On AI technology and future of software
The year 2023 became a year when large language models (LLMs) made their way into the lives of many people. It became clear that producing fluffy texts could mostly be done by machines, provided the topic had already been present in the huge text corpora used to teach the models.
Even some basic common-sense logic, it turned out, is already crystallized in such a pile of statistics. To the naked eye, it really feels like magic to get a page of text from a one- or two-sentence prompt. Software developers, judging by recent blog writings, are also deeply concerned because LLMs possess programming language ‘stats’ in their gigantic associative memories.
Who should be concerned?
In order to answer this question, we need to understand what software development is. Even if one has been in software development for decades, one can easily miss one of the most essential attributes of what software is. But first, in no particular order, let’s list the usual, non-exclusive reasons why people are engaged in the practice:
- …earning money
- …solving people’s problems
- …having intellectual fun
- …etc.
From this, we see that software is usually not the goal, but one of the means to achieve something else. After getting clear on the ‘why,’ we better move to the ‘what’ and ‘how.’ Here is where the very common misconception lies, and there have been many attempts to debunk it. I do not want to list the misconceptions here and will go straight into discussing one often forgotten attribute of software, which is also key to understanding the kind of ‘danger’ AI brings to developers.
Software is about knowledge
For simplicity, let’s take a look at the DIKW (Data, Information, Knowledge, Wisdom) hierarchy. In short, if an LLM can write you a piece of software, it means that the LLM already possesses the knowledge needed for that, because humanity already had that knowledge, albeit buried in some obscure (or not so obscure) places.
The whole software industry produces a lot of code duplication for many reasons. Now we have a new kind of search engine, which helps us find what is already known. This means that there will be less mindless development, right?
Well, at first, there will be more mindless code and more duplication of the same knowledge. If you want an analogy, the pre-Internet diversity of network technology can be a hint. This will continue until humanity finds good approaches to optimize knowledge management for the common good. Think of Tim Berners-Lee and the ‘WWW of AI’ yet to come. Hopefully, this time it will not take so long to realize. Until then, there will be an era of old and new coexisting, solving old problems with superficially new means. This will last until automobile shapes no longer remind us of horses. And software development will finally leap to the knowledge level, no longer constrained by specific syntax.
Of course, those software developers whose expertise is in form (syntax, frameworks) and not in content (domain knowledge, including computing domain knowledge and perhaps mathematics as a highly compressed form of expression and a way to process technical knowledge) will suffer.
When I hear about things changing rapidly, new libraries and frameworks coming and going, and software developers needing to learn, etc., I see that all this is a concern for people who focus on secondary matters — the form of software — instead of the essence. Those who have been focusing on the content will feel liberated and productive in their development efforts. Because they were already doing things correctly.
My prediction is that LLMs are just the first step in building the tower of Babel of knowledge. Most likely, there will be efforts (which might already be underway) to store knowledge in a more condensed and explicit way than large language models. There will be (and already are appearing) tools to build ontologies — those backbones of knowledge. Or do you really think, for example, that the next ISO standards will be neural networks or 1500-bit vectors for some Common AI model? Dystopian scenarios aside, it’s wise to still have knowledge in a concise form and human-readable text.
And by the way, there is also W for Wisdom in the DIKW.
Conclusion
Knowledge and know-how will determine the winners in software development in the near future. LLM technology can help in dealing with the form and recorded knowledge of humanity, but unique knowledge and experience, combined with wisdom, is the compass to navigate the newly found ocean. This includes understanding the laws of technical systems’ evolution.
As usual, there will also be a lot of efforts aimed at achieving fast, short-term profits, as well as jumping onto an uncoupled car moving by inertia.
