Two variables for this age
I've said previously that we are in the age of the last mover. What this means is that:
Any result that can be produced by an algorithm will become:
Free to produce
Instantaneous to produce
Geographically independent, i.e. you will be able to produce the result anywhere
This is obviously not the case yet. There is still a lot of paid software out there. But it's starting to happen. I believe the most advanced piece of software known to man is ChatGPT. ChatGPT is free to use. Another Large Language Model (LLM), GPT4All goes ever further. It is a free-to-use privately hosted LLM. You can download, use, and own your own model for life.
I believe there are two variables at play when determining when an industry or piece of software will become free. The two variables are:
The number of postgraduate students doing research in the field. Let's call this s
The funding available to the field, f
The more students doing postgraduate studies in a field, the higher the likelihood that one of them will develop software and make it open source. Making something open source is an excellent way to drive adoption and to stake your claim in a field. Furthermore, it’s the easiest way to get people to choose your software instead of the incumbent versions. It may, in fact, be the only way to wrestle market share from the incumbents.
This effect of going open source will compound and gain speed as the culture in a field change. If a new student starts their field of study, and they see that previous students gained exposure with open-source software, they are more likely to do it as well. It is now standard practice for students in AI/ML to package software relating to their research into open-source repositories that can be downloaded.
The culture in a field also determines how paid software is perceived. In the field of Geotechnical Slope Stability, for instance, paid software is the norm. The market will evaluate the software solely based on its merits. If its merits exceed its price, transactions will start to happen. It is difficult, however, to imagine a new piece of paid software for performing tensor calculations. We already have PyTorch that is widely used and free. The new software will be scoffed at, not because of its merits and features, but because it doesn’t fit into the culture.
The second effect, f, is the amount of funding available to a field. When software companies get funded, adoption is the key metric to determine whether the software is valuable. Adoption will inevitably be faster when there is no price barrier. Funders will then push their companies to provide some value free of charge.
A good example is Pydantic, a data-validation Python library. They raised funding for developing a new version of their free software, and they will be offering cloud services on top of it.
The question I’m wresting with is this: If you find yourself in a field with a sparse number of postgraduate students and low levels of funding (compared to AI at least), should you stick to the old principles of offering only paid software, or should you be the last mover in your field?
