NHS – why is that? Well, one problem is we rush to talk about answering healthcare problems with this ‘sexy’ thing called AI. The reality is you have to solve the un-sexy part of that equation, which a lot of people don’t talk about and that is infrastructure. Regardless of how well a AI tool should perform in theory and having a well-defined data strategy to capture data to train the AI system, its practical application requires deployment into the existing IT infrastructure. Not only is infrastructure not ‘sexy’ but the return on investment is not immediate in the short- or medium term.”

“It is this combined with data science being outside the core healthcare staff groups, as recognized by the Topol Review that results in limited AI deployment at scale across a healthcare organization. Does that mean we should not try? Absolutely not. I think we do need to challenge ourselves on what we can do in this constraint-based optimization problem – for me, there was an immediate part of that equation where I could help – using my Corporate Social Responsibility hours at KPMG to volunteer my data science skill set at Great Ormond Street Hospital (who have invested heavily in their infrastructure) to help build and crucially, deploy AI systems to inform clinical decision-making and operational efficiency.

Rebecca will be joined by fellow clinicians, healthcare leaders, C-suite executives, technical experts and many more in the upcoming AIMed Pediatric virtual conference hosted in association with the International Pediatric Endosurgery Group (IPEG) and the International Society for Pediatric Innovation (iSPI) between 10 and 11 November. Register your interest or get a copy of the agenda here today!


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.