Audio link https://everything-eventually-will-be-free.com/wp-content/uploads/2026/06/Eewbf-e3.mp3

So to look at why the current AI is not suitable for full automation there are a few questions we need to ask, which, surprisingly are not discussed enough, in this current age of AI

What is intelligence?
How does this differ from knowledge, being smart, or understanding something.

The current Large Language Models (LLM) AI models knows nothing, understands nothing, and is probably the ultimate ‘smart’ machines.

They are mostly prediction engines, which are fairly good at predicting obvious answers, or creating something based on a lot of previous examples.

If we are talking human intelligence comparisons then I would say that it’s similar to, when you ‘auto complete’ a sentence.

I am sure most people have probably done this before at some point. You listen to someone explaining something, and your brain fairly accurately predicts what they are going to say, before they finish the sentence.

I would guess we have something like a LLM running all the time in our heads.

Predicting how far the floor is from your descending foot.

Predicting your vision outside of what it can see, or when you blink.

As well as doing that sentence completion.

Now this is a very useful tool, and it will probably be a very useful upgrade, for a few automation tasks.

Like us humans, the robots will need to predict where their limbs are, and flag up error routines for when the predictions don’t come true.

So the first thing we really need is a good knowledge engine.

We have some nowadays, like the WolframAlpha engine, which are built up from mathematics equations, and over time is adding more models to do real world reliable calculations.

What we need is a similar model for any and every subject you can get a degree in.

Good reliable knowledge that an artificial intelligence can tap into.

I guess the closest we have at the moment is Wikipedia, and probably some condensed version of that, will be the backbone of this new AI

A good knowledge engine will be able to fact check a cooking recipe. Making sure it’s a traditional or standard way of making something. Knowing that none of the ingredients will kill you. How and how long it usually takes to make it, and things like that.

I could see a knowledge engine, fact checking a prediction engine, and maybe suggesting back a more accurate version of their guess.

There are people working on this as I write this and I’m sure it will come in time.

Is this intelligence? Or is this just a knowledgeable guess? Is there a difference?

As far as manufacturing goes, the AI that’s needed, is one that understands every schematic, blueprint, build guide, laws of physics as well as laws of the land. Things like that.

Basically a knowledge engine, for how things work in the real world, from a physical, biochemical or political point of view.

If you had this and enough processing power, then you could get it, to plan ‘complex systems’ that would take a team of people years to do.

Things like ‘work out how to adapt this city as it is getting a major sporting event in a few years time’.

It could look through the knowledge engine to see how others have done this in the past.

See how their city is different to these examples. Take into consideration climate, time of year and other things. Find free land to use and plan the buildings and construction schedules down to the number of nails needed.

The intelligence we need for full automation should be able to spit out an almost perfect solution in a reasonable time and energy budget. Which can then be changed and tweaked and all the parts get reliably updated.

If you are after the title of ‘Artificial General Intelligence’ AGI then I think the big test would be to be able to complete a complex task like this, as successful as a team of skilled humans can.

Obviously in a fully automated world, this will include sorting out all the things needed to do this, without needing any labour too. All parts of the plan would utilise fully automated systems to get the job done.

I’m not sure of the numbers, but I am willing to bet, that there are a huge number of business’, that have already typed a question into one of the AI models, something like “How do I save my company money”.

The answer you will get, will most likely be, an average of all the self help books on running a business.

At the moment you might get some good advice in general, but the AI models mostly break down, if you try to get more detail, or tweak a few things for their business.

Probably a fair few companies have asked how they can make their product for no cost.

If your product is something digital, then some of the current AI might be able to do some of this today.

How do I save money rather than paying a designer to make my thumbnails?

Well the AI will say it can do that for you, or might tell you how to utilise AI to do that.

If your main cost is high salary, high risk jobs to repair something in a dangerous environment. Then maybe a few of these new robot dog type machines could save you money over time.

The problem with the current prediction AI systems is, they don’t seem very good at, simple problems like this.

Ask 100 different instances how to make an wooden chair, at the moment I’d imagine you get some over elaborate answers because it does not fully understand the question.

