Initial Patient-ChatGPT exchange
Follow-Up Patient-ChatGPT Exchanges
While my consultation exchanges with ChatGPT didn’t exactly happen, the vision of an AI chatbot that can answer patients’ queries based on their health conditions — both past and present — can very well be realized in the not-so-distant future.
In fact, this is exactly what I predicted in my recent lecture “From ChatGPT to Precision Medicine, how AI is improving lives one neuron at a time”
In early February 2023, I was honored to lecture at the University of Chicago Booth School of Business. This is part of Professor Mladen Kolar’s Machine Learning class where he would invite industry experts to discuss some of the latest trends in AI and Machine Learning. As this was the first time I returned to my alma mater since the pandemic, I was particularly excited. With a group of curious students asking insightful questions, I had a lot of fun discussing how AI is transforming the healthcare industry.
For those who are not able to participate in the lecture, I would like to capture some of the highlights here.
Precision Medicine GPT ( PMGPT ): The Future of Medicine
The success of precision medicine -- tailoring of treatments and therapeutics based on patients’ profiles -- is fueled by the vast amount of data created by advanced biotechnology, health sensors patients use at home, and the collection of information about patients’ journey in healthcare with hand-held devices.
However, most of the current modalities — vaccines, diagnostics, gene and cell therapies, mobile APP for diabetes management — typically get their data input from only a single source of data such as liquid biopsy, MRI, genomics sequencing, or electronic health records. Witnessing the power of the Large Language Model (LLM) capability of ChatGPT, we asked the question “what if we can ingest data from these sources, and build an aggregate foundation model that would enable physicians to prescribe a personalized treatment for their patients?”
A Machine That Can Finish Your Sentence
The underlying idea of GPT-3 is a way of linking an intuitive notion of understanding to something that can be measured and understood mechanistically, and that is the task of predicting the next word in text.
Ilya Sutskever, Chief Scientist at OpenAI
Powering ChatGPT is the “Generative Pre-Trained Transformers” ( GPT ), a neural network that is allowing the chatbot to guess the next word in a sentence. Incorporating large contexts found in languages, the transformer architecture is designed to address the limiting sequential processing of existing language models such as Recurrent Neural Networks ( RNN ). Trained by a massive amount of data ( 45TB of data, 135 Billion of parameters ) from the internet, GPT is able to learn skills such as language translation and question answering. By mastering these tasks, GPT would also learn how a natural language is pieced together.
DeepMind’s AlphaFold: A solution to a 50-year-old grand challenge in biology
Essential to practically all life functions, proteins are large complex molecules, made up of chains of amino acids, and what a protein does largely depends on its unique 3D structure. Figuring out what shapes proteins fold into is known as the “protein folding problem”, and has stood as a grand challenge in biology for the past 50 years.
In a major scientific advance, the latest version of DeepMind’s AI system AlphaFold has been recognized as a solution to this grand challenge. With the capability of accurately predicting the 3D shapes of proteins from their amino acid sequences, the release of AlphaFold is now paving the way for the development of new medicines or technologies to tackle global challenges such as famine or pollution.
Similar to ChatGPT, AlphaFold is also built from the Transformer-Based neural network. Faced with the challenges of tackling the large contextual information of a protein’s 3D structure — not unlike that found in languages —, the researchers at DeepMind turned to the Transformers architecture and they were able to model an AI predictive system based on amino acid sequences.
Vertical AI, the next set of startup opportunities beyond ChatGPT
Go build a vertical AI startup!
Sam Altman ( OpenAI’s CEO ) replied to a conference audience’s question when he was asked about the next startup companies founders should start.
Indeed they should!
Witnessing the power of ChatGPT through its ingestion of publicly available data, we realize that there will be tremendous untapped opportunities for different vertical industries. As shown in the ChatGPT-Patient exchanges, the chatbot, when built with all the various sources of data, can help prescribe a very personalized treatment plan, delivering great value to the healthcare industry.
However, gaining access to these privately-owned data would remain a challenge. Having proprietary access to such data will become any vertical AI startup’s competitive advantage. Whether this data is built through business partnerships or through synthetic data generation, founders will be able to create a strong moat for their startups.
Good AI portfolio
Therefore, at Good AI, we place a high premium on startups that could gain proprietary access to this rich set of data. Whether they are building System-On-A-Chip (SoC) for precision gene/cell therapy, creating an AI Drug Discovery platform for microbiome-derived therapeutics, reprogramming human cells to precision engineer the next generation of medicine, or engineering enzymes to improve environmental health for millions of people across the world, our portfolios all have established these data access competitive advantages.
At the end of the lecture, I had the chance to answer questions regarding patients’ data privacy and government regulation. While these concerns would certainly pose challenges for the widespread adoption of AI for the healthcare industry, we believe they are also inspiring a new class of mission-driven founders — as we have witnessed from our portfolios — to tackle these problems head-on, delivering the ultimate promise of improving lives one neuron at a time.
The lecture presentation can be found here
Interesting from a Venture Capital perspective which startups to look out for in the next year.. Great read, Thanks!