Three Layers of AI with an Analogy
In my earlier post, I mentioned about three distinct activities described under the broad term “AI”. They are -
(i) using web-based text engines like Chatgpt, Claude or Gemini and their extensions as coding tools,
(ii) downloading numerical models directly from Huggingface and building applications on top of them, and
(iii) developing and training mathematical models for new applications.
Maybe an analogy will help in understanding the differences. You can run Linux or some application software in your computer by downloading and installing it. Similarly you run BLAST at NCBI (or even locally). That is equivalent to the first layer of AI, and it is the layer everyone is fascinated with. However, it gives very limited access to the entire set of possibilities.
In case of software, more curious souls go to github, check the source codes and maybe download and compile them in their machines. That works only if the software is open source. The AI equivalent of github is huggingface, but instead of code, you download trained “models” from there. This is what I present as the second layer.
In a recent trend, many people are installing AI models from Huggingface locally so that they do not need to rely on companies like openAI (chatGPT). This has several advantages (cost, flexibility, privacy) and disadvantages (learning curve, hardware cost, unavailability of commercial models). For non-LLM models, such as those for genomic data, this is the primary way of exploration (although Evo also has a website to give some access).
The software world analogy of third level of AI is to write new bioinformatics algorithms or writing the code for Linux. In case of AI, you derive new mathematical models, train them with data and then test on new problems. Possibilities are unlimited, and I am seeing increasing number of computational biology papers in this area. I will cover many of them in this blog.
This summer (July 13-24, 1-3PM PST), I plan to have an online class on Machine Learning and AI Concepts in Python covering this last aspect. This is the second summer of this class. Please feel free to get in touch (info@coding4medicine.com) if you are interested.
