Thoughts on the AI Hype Wagon

 I've been around long enough to feel like an old fart sometimes, but this has allowed me to see the same trends play out over and over - which means I can recognize the pattern.

First, Java was going to take over the world.

Linux was going to quickly kill windows.

The dot com boom and but.  

Microsoft .net mania.

4g (and then 5g) was going to revolutionize everything.

Web 2.0.. something something.

Cryptocurrency will make government issued money and gold obsolete.

Now AI will make humans obsolete.  

What will happen here is the same as what happens every time.  There will be some value in most of these things after the hype dies down, but it will be nowhere near as much as predicted and it will take a lot longer than anticipated.  

For example, I was using Bitcoin when it was still experimental and well before the average person had any idea what it was.  Back then, local mining on a 486 CPU was something one could reasonably do, but few people actually believed it was anything more than a computer science experiment.  It took more than 10 years for most people to hear about it, and that was only because flashy startups with venture capital funding spent tons of money for advertisements that pushed it as some kind of "investing".  

As the hype took over people started blindly parting with their hard-earned money to buy what is really just solutions to arbitrary mathematical problems meant to be difficult to solve.  Sure, cryptocurrency is scarce, but so are pieces of paper with my handwriting - that doesn't make them particularly valuable.  

The first problem with AI is that it is poorly defined.  The term Artificial Intelligence is being used as a catch-all for everything from image and text generators to machine learning systems that fine-tune parameters, solve equations, predict chemical reactions, or analyze medical images.  These are all differing things with differing amounts of usefulness.  

 Machine learning for things like predicting chemical reactions or predicting cancer from x-ray images are very useful if they work, and it's easy enough to test how well they work given enough time and data.  

Generative AI is a totally different thing, and testing it is a much more fuzzy affair.  The fact that you can chat with it is spiffy and all, and it can be impressive how life-like it seems - but that's also just a mirage.  Generative AI has no reasoning skills whatsoever.  It can take massive amounts of information and turn them into statistics, which can then be turned back into written word - which is amazing, but also fraught with peril.  

The worst part is that most users and potential users don't even realize this.  They are simply introduced to a seemingly magical tool, and they believe it can do whatever people tell them it can.  They don't know that it may make up facts, or that it is just a summary of data generated by humans that could be full of errors and bias.  

The point is this: 

1. Generational AI and domain specific statistical machine learning are, for the most part, two entirely separate things that shouldn't be confused.  

2. Generational AI can't "think", it doesn't have powers of reasoning, or any kind of symbolic logic ability.  I doesn't "know" what it's saying.  At the end of the day, it is a powerful tool to search and summarize the vast amount of data available online, but it's just that.  

For example: ChatGPT and it's ilk might be able to help a lawyer by searching for case files, but those results will need to be checked by a human.  It might be able to suggest a legal strategy, but whether that strategy actually makes sense, or applies in the jurisdiction in question is a totally different issue that, again - will need to be checked by a human expert.  

AI tools will become useful sometime after they become mainstream and the hype settles down, but mainly after people understand their limitations.  The current trend of "slap AI on everything" reminds me of how Microsoft started calling everything ".net" for years.  The current type of AI isn't going to save most businesses 30% of their costs, nor is it going to replace a large percentage of human jobs.  To the extent that it does replace them, those will be jobs that were drudge work anyway.