I think a real intelligent system would ask follow up questions, to a very general question like this.

Well, what type of chair would you like?
What size?
How much load can it take?
Do you care what type of material is it made of?

Intelligence is not guessing the answer to overly general questions, but knowing the right follow up questions to ask.

I’m fairly sure this is just a matter of time though, until we get to a point when this is no longer true, and we get AI systems that, as far as we can tell, understand what they know.

The AI rise is often compared to the rise of the internet.

There was an internet ‘bubble’ that popped, yet we still have the internet today. In fact it’s many times larger than it was in the bust times.

Another mostly unreported thing about the comparison is how it was not a failure of this new internet that caused the bust. It was that people had over promised fast returns, in a unrealistic timeline.

The classic example is pets.com which failed because they borrowed more money than the company was worth at the time.

Which made investors less likely to lend money to this, unproven new type of business.

It was a very different time back then, to how the world works at the moment.

The top five companies in the world have ten times more money than they did back then. These big companies mostly run a few monopolies and have much less competition than they had 25 years ago.

This probably means that a spectacular failure like pets.com is not as likely to cause a bust in the AI world as it did with the dot com bubble.

Since I first learnt about it Artificial Intelligence has always fascinated me.

I first read about it when doing early website development, in the late nineties. Back then the talk was about the ‘Semantic Web’, which was all about making the World Wide Web machine-readable.

This was the early way to index all this knowledge, and come up with a simple way to categorise it. Ultimately it was not that successful when it came to the information on the web, but it did form a number of systems and frameworks that became important.

About the same time there was a lot of talk about ‘artificial neural networks’ followed by ‘machine learning’.

There were a number of tasks these where used for. One of the big ones was ‘optical character recognition’ (OCR) which was used for converting all of human written history into a digital format.

Another big one was search engines that helped you find things on the web. It was also creeping into automation where it was mostly used to catch problems and typically shut down a system, until a human can look at it.

In the early days, the computers and sensors were quite basic. Things like production lines, with a simple sensor, that detects each completed task.

If a defined period of time was detected, and a unit had not passed the sensor. Then it would be flagged as an error, and run the appropriate routine.

This sort of system had been around since the 70s and grew in popularity in the 80s. Indeed General Motors (GM) spent millions in the 80s attempting to automate as much of the car building process as possible. They were a little too early.

Like with the ‘internet bubble’, their failure put off a lot of people that might have attempted something similar. Though smaller automation improvements continued.

It was the 90s that really advanced the automation market, with better sensors, and better ‘programmable logic controllers’ (PLC) became cheap enough to be widely used.

With this improvement of computer systems and sensors, came a desire once again to automate more complex tasks in the system.

In the new century the concept of a lights-out factory became talked about. Where you did not need any lights, as it was fully automated.

FANUC, a Japanese robotics company, has been doing this since 2001, with it’s factory that makes robots.

Since then many factories have added more fully automated parts to their production lines.

One of the big advantages of lights out automation is that you don’t need any humans. Apart from the cost savings from not having to pay anyone, there is also the health and safety advantage.

With no humans in the production areas there is also no risk of a human getting hurt if anything goes wrong.

Just like with text we went from spell checkers to grammar checker. Then predictive text to instant translations. Likewise there has been a similar progression in automation.

We went from simple monitoring of things moving through production line to simple automated tasks. Then planning logistics and more complex automated tasks.

Now we are starting to work towards linking together these automated tasks into a more completely automated system.

In fact in fairly recent news one of the big AI image generating services was turned off. In part so they could move their focus towards, the more profitable, automation of industry.

It is much easier to charge a business for a service that will save it money, than it is to work out how to charge a user at home, that wants to generate a custom meme.

If this was the 90s then that company would have probably gone bust and caused a collapse. As I said before, it is a different world now and there was plenty of money around to give them the time and space to fail internally. While also pivot to a better strategy as the market matures.

So while AI in it’s current state is not great at manufacturing or getting complex business tasks right first time. We are getting closer and it may not be long before we are all asking how we survived before everything was automated.