2023: The Dawn of Exponential Convergence

Hello again friends and colleagues,

Happy New Year!

By the way this week's newsletter is purely human-generated :-)

Looking back, I believe we will all remember 2022 as an inflection year for humanity in general and specifically for healthcare. Exponential technologies (ones that compound improvements rapidly enough to improve thousands or millions of times in the span of years leading to a qualitative disruption performance) that have transformed industries like communications, media, electronics, computers, and information processing, are now moving into all aspects of life. Moving into domains that have been till very recently considered immune to machine disruption.

Machine Learning

One of those exponential technologies is machine learning. 2022 was the year that humanity's monopoly on intelligence came to an end! Wow!! What a statement. But it's true. There is no longer doubt that machines can now at least imitate intelligence in a way that is indistinguishable from the real thing. And for a lot of things that is good enough to be hugely disruptive. Large Language Models like GPT in its different flavors, most recently ChatGPT, can write essays that teachers and professors can't distinguish from ones written by their actual students (except when they are too good to have been written by the students :-). Text-to-image models like Dall-E and MidJourney can create "art" that has won in competitions. And other models can write computer code that is competitive with the best human programmers (AlphaCode), negotiate like the best human negotiators (Cicero), and many more. I think the best proof of how good these models really are is that experts and professionals have looked at their creative output and decided it's a threat to their livelihood, and are suing the companies that created them for using their work to train the models without permission.

Machine learning will be hugely disruptive to healthcare. General models like ChatGPT have not been trained specifically for healthcare but they are still good enough to pass the United States Medical Licensing Examination (USMLE) sample test! AI models that can read chest X-rays, retinal images, and pictures of moles, ... are already out there. But the models can only do one thing. Large Language Models like ChatGPT but ones that have specifically trained on healthcare data are starting to appear. Within the last few months, Stanford and MosaicML created PubMedGPT, and the University of Florida working with NVIDIA created GatorTron. Those models have been found to perform better on a variety of healthcare-specific tasks like clinical concept extraction, medical relationship identification, and answering medical questions! And that is just the beginning. I expect the first multimodal models (ones that can work with text, images, sound, and videos at the same time) to debut in 2023. Some are saying that GPT-4 which is hotly anticipated will be a multimodal model but that remains to be seen. Regardless, someone will do it. And then it will appear in healthcare-specific models. This will be a game-changer. Medicine in multimodal. USMLE questions that relied on interpreting X-rays or ECGs were taken out of the experiment I mentioned above where ChatGPT passed the test. Soon this will no longer be necessary.

Machine learning is improving at an unprecedented exponential rate. Models are getting bigger by a factor of 10-100 times per generation which is every few months! The rate of improvement is and will continue to be shocking. Societal technological transformation has moved from centuries (industrial), to decades (information), to years (mobile), to months (machine learning). Even experts in the field are experiencing whiplash!

Machine learning will transform medical practice, healthcare delivery, drug discovery, and medical devices, to name a few areas that be impacted.

Gene Sequencing

Another exponential technology, that has been incubating for a couple of decades but will likely have it's own disruptive appearance in 2023 is gene sequencing. The price and speed of sequencing technology has been improving exponentially as well and sequencing a full human genome will soon be a few hundred dollars to the end-user and available in hours compared to a cost of tens of thousands and days or weeks to complete just a few years ago.

Genetic information will be used to identify healthcare risks years or decades before they cause illness, inform much better medication selection decisions, and will accelerate research into many medical conditions and their treatment most notably cancer.

CRISPR

A third exponential technology is gene editing also known as CRISPR technology which allows very specific changes to be made in genetic code as a treatment of some genetic conditions. The first such treatment was approved in 2022 albeit at a cost of $3,500,000! Many more CRISPR therapies are in pipeline at such companies as Editas, Intellia Therapeutics, and CRISPR Technologies (I have small investments in all three companies).

The Exponential Convergence

The last thing I will mention about exponential technologies is that on top of the disruptions that each of those technologies will bring about individually, we are also witnessing exponential convergence! These technologies are merging and feeding off of each other. A good example of this is AlphaFold which is an AI model created by DeepMind which used machine learning and gene sequencing to identify the amino acid sequence of the proteins encoded by the genetic code and predict their 3-dimensional folding structures. This is going to be huge for drug discovery. DeepMind which is a Google company has spun off its life sciences business into a separate company called Isomorphic Labs.

Wishing you a very happy and prosperous New Year! May 2023 bring you joy, good health, and success.

See you next week,

Sam