Challenges Integrating GenAI into Various Departments

Did Moderna encounter any challenges in integrating generative AI like mChat and GPTs into its various departments? How did they overcome these challenges?

I think first I want to lay out the foundation that our main reason to embrace AI is to make our company safer. We need to make decisions in every part of the company that are dependent on vast amounts of data under the pressure of time. That can be true in manufacturing that can be true in quality control that can be true in legal proceedings in commercial. So, one field in which AI has been hugely helpful is to help Garner vast amounts of data, extract insight and give have the capabilities to make decisions. And so, I believe that it makes your decision safer. Because before AI, the way that humans dealt with this was to have to go manually, if you will, for vast amounts of data, and try to make the most of this. And of course, we're very challenged when it's a 4000 page report in you have two weeks to read this to give a recommendation. It's challenging, and you don't get them one at a time, either. And so that was a risky situation. And I say this because we are asked very often, how do you mitigate the risks of AI? But I want to ask back, how do you mitigate the risk of not AI? How do you mitigate the risks of making important decisions on 1000s of pages of data with your command center, and your good way? What's the risk mitigation there. So at least we have a new tool in the toolbox. It is wandering toolbox; it doesn't make decisions for us humans make the decisions here. We mostly use it on past data, to understand ways of working and to train ourselves as a research project. We don't operate with AI driven workloads that are critical to the business as it stands. This is our research domain. But we learned so much from the conversation with AI on best data in the way that we should handle data to make decisions, it has already become a vital part to us.

So, in that frame, in that setting that I've just explained that our number one reason for AI adoption at scale, is to lower the risk of the human pressure and decision making in everyday life in every part of the company. The main challenge we've had is imagination, we have rubbed the bottle, the genie came out what are we going to ask for. And so, we keep finding new ways to use AI. And of course, this cannot be centralized, this needs to come from the people. So, we have, for instance, those very complex machineries in our factories. And they are coded in languages that were created for them decades ago. And even though the machine is state of the art, the coding language for it, is maybe 20 or 30 years old. But we still have the playbook for it. And we can instruct this in the knowledge base of a GPT. And asked to perform an operation on our machines, that then translates into the code that the machine uses have been using for decades. And orange veneers can reprogram the machine without having to require the help of a consultant from another company who has knowledge in how those things were programmed two years ago, they are now empowered with the technology to deliver their own custom actions, you know, bursting standard operating procedures, following all the guidelines, and doing these directly themselves with full ownership of what the outcome is going to be. And I would never have thought of that use case it has to come from someone who operates or manufacturing assets, to come back to us and say I have found a complete game changer in AI to help me with my day to day life as an operator in manufacturing.

This has been a very interesting process because of course, we're there at the frontier of GXP and validation. And so, this is the research project, we don't have, as of today, the machine that was directly or indirectly coded by AI. But we've used AI to understand better the way that they are coded to obscure people to train them to give them autonomy in the way that they proceed with the code of the machine. And this also has become the conversation dramatically. So of course, the next horizon for us will be matter, we understand what he's good at, now that we have used it to elevate our own capability on, you know, at our different depths, how will we meet with the demands of the regulation, and the reasons for which they're in place? The accountability, transparency, explainability, with something, you know, that is as novel as your active AI and AI all together to make those use cases come to production at some point when we feel that we're safe doing this and safer than just the humans with their common sense. And they're going to be right. And that's a very exciting time to be in charge of AI products and platforms because this is going to be defining for generations to come the way that we're handling this right now.