By Arthur C. Ouwehand – Global Health Nutrition Sciences, International Flavors & Fragrances, 02460 Kantvik, Finland
…With that, I could end this blog.
But it would be 499 words short of the target length. So, let’s rephrase the question. What can generative Artificial Intelligence (AI; e.g. ChatGPT) do for probiotic science?
The strength of generative AI is that it can manage large amounts of data. Scientists, engineers, manufacturers all have large amounts of historical and new incoming data. If you combine that data, look for patterns, correlations or specific targets, you might be able to identify new strains or find new health targets for new or existing strains. In the future, it may be possible to create personalized probiotics that are tailored to an individual’s genome, microbiota, and lifestyle. Generative AI could potentially help identify the most suitable probiotic (or nutritional requirement) for an individual. It may also be possible to help optimise the fermentation process using data collected by the sensors in the fermenter and correlating this with the help of generative AI to yield and stability. This does not come easy. It requires the system to be trained, it requires a lot of computational power and is not something you can do with ChatGPT but requires other systems. That will require time and money. But maybe that effort is worthwhile if it gives you better quality probiotics, smarter design of clinical studies or new health targets (= new markets).
So, this is not something for today but for the (near?) future. What can be done today using e.g. ChatGPT? Writing is a very feasible target. This blog could have been written by ChatGPT and it would have done a very decent job. However, that is not what happened. This was not written by artificial intelligence but by ‘actual intelligence’, i.e. me, with all my quirks and limitations, with the help of artificial intelligence. ChatGPT nicely generated text on the topic of this blog. It was not exactly what I wanted to write but it got me started. When writing, not being a native English speaker, I may generate text that contains what I want to say, but not very eloquently. ChatGPT can reformulate again and again until I get something I like. It can also otherwise check the text one writes e.g. for clarity. There are other things where it is very handy: It can summarise text, e.g. articles, though you need to feed it plain text. It can translate, so maybe an alternative to Google Translate?
This may sound worrying to some. However, we have already entrusted computer algorithms with many difficult and boring tasks. It works very nicely for e.g. statistics, especially in the management of sequencing data. While handy, is it trustworthy? As mentioned, ChatGPT generates very nice and convincing texts. But what is the origin of these texts? In scientific writing, it is important to include references to support claims and ideas. No problem, just ask ChatGPT and it will provide you with ‘references’ and indicate where they are relevant. However, like the text, ChatGPT seems to sometimes compose the reference; authors from here, title from there and a DOI link from yet somewhere else. Obviously, that is a problem. Why do you do that dear ChatGPT? “I am a machine and have been trained with data from the internet. Always check with an expert”.
Where does this leave us? ChatGPT is a handy writing tool; no more writers block, better formulated text, that can lead to better productivity. However, ‘Trust, but verify’. One more thing; assume that what you do with ChatGPT is basically public. So, you might not want to ask it to summarise your top-secret internal report.
Arthur Ouwehand
Dr. Arthur Ouwehand is a Technical Fellow at International Flavors and Fragrances, Kantvik, Finland. Dr. Ouwehand received his M.Sc. degree (1992) from Wageningen University (Netherlands) and his Ph.D. degree (1996) in microbiology from University of Gothenburg (Sweden). Since 1999 he is functioning as Adjunct Professor at the University of Turku (Finland).