Bayer Pharmaceuticals

Oliver Hesse

Computational Discovery Shaper

While maybe not the newest kid on the block, generative AI is certainly one of the hottest technology topics of the time. Specifically, the release of openai’s ChatGPT sparked a plethora of discussions that range from fear to enthusiasm. But besides the sometimes very public discussions around AI there is also an impressive amount of work going on behind the scenes in developing new capabilities and finding use cases where this set of technologies can be applied. Here, the release of AutoGPT has created another wave of enthusiasm and awe, as it combines the power of ChatGPT with other applications and tools.

While many of the discussed and shared use cases may not be revolutionary atfirst sight, there is significant potential that these technologies, and the ones to come and build upon them, will have a profound impact on the way we work and do business. And this holds true for Biotech as much as for any other industry.

The first and obvious capability that generative AI offers is the interaction in natural language. I can simply tell the system what I want and it ‘understands’ my intention. This is fundamentally different from the need to program or learn a certain set of rules and commands. While prompt engineering is a particularly important topic (generative AI uses your input as a starting point, and the better the input, the better the response) the fundamental idea of being able to ‘communicate’ in natural language is a central characteristic. With this I can also ask the system to e.g., summarize a particular topic or text that I provide or to generate some text that I need, from a friendly email to a journal article like this or potentially a document for regulatory authorities. The ability to synthesize enormous amounts of information in seconds and respond in natural language will be an invaluable productivity booster on many fronts.

Key to the success of using generative AI in a business context will be the availability and access to internal data to get meaningful responses to a particular business use case. But imagine the possibility of asking your ‘personal assistant’ a question like ‘What was the difference in manufacturing from product a vs b’ and get an answer in seconds. Similarly for questions about your regulatory filings in different markets or an analysis of the competitive landscape in a specific area.

But the applicability does not stop here, it is merely the starting point. The tools are really good when you have structured input and or output. The prime example here is certainly coding. I can simply tell the system to write me some python code to do x or to review or document my code that I generated. While coding may not be at the heart of work in biotech, there is nevertheless a lot of applicability, and this is where tools like autogpt come into play where you can combine various tools.

To make this more tangible imagine the following scenario where you simply instruct the tool to plan your process development: to gather information about your new product, summarize the findings, develop an experimental plan, provide the respective programming of your equipment, check for the availability of resources and materials needed, order them if necessary, compile all this into a summary and send it off to people that need to be informed. 5 min later the tool is done.

Sounds like Sci-Fi, but that is the power if you combine the capabilities of multiple tools through something like autogpt, where you can instruct in plain language. Here, various agents then start creating and executing tasks and creating new task and due to their ability to ‘translate’ from natural language to structured code for e.g., API (Application Programming Interface) calls (i.e., interfacing with another system) they are able connect all these different systems without the need to explicitly code all these transactions. And you can still have the human in the loop to e.g., allow for certain transactions.

Now this is of course a) just one example of what could be done and b) will not be done tomorrow, but it demonstrates some of the powers and the potential that these new tools can provide.

Despite the exciting potential of generative AI, there are also risks and challenges. One major concern is the lack of transparency in the algorithms used by generative AI. This can make it difficult to understand how the algorithms arrived at a particular result and assess the validity of the generated solutions, which is particularly important in the context of developing medicines where patient safety is paramount. While the regulatory framework in pharma is often seen as an obstacle to the implementation of novel approaches, this should not be seen as an excuse to not assess these technologies more and better understand their potential and pitfalls.

In conclusion, generative AI has the potential to transform the way we work and do business in many areas, including Biotech and we only start to understand how transformative these tools will be. We still need to learn about the best way to use them, but we can certainly not ignore them.