AI
"AI needs large amounts of energy":
This keeps coming up, the idea that AI need lots of energy, and as a consequence, it is necessary to increase power generation to provide the tech.
This is only partially true. AI in itself does not need energy, it is simply information, and processing of information. AI does not store or use energy. The demand for energy is a result of current computer technology being inefficient and thus waste a lot of energy from thermal losses in semiconductors.
It can not be ruled out that improvements in technology will improve efficiency quite considerably, and thus there will not be any need for this power, for AI.
Have been working on the web pages, using it to look into AI-supported programming. And this works very well, this is the future.
Kilo Code is now working on the code, while I write this.
I no longer need to write code, I can just orchestrate and give instructions, think about how things should be, give instructions.
This tech is still very young, and that shows. It isn't very smart, it makes a lot of mistakes. But compared to before.. The productivity gain is immense and really is a fundamental shift in how software is made.
This is our future.
To instruct and think about how the result is going to be, while software generates and creates what we imagine can work.
We no longer need to have insight into all the specific details of implementation, it is all handled automatically. We can focus on systems and functionality.
This shifts our mental focus on to the higher level, enabling us to be more productive and create more value with less effort.
But it does not replace skills and expertise. The automation can not create something that is new and unique, it can not see purpose.
AI automation replaces tasks, not jobs.
Evolution in steps.
Advancements in technology happen in steps, it is not a linear progression. Current AI hype and evolution was triggered by the invention of LLM, which enabled software to understand human language by transforming what we write or say into data patterns which then can be used by the software to perform operation. Such as searching for text, generate images.
What happened next was an evolution on use of that technology, with the creation of tools and agents, tools that utilize LLM AI models and work on top of those, making an interface between the user and the AI models.
It is not necessary to be a software programmer to use the technology. It started with the chat interface, where we simply write questions and then get some sort of answer. Now there are many providers of AI models, of tools to build AI agents, of tools that use AI agents for a specific purpose.
And things change very fast. It was at first a challenge to run your own local AI model on your computer, now there are applications like LMStudio and Ollama that can do this, no programming required.
The coding agent I use to generate and write code is an example of practical use of this AI technology. It works, sort of, but certainly has its limitations. It is an immense boost in productivity.
Generally, everything we see now in AI is based on the LLM models and is affected by their characteristics and limitations. We will continue to see tools and agents and applications improve, while the underlying AI will always set its fingerprint on how those tolls perform and work. The next step in evolution will be AI models that can reason and have logic. Current attempts to fix those issues are only patchworks.
Limitations.
Coding with AI reveals clearly limitations of this technology based on the LLM.
It is not accurate, it does not produce consistent results, in some case, when it doesn't know or unable to find a working solution, it just makes something.
It struggles with design, obvious you may say because it can't really see. It is not good with layouts.
It is quite good when there is a well-known, established solution. Trying to have it code something more special is more challenging, and increased complexity, especially if it involves logic thinking, is hard.
Nature of AI is that it is not accurate, it is approximate, it sees patterns rather than exact numbers. AI has been around and in practical use for some time, before chat-AI became a thing, in applications like image recognition and analysis.
A good example is the image I have of the inside of my knee, a MR-scan. I can not see anything useful form that, the doctor could see something, after looking at it for quite some time. AI can scan and analyze such images in short time, and come up with likely answers. answers that are equally good as what the doctor can make out. Because this is not an exact analysis to begin with, the answers are more like "it is likely that this part is destroyed..", "..yes, surely here is the problem.. perhaps..". It is a problem involving patterns, and the outcome states likely results, rather than a specific, consistent answer in numbers.
The LLM opened the world of software automation to the public, sort of. And the inherent faults and limitations becomes problematic, because the AI does not necessarily provide a correct answer, and that is not always obvious to the user.
No, new technology does not take-our-jobs..
Never did. Advancements in technology is what improves our world, because it increases productivity and thus quality of life in general. I mean, just look at history.. How was life, before technology.
What AI does, is change jobs and how things are done. But that is not new.
