I have been writing a lot in the past 12 months and most of it has not appeared on this blog. My output on the blog is reduced as a result but this is normal. Writing is tiring.
About 18 months ago I left my second start-up Fosanis, slightly burnt-out and paradoxically highly-motivated.
At Fosanis I joined a company as it was almost out of cash. I brought it a sizeable grant and we used that to bring in first investor and then later VC funding. Inside of six months, we built a CE-certified medical product and launched it onto the German market. Alongside this, I hired a tech team of roughly 14 people and we built an entire data science infrastructure to improve product development.
The work load, and a sense of frustration with some of the direction, led to the burn-out.
A general sense of dissatisfaction with how Berlin-based medical AI startups were building their products was the source of my motivation.
Most startups are building products which are not fit for purpose.
Let me be clear, I am not talking about Fosanis. We were extremely careful in how we built our product. But I saw the competing interests and I know how hard it was to resist the pressure to cut corners.
AI is beautiful and it has the potential to revolutionise the delivery of healthcare. But unless you build it right it is worse than useless, it is dangerous.
One academic group has studied FDA certified medical AI products and the vast majority of certified products have never undergone a clinical study. Either those devices are being used for trivial automation tasks, or they should be thoroughly studied for the risk of causing harm.
My response has been to write some opinion papers:
- In 2020, I published a guide to building medical AI products.
- Right now, I have a best practices guide for medical AI undergoing peer-review. An old, and sadly error-filled, preprint is already available. I will post the final version once it has been accepted.
- I have another article on medical AI translation which I am publishing with the co-author of the first paper, Vince Madai. Again, I will link to that article once it appears.
- I have just finished working on a project on how to Validate medical AI. This one still needs to be written-up but I am looking forward to publishing it.
At this point, my academic contribution is done. I enjoy writing, but not as my only career outlet. I will continue to blog as I have learned the benefits in terms of forming my ideas.
I will be returning to building things. Hopefully the lessons learned in developing those papers will inform my future projects.