Machine Learning Intro
AI/ML is all about automation - teach machine how to learn and then let it take control. And learning is about feedback loops.
graph LR
practice --> experience --> theory --> knowledge --> practice
The feedback here is the line from theory to practice. It is called positive loop, because you add knowledge. With added knowledge you can practice new things, and the loop goes on until you don’t learn anything anymore. Your knowledge is stable now. If you don’t want your knowledge to fade, you need to repeat the loop.
This was automation control theory in 30 seconds. Everything from Elon Musk’s SpaceX rockets to air conditioner in your room is powered by such feedback loops.
graph LR
t[measure temperature] --> b[below target] --> increase
t --> a[above target] --> decrease
increase --positive feedback--> t
decrease --negative feedback--> t
Monkey in the Loop
When I am not switching tabs in Firefox, I am fixing some problem, coding, or trying new ideas. And my loop in Linux is basically the following.
graph LR
p[get problem] --> t[open terminal] --> r[read error] --> e[edit file]
e --> execute
execute --> r
And most of the time, depending on the error I get, I do some operation that I already did before. So I usually hit ctrl+r (history lookup) before writing my command.
For the AI based Linux, I should somehow give AI access to errors, file and safe execution environment to test run such loops for me. I just want to click confirm button to keep my dopamine monkey entertained.
Once my monkey passes through a few of such automated loops with supervision, there is no way for it to get back. The reason I am writing this is because I already passed through such loops in my imagination, and now I want it to become reality. )
Linux Augmentations for AI
First component is some kind of shell (GNOME, i3, …) supersystem. I don’t know how to implement this (yet). The goal is to let you and AI both see the same system and interact with it in parallel, sometimes replacing or correcting each other.
I see the UI for this as a Quest Log. Like in RPG games. Consecutive entries, each entry has subsequent stages, each stage has tasks. The “Quest Log” records the problems, what needs to be done, simplifies it to tasks, shows the progress and records actions.
The Quest Log, or the list of “error”, “goal”, “action”, “result” entries is a source for one-shot training data . Each entry is one shot. “One-shot” means you need to repeat something only once for the system to learn it. This makes it possible to switch between different AI/AGi networks in the future. Filter irrelevant errors and stuff like in git history. This database of your RPG experience can be copied and then reuploaded into new improved AI model architecture. In the end the goal is to run this network locally.
How much data should be included in one-shot context - I can not explain. Rather than feeding everything to AI, I like the idea to extend the context on demand in a natural interaction. The AI could start with the same context as you - the error message. Then gradually figure out what is missing on the way. So the learning process should be same as humans, without AI superpower to see everything. Just to make sure we trust each other, and can grow and update each other in parallel.
Where does this come from? (some distant notes on human touch)
I went to LEAP conference in Saudi to speak with big companies like Aramco about how do they approach Open Source sustainability problem? The Open Source is the core layer of our digital infrastructure. All AI solutions are based on top of it. Internet is impossible without it. Open Source provides tremendous value to each tech industry by cutting costs and eliminating waste. But we see that the value benefits don’t reach the maintainers.
The code is only alive in the heads of maintainers, and without the means to support them/us, we lose the code, the expertise, but the most important - the fun and the community. The feeling of well-being in people depends less on the income, but more on if you have the people who understand you around.
When AI automates everybody in two years, how do we all earn money for living? What should be our gameplay? Who is working on planning that, rather than waiting for the disaster to happen?
These were questions I asked at Aramco Digital’s booth of nourOS. “If there is somebody who works on that gameplay for humans well-being, I want to be the part of it. If not, then I want to create it”. I filled the form, writing some strange things there, and it got lost somewhere in the depths of the internets. Maybe I got my email wrong.
The nourOS is basically the AI that gives you the prompt to replace clicks, typing and UI work that people do every day in hundreds of thousands of Enterprise applications to get the jobs done. I asked what kind of Python Notebooks and libraries they use, if their AI uses Open Source model, if the OS is the real OS that runs on top of hardware. Of course, people at high level conferences usually don’t know such details, but in general I’ve got the impression that this is Linux based OS that runs on hardware, and yes, it uses all the AI stack, which is de-facto Python. I told that we can not solve the Open Source sustainability problem, because there are no bridges across canyon that lies between finance, CEO, ESG, CSR and people writing valuable Open Source code outside a company. I exchange I’ve been told a story about one brilliant AI person with no college/university degree, who learned everything about technology from Open Source community.
As authors of Open Source software, we can not solve Open Source sustainability problem on our own. It is an impossible challenge for code hackers, the elusive challenge for legal hackers, and even hackers in finance are not able to solve it. That’s why we need more hackers from different areas to define the future we aim for, and use AI, practice and simulations to get there in the incremental improvement way.
Why do I speak about Aramco? Last year I didn’t even knew this name. The oil and gas companies know a lot if not everything about maintenance of physical infrastructure. The economics and gameplay of pipe maintenance is beyond the simplistic “bottom line”. It requires metrics, skills and budgeting that are not market models. And I expect people from Aramco Digital (behind nourOS) to become that skilled in the maintenance of our digital infrastructure. In the end our goal as humans is not to “earn all the money”, which we print anyways. Even if the by definition, the goal of commercial company is to exactly to earn as much as possible while not being slashed by regulators. With AI one single company could just win all the money quickly and that’s it. As humans we need a smarter goals now. Does building the capacity for the Earth maintainers sound good?
The conclusion
(the negative feedback loop)
When big corporations take over the Open Source, I stop to see real people around me anymore, and that brings heavy emotions about my own future. In my 42 without family, without a job, sponsored by my friends, I don’t feel like I am a good example for my nephews, or anybody else at all, and that’s not attractive to girls as well, so probably no children. Very very sad.
(the positive feedback loop)
To make a better future, we need to practice it. If our future is maintaining our planet, and enjoying the life as the maintainers, we need to get the metrics of what the work-life balance is, what the feeling of well-being is, and continuously improve our systems. Analyzing, coding, testing, and deploying in small changes. The gameplay doesn’t need to be the survival game, and we have all resources to change it for the better. Even if the survival drive is the only thing that pushes us forward. I don’t know how to get there, but I know that should stop on the path. Not alone, but with others. So no matter where you are, and what you do, there is hope. And no, hope is inappropriate word here. “If there is a will, there is a way”, - that is how we get there.
(the AI/AGI/ML thing)
So how do we integrate AI things in Linux faster to free our time and talk more bigs things like that? 
P.S If anybody read the whole text, plz. tell me that I am sane - I am afraid to reread it to confirm the inevitable.