Unpacking the Influence of Artificial Intelligence on Biomedical Professions

I only realised this several years later, but when I was training to be a medical doctor I did not have to think once about what kind of work I would be doing after I completed my training. Granted, one must think at some point about what area of specialisation one chooses, but that too was more akin to choosing from a set bundle of options rather than really exploring what all the options are.

As my career trajectory took me further away from the clearly marked racetrack of clinical practice to the open wilderness of the business professional world, I started realising how ill-equipped medical training leaves one if they decide to do anything other than what doctors do. It’s not really a flaw in the system: one of the perks of training to be a medical professional is the guarantee that one does not have to put much effort into finding gainful employment in a respectable and lucrative line of work.

That being said, there are and always will be those who decide to take the path less travelled and at some point wish that they were better at “selling themselves”, a skill that is hardly ever required to be a good doctor. While the medical profession, alongside professions like law, accounting, and architecture, are extreme examples of this, the phenomenon is not too unfamiliar to those pursuing degrees or possessing qualifications in biomedical sciences.

Academia or Industry?

A big question is – Academia or Industry? Academia is the easier choice, offering more certainty in exchange for narrowing of focus, while industry beckons with vaguely defined promises of glamorous career opportunities! This is all to say that even before AI marched onto the scene with the threat or promise, depending on one’s viewpoint and situation, of upending the job market for professional work, it wasn’t exactly smooth sailing for those looking to start their careers after obtaining advanced degrees.

The Impact of AI on Biomedical Industry Jobs

So, let’s begin to unpack what impact AI has made and is likely to make on jobs in the biomedical industry. One can only speculate as to how pervasive the use of AI will be in the industry. We’ll leave aside its impact on recruiting and job searches for now as it is a whole big topic on its own that is best covered separately. Focusing on how it is actually used in the industry, broadly speaking there are three main types of AI:

• Computational chemistry models – Simulative AI

• Outcome prediction models – Predictive AI

• Large language models – Generative AI

Simulative AI

These AI models have helped turbo-charge computational chemistry for several years at this point. Examples include AlphaFold and Boltz models. At their core, these models can not only understand molecular structures but also predict what would happen when certain types of molecular structures come near other molecular structures, helping simulate ligand-receptor interaction.

This is especially useful when dealing with protein molecules with molecular weights running into hundreds of thousands of daltons, with hundreds to thousands of possible interactive surfaces on them. Naturally, these models have played a big role in high throughput screening of lead candidates for pharmaceutical drug development while lowering the cost of generating “hits”. This process occurs at the very beginning of the drug development process, and all major pharma companies as well as the majority of biotech startups employ labs and scientists who specialise in working with these models.

While the current slate of AI founders love talking about how their models will “cure cancer one day”, computational chemistry models have already been helping us discover more and more cancer treatments for over a decade now. Ironically, the massive amount of attention and funding that language models have been sucking up over the last few years may have had a slowing effect on R&D for simulative AI models. Whether that’s true or not, it is unfortunate that LLM makers continue to obfuscate the role that AI has been playing in drug development long before the first LLM-based chatbot began selling £20 subscriptions.

Predictive AI

When a drug candidate moves out of the lab and into in-vivo trials, where it is introduced for the first time into living systems such as small animals, we are looking at a much more macro scale than molecular interactions. A tremendous number of variables, a vast proportion of them unknown, determine what effect a drug might show on a disease model of, say, a mouse or a guinea pig. Scale that up by several orders of magnitude and then one can begin to imagine the challenge of computationally modelling the expected effect of a drug in a human being with any degree of certainty or accuracy. While we are several years, if not decades, away from “synthetic humans” who exist solely as computer code on whom we can run “drug trials”, we are closer to predicting roughly how an untreated healthy or diseased individual might fare on certain metrics of interest so that we may use it in place of control arms in clinical trials.

What predictive AI models can do today is not any less impressive either. Just as AI-based weather modelling and hence forecasts have gotten increasingly accurate over the years, AI-based prediction modelling can be used for things such as early detection of chronic diseases, predicting risk of rehospitalisation, and optimising clinical trial recruitment.

While not creating anywhere near the same amount of buzz as LLM-powered chatbots, industry functions such as Clinical Development, Health Economic Outcomes Research, and Patient Engagement have been working on adopting predictive AI models for a while. One can only hope that this sort of AI receives some of the love that is being showered by the tech industry on LLMs and image/video generators.

Moreover, predictive AI models are being harnessed to interpret vast quantities of clinical and real-world data, enabling researchers and clinicians to identify patterns and correlations that might escape traditional analysis. These insights can inform personalised medicine strategies, tailoring treatments to individual genetic profiles and lifestyle factors. As regulatory frameworks adapt to these technological advancements, the integration of predictive AI into routine clinical workflows is expected to accelerate, further transforming patient care and drug development processes.

 

Generative AI

It is the newest kid on the block but boy is it the most popular! Following the release of the 2017 paper by Google scientists that set the stage for the development of what we today call generative AI centred around large language models, it took just 6 years for AI chatbots to reach consumers. Since then, the greatest advertising campaign in human history has pushed the use of LLMs on everyone from CEOs of multi-billion-dollar companies to Jack and Joan going on their weekly grocery run. Quite distinct to the other two types of AI I mentioned, generative AI was not created to solve a particular problem, rather everyone has been left to figure out how they can put it to the best use to them. This is not necessarily a flaw of the technology, but it is the source of its biggest strength as well as its biggest weakness. The strength comes from the virtually infinite number of use cases it can be deployed for. The weakness on the other hand is too fold, firstly, it relies on the user to have sufficient patience, knowledge, and ingenuity to figure out how to extract the most value from it, narrowing the field of potentially high-impact / high-value use cases being discovered. Secondly, by trying to be anything anyone wants it to be in most cases it struggles to be good enough to keep anyone happy all the time. While these observations may appear mostly philosophical, they have real-world implications for professionals working in the pharmaceutical industry.

1. There is expectation from leadership teams within companies for AI to be mass-adopted without always having a clear idea of how it should be done.

2. There is an expectation that with the use of AI work can be done more quickly and cheaply without loss of quality.

3. It is expected that each employee figures out how to use AI to make themselves and those working under them more efficient and productive.

The above statements are generalisations and are meant to illustrate of the trend in the industry in the last few years. Obviously, individual companies can be found at different places on the spectrum of AI sceptics to zealous adopters. What this does mean for those seeking jobs in functions of the industry where Generative AI can be deployed, which is most functions, knowing and adapting to the “AI culture” at the company or at least in the team would be an important contributor to success.

For now, the generative AI is used in industry functions such as Insights and Analytics which includes Market Research and Competitive Intelligence where it can speed up some of the core processes involved in secondary and primary source research from research design to interpreting results. Good use cases for Gen AI have also been found by those working in Marketing, Medical Communications, Patient Engagement where it can act as a “ideas generator”, help in creating and refining copy, and answering basic but common questions from customers. Of course, more straightforward and less specialised uses cases abound from transcription and translation to composing emails which together can help save a lot of employee time to be used for more human-dependent tasks.

What does this all mean?

If you have read this far and found it interesting, suffice it to say that I have barely scratched the surface on this topic. Choosing the right industry role for yourself can be a daunting task and a general idea of what the industry has to offer is a good starting point but one may still feel lost.   For those looking to start your career or feel like you want to explore what else is out there for you in the biopharma industry, feel free to take a look at our Services page.

Have a good one and see you next time!