AI? You Mean Chemometrics?

With Rasmus Bro

How should we define artificial intelligence (AI)?
 

I find that the term is currently too ambiguous. When you talk to the general public or mid-level and high-level leaders, they are thinking of something completely different when they say “AI” compared with what a person working with it would think. And the historical version of AI (which has been around for many decades) is quite different from the current iterations. In short, I think AI is fine to use as a buzzword, but we all have to be much more specific if we are working in a “real” context.

In fact, actual AI is very different from field to field. When I talk to peers, I use more relevant, specific words. For example, I often go with “data analysis.” That effectively takes the magic out of it so we can discuss more “honestly.” Note: I don’t really mind the hype and I am happy to take advantage of it! But since I am here talking as a scientist in the field, I have to be somewhat sober. So, any kind of data analysis – that’s AI!

What is now referred to as “AI” or “data analysis” in a chemistry context is actually chemometrics. The field has existed since the seventies and was driven by the richness of the instrumentation. Chemists get very excited when someone says they can measure 100,000 analytes in one go. And they go out and do that on three samples and expect to learn 100,000 things. Then, later (oftentimes years later), they realize that they need data science to harvest the information. This sort of development has been seen over and over again; for example, in various omics fields.

What problems in analytical science could we solve with AI? 
 

As instrumentation and other aspects continue to improve, we’ll continue to develop new tools that help us extract information in a meaningful way or introduce more automation as methods become more routine. With data science, we can handle the increasingly more complex and rich data that we encounter with developments in instrumentation and the internet/connectedness. Often, the biggest problem is not the actual data science but understanding the goal, the quality and validity of the data, and other related matters.

So, when people ask about the problems we could solve with AI, I say, “I am really sorry but I’ve been using AI for over three decades. This is nothing new.” My first paper in 1994 was about making neural networks easier to interpret. My research today is concerned with developing new algorithms within statistics, mathematical modeling, machine learning, and so on, to solve specific problems within analytical chemistry. For example, I have developed several methods for unmixing complex signals from complex samples. All of this is “data analysis” – AI!

Really, data science has existed for more than 100 years – and has been referred to as “AI” for perhaps 80 years. We keep increasing information and that needs to be processed – hence data science is needed. The specific needs differ from field to field: the development in process analytical technology is different from genomics, which in turn is different from flavor research, and so on.

Can AI live up to the hype? 
 

It cannot be denied that there is a great deal of hype around AI, largely driven by the recent emergence of generative AI tools, such as ChatGPT. How these tools will change society, especially in the education and teaching domain, is fascinating. But I’m afraid it just doesn’t have a great deal to do with analyzing analytical chemistry data! And my concern is that we end up spending a lot of money on silly projects because mid- and high-level leaders and decision makers are mainly concerned with the hyped version of AI – generative AI.

So, let’s be specific, avoid conflating actual AI – data analysis and chemometrics – with generative AI, and be wary of being led by hype rather than solid research needs.

Rasmus Bro is Professor, Design and Consumer Behaviour, Department of Food Science, University of Copenhagen, Denmark