I am old enough to remember a time when you had to physically go to the bank to pay your bills. A time when, if you were out of the house, no one could reach you or you could not reach anyone. Mobile was not really a thing until around mid-90ies.
So did someone at the bank loose their job? If so, was that a bad thing, overall? Want to go back?
It was advancements in technology that made those changes, they happened because it was possible to do, and because the new solution offered improvement.
We are now in the middle of an automation revolution, driven by the possibilities of new AI technology. Some jobs will disappear, to be replaced by something new. This will drive huge improvements in productivity, making services and products much cheaper. And it will open up new opportunities for business.
For those who want to venture into AI automation and supportive businesses, there are both opportunities and possibilities for huge failure. Making software based on current AI tech is risky. It is very likely that it will be obsolete before you accomplish the launch of a finished product.
And if you put all your effort into learning "prompt-engineering", you are in for a surprise; your newly acquired skill is soon to be rendered obsolete.. (Not completely, at least not yet, but for common user-level it very soon will be.)
Can now install and run AI models on Ubuntu, where you can chat with it (why? to see if it works, perphaps..) and through API make it part of something useful you make.
The future computer will have AI models installed as part of the system, then applications will use whatever you have on your computer as an engine to do cool things.
Decision-making:
AI is suitable for solving problems that involves large amounts of data, and where the outcome is not overly complex in logic and does not involve complex mathematical constructs. And where the outcome, if done manually by a human, is not one specific correct answer.
Writing code for a website is such a task, well, can be.
Translation of a text between languages is another suitable task. The resulting text can be slightly different, but still be equally good for its purpose.
Decision-making may seem like something we should never let automation do. Especially important ones, that affect our daily lives and affect a lot of people. But when we look into a decision-making process, it becomes clear that the process in itself, has properties that makes it very suitable for AI automation.
The important part is to decide what outcome we want from a decision. Then we use available information to make a decision that we believe is most likely to give that desired outcome.
This process in-between can be automated, and then it can be done better than manual, because the process itself can now handle much larger data and is not affected by bias in the same way.
This has some huge implications for how society and work and professions will evolve in the future.
Management roles will change, to focus on outcomes. Mid-level management will disappear. This will lead to more cost efficient and productive businesses and organizations.
Eventually, AI will replace politicians, eliminating the largest obstacle to human growth and prosperity. People can vote for outcomes, rather than corrupt personalities. This will change how we manage our societies.
Why we don't like AI. Or, more precise: What we don't like.
Everyone not into computer tech was presented to the new AI tech when chat-gpt launched, Everyone could now ask a computer for answers, and actually get something meaningful back. Students quickly discovered it could write their assignments. If there was a problem you had no idea how to solve, you could just ask, and get a text which explained how it works and how to do it.
Then there was image generation, just type a description and have an image of anything imaginable. Soon this evolved into video generation, no surprise, since video is just pictures in a sequence.
Text could be generated automatically. It could write stories, no need for authors anymore.
Then someone pointed out, there are limitations. It is not "smart", really. And it can provide answers that are wrong, problematic if they are presented as factual, and the receiver of the information lack knowledge of the subject at hand, so can not really tell if it is correct or wrong.
The images had a distinct character, and sometimes obvious flaws. "AI image".
(to be continued..)
The point I am trying to make here, is that AI is not suited for creative work. If you believe AI can replace an artist, you do not understand what an artist is.
Automation can assist creative work, making one artist much more productive. But that is different from having AI make endless posts and videos and images to be put out on the net, this is content with no value, it does not contribute to anything, it is just noise.
We say no to this. It already has a new word; "AI-slop".
Easy money is promised; just make a system that generates 1000s of posts and images, all automated, and see revenue coming in. But it doesn't work, because this content has no value, so no one is interested in watching it. Coming next, we will see the death of social media as we know it today, as a result form this. Though, maybe that is actually a good thing..
For management, it is tempting to believe employees can be replaced with AI. We have already seen some unfortunate consequences, where companies end up with products and services that no longer provide the value their customers paid for, and thus, they leave.
(continues later..)