Why the AI Bubble will burst

There are many current articles out about how AI is a trillion dollar bubble on the verge of bursting.  For example, this article is a recent Bloomberg article about the bubble.

There are three key aspects to this AI bubble: the business hype, the technical reality and the historical perspective.  Let's look at all three areas:

1. The Business Hype.  Clearly the stock market loves AI because they love the smell of money.  This is the same frenzy we saw in the Dot Com bubble but with more strategic implications.  Every country in the world knows that AI dominance will be a key strategic advantage when military drones and robots are doing most of our fighting so the race is on! 
   One interesting point about the Business hype is that it had both a Positive ("AI will make you Rich") and negative ("AI will steal your job") spin!  The fear mongering only increased the business frenzy because anything powerful enough to steal your job or destroy the world is worth investing in! LOL
   
   Result:  They Business Hype and "Gold Rush" aspect of AI will explode.  Many people will lose money and many startup companies will go down in flames.  That is the nature of hype cycles and this one is no different.  You notice how not many people talk about AI achieving Artificial General Intelligence (AGI) any more?  Yeah, that pipe dream (or nightmare depending on your perspective) was just part of the hype.  
   
   
2. The Technical Reality.  Large Language Models (LLMs) are a leap forward for AI.  That is undeniable and they are useful tools in many situations as long as you don't trust their outputs.  They are just a neural network trained on billions of pages of Internet data to spit out the next word in a sequence.  They know nothing and none of their output can be trusted beyond being an opinion of a stochastic machine.  Many people call LLMs a "stochastic parrot".  This is a fitting term.  As I have said in a previous article, the answer will be Semantic AI that will blend the best probabilistic techniques with the best deterministic techniques to deliver explainability, trust and reliability to AI.  
   
   Result:  In the near term, Agents will mostly be toy examples or parlor tricks as LLMs cannot be trusted to be autonomous.  Neural networks will continue to make progress by being fine-tuned in specialized domains; however, this is still not a trustworthy solution so they will also fail but it will take longer to be called out on it. Choi's paper entitled, "Off-the-Shelf Large Language Models Are Unreliable Judges" is particularly salient and exposes a serious LLM flaw.
   
   
3. The Historical Perspective.  We clearly did not learn our lesson from the Dot-Com Hype of the late 1990's (see: Dot-com bubble - Wikipedia).  Just because lawmakers and layman don't understand software, they must not fall for the snake-oil salesmen! It is unlikely that this wave of AI hype will be any different than the previous waves of technological hype - short-term euphoria, crazy predictions of revolution and finally things settle down into evolutionary progress.

Overall, it is a good thing that this bubble will soon burst...  because then we can get back to making software more reliable instead of pretending that LLMs are good programmers.