Does AI feel pain?

There is much talk these days about AI eliminating roles and simplifying day-to-day activities, but is this true, and if so, can we rely on AI to completely replace jobs without human input?

Ask AI a question and it will give you an answer, virtually instantly, and whilst it is often accurate it can give some unexpected results. This suggests two things: first, that AI ‘knows’ many things, and second, that it is necessary to check the output to ensure it is meaningful. This blog will look at the first of these ideas, namely, what method does AI use to ‘know a lot’, and what can we learn from this?

There is a Russian proverb that says, “Without effort, you can't even pull a fish out of the pond”, the equivalent of “no pain, no gain”, which seems, with the advent of AI, to be less relevant these days, at least in some spheres of life. For example, how does AI learn to develop software so well, and is it a simple, ‘painless’, activity? In summary, AIs have a machine learning phase where they read vast quantities of data, spot patterns and then convert these to relationships. Interestingly this is the same process humans use, although a human will use less learning data and use other relationships it has learned previously to create the necessary relationships that we would then call experience.

So, both the AI and the human learn, and whilst the AI has no concept of ‘pain’ it does need to go through the same process as a person. Returning to software development, many mid and senior developers generate much of their code with the aid of AI, seemingly eliminating the need for the ‘pain’ part of the process. And it is reasonable to suggest that it is their experience that enables them to determine how efficiently the AI generated code solves the problem.

In fact, the use of AI to generate code is now so prevalent that there has been an associated drop in the number of junior development roles. Apparently, hiring a ‘junior’ and developing them into a mid/senior developer is not currently in vogue. Perhaps there are so many seniors out there that offering a ‘competitive package’ means you don’t need to experience the pain of having to train someone. Your senior can use AI to generate basic code, then tweak it with their ‘special skills’, and voila, no need for those pesky kids.

This seems odd on many levels and speaks to a decline in a requirement for quality and a focus on efficiency without reference to anything beyond tomorrow. This new route generates results in the short term, but it does make existing seniors, regardless of their relative expertise a valuable commodity, that in time will become scarcer, leading to a further decline in quality and the loss of the efficiencies that drove the whole process.

Here’s the problem. The ability to interpret code and determine its efficiency and elegance, depends on experience. That experience comes from consistently and repetitively putting some effort into learning, practicing and honing a new skill – experiencing the pain of things not working, of sometimes failing, and of finding ways to make things work. Without this process the person does not create the mental pathways to figure out how things work, and what works best. Without experience, a developer might instead be able to interpret pre-written code and solve basic problems, but will find it hard to develop new ideas without external input.

Time, effort and intense focus can solve many problems, but practice, a lot of failure and determination are also required. Just looking at a problem for two minutes, thinking “I can sort of see the solution” but then using AI to provide that solution, is the same as the surgeon who learns by reading textbooks. It is necessary to understand the theory but essential to practice skills, spend time thinking how to do something, figuring out why some solutions won’t work, and why others do, and above all, spending time ‘doing it’. These activities build ‘experience’ which you don’t get by reading or copying; experience is gained by actually doing things.

Presumably, pre-AI, most of those senior developers became seniors by starting from scratch and doing things, but where will the new seniors come from if there are no juniors? This is a perplexing problem and until it is resolved will only serve to heighten the reliance on AI generated code, and a more homogeneous and less varied output.

But that’s another story, and one we’ll discuss in our next blog: Creativity, AI and Cubism.