The explosive, seemingly overnight growth of generative AI tools like ChatGPT over the last year has pushed AI to the front of nearly every R&D leader’s agenda. And it’s easy to see why. AI’s integration within R&D processes and “lab of the future” initiatives has the capacity to dramatically accelerate innovation cycles and give companies a powerful leapfrog opportunity in a marketplace that is increasingly multi-dimensional and complex. Understandably, every company wants to be the first to add this powerful tool to their discovery toolbox, so they rush to their IT department to develop an AI. Then, they run into some surprises… (Read about these surprises in our eBrief.)
When a transformative technology comes onto the scene this quickly, there are bound to be some misconceptions. And that’s definitely been true for AI, especially in the physical sciences, and in particular the Chemistry and Materials Science R&D space. R&D typically has a data landscape that is vastly different from other business functions, and this means it requires different data management strategies to enable the use of AI. With our end-to-end R&D platform, Albert is at the forefront of R&D digitization and “lab of the future” efforts in the Chemistry and Materials Science industry and works with customers around the world on empowering their labs with AI-driven insights.
Over the last year, we have answered countless questions from customers and debunked many of the same myths over and over. To help R&D organizations walk into this process with eyes and minds wide open, we are debunking a few of the most common myths here.
Myth #1: AI will replace Scientists
While there are some truly remarkable applications of AI, these systems are generating outputs that are still very much within the domain of knowledge they were trained upon and struggle with anything beyond that domain. In R&D, where innovation often requires novel ideas and approaches that diverge from existing knowledge, the unique ability of humans to think abstractly, ask new questions, and draw from a wide array of experiences is crucial and will remain so even as AI continues to advance. So, there will always be a need for scientists.
Myth #2: You can use AI with only data from a few experiments
Many people think that if you have data from a few experiments, you can give it to an AI, and it will immediately understand the nature of the chemistry in your problem and be able to tell you what to do next. The reality is that to get to the type of AI we are seeing in other business units today, you may need months but more likely years of diverse examples from across the R&D ecosystem to train the AI on before it will be reliable enough for all scientists to use at scale. For example, most of the AI tools people think of today have been trained on hundreds of millions or more examples accumulated over many years.
Myth #3: AI can perform miracles
Organizations often expect AI to come up with something they never would have thought of. While AI is great at pointing you in the right direction to look for novel answers or in recommending small changes to what you’ve done in the past, it can’t tell you something it’s never seen before without a high likelihood of nonsensical proposals. The most pragmatic way to approach AI in R&D is to expect it to propose recommendations that are reasonable and based on historical data that already exists. The key function of the AI then is to enhance the so-called ‘network effect’ between researchers — the work that any individual does today, informs the potential ideas (i.e. the proposals generated by an AI) of all their colleagues tomorrow. This is where the greatest opportunity for accelerated innovation exists today.
Myth #4: AI doesn’t require human intervention
Unless you have millions, or billions of examples to train the AI on, you will need to help guide it yourself. This is especially true for Chemistry and Materials Science R&D, where domain expertise plays such a significant role. Early AIs behave like toddlers; they need constant input and correction, explained in diverse ways so they understand the guardrails and constraints. While this can take time in the beginning, the investment in teaching the AI domain knowledge will pay off immensely down the road.
Will AI Actually be Useful in the Lab?
While the journey to AI may take longer in R&D than some people realize, it brings with it an enormous number of benefits in productivity, efficiency, and insight—many of which can be realized in as little as a few months. The key is taking the right steps now to ensure your organization is positioned for success. To learn more about the critical steps you must take to successfully implement AI, download Albert’s comprehensive eBrief: Artificial Intelligence in R&D – Critical Path to Success.