AI in Biotech: Why I Still Don't Buy the Hype

Aside from my bioinformatics journey, I work as a “brand ambassador”, which is a glamorous way of saying that I set up corporate networking events. Last night I had an event in downtown Boston for a bunch of board members, and got to listen to them all talk about what their companies were up to (sadly there were no biotech companies). At a certain point in the night, I got bored of hearing about “smart, scalable data-driven solutions to next-gen problems backed by AI blockchain models,” and decided to play buzzword bingo with myself to pass the time. I’m sure you can probably guess what word that I checked off over and over and over again. The AI craze is absolutely real, but I wonder how many of the executives dumping money into it actually understand what they are implementing it for? 

It seems like since covid, the surge of AI has created a corporate hysteria, driven by grossly exaggerated headlines, and overall lack of understanding. Articles warn of AI revolutionizing every industry overnight, causing companies to scramble to adopt it, without a clear plan in mind. In fear of being left behind, it seems like many organizations have implemented AI haphazardly, deploying tools they don’t fully understand or systems that solve problems they don’t actually have. An example that I see often is companies grossly overpaying to implement AI chatbots into their website, with the hopes that it improves the customer retention rate. Alternatively, they could just make their website cleaner and easier to navigate, but no, a basic chatgpt extension will make all of their worries disappear. Don’t get me wrong, AI is powerful, but I feel like it is being treated more like a marketing term than an actual tool. Every other press release in biopharma mentions “AI-driven drug discovery” or “machine learning–based biomarker prediction,” but I feel like half the time, when you dig into what they’re actually referencing, it’s just a glorified regression model or some clustering script running in Python. 

In the biotech world, AI is being treated like stem cells or CRISPR when they first became a big deal, except instead of being promised to cure diseases, it is promising to find cures automatically. And that is not to say that the aforementioned technologies were not ground-breaking, because they absolutely were, but they were both plagued by headlines (many of which were probably written by non-biologists) overpromising their impact. The funniest part of the AI situation, is that many biotech companies have been using AI for a while, they just never slapped the label on it until it became a marketing tool. Take protein structure prediction for example. Long before “AI” became a buzzword, computational biologists were using HMMs and neural networks to predict secondary folding or sequence homology, they just weren't marketed as “deep learning” yet. Another example is image-based cell assays which have relied on feature extraction and classification since before I knew what a nucleus was. They used to call it “morphological profiling” or “computer vision.” Now, that same pipeline with a convolutional neural net slapped on top gets rebranded as “AI-powered phenotypic screening.” 

Despite some of the AI slander in this post, I still believe that AI might actually deliver real breakthroughs in biotech, but it has to be implemented thoughtfully. A possibility I’ve thought about before is using language models to mine archives of biomedical literature in order to find hidden drug-target relationships, or combining this data with binding affinity databases to create more precise pathway relationships. But I would guess that things like this are slow, and take time to lay the groundwork and implement. 

So when I bash AI, I’m not bashing the technology, I’m bashing the way it is being used as a marketing tool to appeal to the confused, panicked investors. Or I’m bashing the way it is being treated like a shortcut, when we all know that the best science has never been about shortcuts. Science is built on careful work, testing, and repetition. AI can make that process faster, but it doesn’t replace it.

That leads me to my final point about the panic of AI “taking our jobs.” In bioinformatics especially, most of us know that the work isn’t just about running code or pulling random bits of information from the internet. Bioinformatics is about thousands of tiny choices that shape our results, from what reference genome to use, to how we want to filter and sort differentially expressed genes for our specific situation. These are choices that no AI can replicate without human context. So when people say that “AI will replace scientists,” I think what they should be expressing is that it will replace the repetitive stuff that scientists shouldn’t be doing manually anyway. 

To wrap things up, I’m not worried about AI replacing me or any of my fellow students or colleagues. I am worried about the hype overshadowing actual innovation, and about investors and executives chasing buzzwords without understanding the science behind them. The best use of AI in biotech won’t come from panic or marketing, it will come from collaboration between the people who build the models and the people who understand biology. Until that balance is found, I’ll continue to be skeptical about every headline I see about AI.