A person writing code in front of a robot

Here’s the code necessary to run a super-human AI from the year 2124

Huh?

Programming doesn’t have to be complicated.

I’ve already written the exact piece of code that will be able to run an Artificial Intelligence with super-human abilities from the year 2124. And I’m willing to share this with you, free of charge, completely open-source.

Ready?
Here we go.

Python
import ai
ai.load_model()
ai.run()

It is only three lines long.

It is readable.
It is sleek.
It is art.

What’s in a line?

These three lines of code are enough to run an incredibly powerful AI model in the future. In fact, these same three lines are capable of running a ChatGPT-like model right now as well. How could that be?

Depending on your level of expertise with programming, you are either very confused or only mildly so. Granted, this is very cheesy. You may have been clickbaited to think I had secret answers to the future of AI. That’s not what this article is about. I am serious about this however: the future of programming does not have to be complicated. This article is all about the power of abstractions. With these three lines, we abstract away all the low-level algorithms, the data storage and even model training, leaving the future AI Engineer more time to think about higher-level problems.

import ai

“Get your AI here! Flying drones, personal assistants, you name it, we got it! Don’t miss out on the future, folks, it’s all here!”

Python is such a powerful programming language in part due to the plethora of external libraries. With this single line, you give yourself access to a previously developed library, in this case a hypothetical one called ai, which you can assume based on its name contains code to run something with Artificial Intelligence. However, this line itself does not start anything just yet (unless it is arguably poorly programmed). Under the assumption that there exists some ai library that we can use to load and run some super-human AI model, we hereby grant ourselves access to it. Today, such ai library could already be created to support Large Language Models (LLMs).

ai.load_model()

We’ve given ourselves access to some insanely powerful (albeit for now hypothetical) ai library. To use this, some model must have already been trained. The model architecture including the parameters are currently already things we store in a single file, and this seems to be an efficient method of sharing various models. So, it is safe to assume this will still be the case in 2124. Based on the assumption that we have such a model file that was trained with super-human abilities (in the default, designated file location), we read and interpret the data of this model file with this one line of code. This moves the AI model from ROM to (V)RAM – or, if computer architectures change in the future, more generally onto the appropriate metal. In terms of today’s LLMs, this line could read a .gguf file on your HDD or SSD into the GPU’s VRAM, either directly or via Ollama.

“Run AI, run!”

ai.run()

This is where the magic is at. We’ve loaded the relevant code for running AI models, and we’ve loaded the AI model itself. Now, all that’s left is to tell your computer that you’re ready to start using the AI model.

“DomoArigato”: a household robot-o

For the hypothetical super-human 2124 AI model that we’ve loaded, this function could do a wide range of things. It might start what looks like a videocall, with a 3D rendered AI character you can talk to – one that can see you, too. It could connect to your DomoArigato: your future household robot that will now wake and do all kinds of chores around the house. Or, it could interpret text in your terminal and let you chat with it. The latter is actually not too dissimilar to how one might run a contemporary LLM.

Layers of abstraction

As you can tell, I am a big fan of abstractions. Back in highschool before I started my studies in AI, I never imagined I would be “smart enough” to be a programmer. Software development appeared infinitely complex and I had not a single clue where to start. At the start of my studies, we were taught programming in C. We created simple algorithms for solving sudokus but we had to make sure memory was properly (de)allocated. In my spare time, I programmed a Monopoly game as a single file of C code without dependencies. In my second year of AI, I saw a huge proverbial mountain ahead of me: “how many years do I need to study this before I can go from a sudoku solver to a fully functional and responsive website?”. Turns out, websites aren’t actually that hard to build. That’s because most of the work has already been done before, and we can easily reuse that. This blog you’re reading now, for instance, is a WordPress website inside of a Docker container. If you have Docker installed, it is no harder than running two commands in your terminal: one to run a database, one to run WordPress itself.

As programmers, we stand on the shoulders of giants.

By reusing code, we can focus more on the specific area of expertise that we bring to the table without doing much of the repetitive configurating or building from scratch. If you’re doing something that someone else might want to do too, odds are there is already code for that. There is rarely any need to reinvent the wheel.

Bonus: will people even remember Python in 2124?

This is an interesting tangent, if I may type so myself. Although the aforementioned three lines of code might be enough to spawn a virtual super-human intelligence in 2124, it is based on the assumption that the Python compiler is still commonly used in the future.

But– yes.
I do believe Python will remain popular for a long time to come.

Shut up and take my money!

Python is already the most popular programming language at the moment. It is a relatively easily readable language that is continuously being improved by the Python community into a full-stack language. The internet is filled with examples of it. And that is exactly what makes it so interesting at the moment. Since the release of ChatGPT, we’ve seen that AI models exist that can help develop code for us. The more examples of a specific programming language one can find online, the better a model can be trained to reproduce this language in a useful way. In short: the most popular programming language at the time ChatGPT is trained will see an increase in use, which makes it even more popular. It is a virtuous cycle. One in favour of Python. This is the reason that, if any high-level programming language is to survive until 2124, I’d put my money on Python[1].

[1] although I don’t think many languages are ever truly going to disappear (with the exception of MATLAB… a guy can dream)

Bonus: try out the AI for yourself!

I’ve open-sourced DomoArigato: the cutesy household robot I came up with while writing this blog post. Whether or not it is actually a super-human AI from the year 2124 is up to you to determine. If you ask DomoArigato, it definitely thinks so!

Head over to the GitHub repository if you’re interested in trying it out for yourself.


Leave a Reply

Your email address will not be published. Required fields are marked *