AI will bust in the same way that the tech boom did

If I told you today that the internet was a failed venture, you would rightfully call me an idiot. Technology is what drives innovation and economic progress within our world. Frankly, the last 25 years has perhaps been a larger leap in productivity and financials gains than the rest of human history combined. The late 90's tech bubble existed because investors didn't know how to invest and tech companies didn't know how to provide value.

AI is another rendition of a technology bubble. Data centers are popping up across the world and companies are spending trillions to build them. Companies are attempting to automate virtually every aspect of their business with little success. Not much revenue is being created, but the more important metric is that not much productivity is being found along the way either. There are theoretical gains, but we don't eat theoretical food. We need something tangible.

AI fails from a recursion problem. This is extremely well documented within technology and children learn the problem at an early age. Play a game of telephone. One person says something and many other people try to repeat it down a line. By the end of the line, the message has changed. AI fails because of this phenomenon. Some kind of looming model is training itself on data that is being funneled through it's own computing. As it takes in data and spits out results, those results continually become data that it is then taking back in. Over time, the model actually becomes worse rather than improving.

It is important also to understand that AI will never be creative. It simply regurgitates what is in front of it rather than creating something new. I'd rather not get too theological in this post, but humans maintain an ability to create that is functionally unmatched.

How does any AI model cancel out the noise or learn the fundamentals of a process when it is learning from so many sources which might be junk? The reality is that it isn't learning properly. AI will fail for quite a long time because technology companies have too short a timeline to understand how these models can succeed.

You must first program a solid foundation. To do that, you must control the inputs. It cannot just be a random scraping of billions of data points.

You must then program the knowledge for more specific tasks. Many tasks require "judgement" that has many different variables to consider with weighted reasons per variable. This is the complexity that must be considered. Weighing variables isn't simply mathematical outputs. It requires a deeper understanding of why each output exists while considering subvariables and layers of depth along the way.

This type of training takes a tremendous amount of time in both humans and machines. AI must learn and grow similar to how a human develops. Unfortunately, AI is even more difficult to train than any human because it accepts all inputs at face-value. It cannot decipher ulterior motives or nuance in a way that a human can.

Large companies believe that they can aggregate data from active businesses. This is a fools journey. Without context, nothing and no one can decipher such madness. I believe that a real AI model will require a "trainer". This is an actual person to walk through jobs and establish foundational norms within their specific trades. A machine can learn faster than others, but it must receive guardrails and interact more honestly with someone who can force it into a better direction.

This will take decades to accomplish. Investors don't want to hear that they are in a massive bubble with a technology that only has minimal current applications. Likely the biggest fault of this bubble is that programming is the #1 industry to see immediate productivity gains from early AI usage. Every other industry is slower to adopt it, requires far more nuance, and will ultimately rubberband away from it in the short-term as the models cause more review work than they are worth.

Time will tell if I am right or wrong, but I am skeptical every time that I interact with a company pushing AI. I know a hype train when I see one, and I'd rather walk to my destination.