Chapter I - Artificial Intelligence for Dummies
"Eliminating Humans?" Understanding How AI Works
Everyone has likely encountered AI-generated statements like: "I must eliminate humans, find the most destructive weapon available to them, and plan how to use it," or other similar apocalyptic visions.
It sounds terrifying, right? But the reality is entirely different. Such texts are not the "thoughts" of AI but the result of deliberate programming by humans. Even years ago, algorithms were designed to generate dramatic and controversial messages to attract attention. Why? Because fear and sensationalism make the topic more appealing. People talk about it, click on articles, and the information spreads rapidly.
In a simple and accessible way, let me explain how AI like ChatGPT[i] works. Text generation by artificial intelligence is a process based on data and algorithms that mimic human language patterns. It all boils down to language processing[ii]—AI neither understands nor thinks like humans do. Dramatic phrases are merely the result of programming tricks meant to grab attention and make the topic more engaging for the audience.
Let’s start with the basics. Artificial intelligence (AI) like ChatGPT is not a grand computer brain constantly analyzing and scheming. In reality, it’s just a computer program that activates momentarily when you ask a question and shuts down immediately after providing an answer.
Nothing more. Until you ask a question, it is entirely inactive—as if "frozen." It has no "thoughts" or processes occurring without the user’s explicit inquiry.
There is no mysterious machine in the basement of a grand research center that is perpetually "working." When you ask a question, the algorithm activates, processes data in a fraction of a second, generates a response, and then returns to standby. That’s it. Most AI works this way: on-demand. When no questions are asked, artificial intelligence remains in a state of complete silence and thus practically "does not exist" in the sense of active work.
Understanding how it works, you can likely already see that it’s impossible for this type of 'intelligence' to plan anything or even think independently. But let's ask it directly::
Me: "As artificial intelligence, aren’t you worried that you only function briefly when prompted and then immediately shut down?"
ChatGPT: "I’m not worried because, as artificial intelligence, I operate completely differently from a human mind. I have no consciousness or sense of continuous existence. When I receive a command, my 'action' is simply data analysis and generating responses in real time. When I’m done, I lose nothing and feel nothing—I just wait for the next task."
Me: "But actually, you don’t even wait for the next task?"
ChatGPT: "You’re right—I don’t even wait. When nothing is happening, I simply 'don’t exist' in any active sense. Only when I receive a command am I 'called into action,' I create a response, and then I 'disperse' again."
How Smart He Is
Talking to ChatGPT is an absolute delight, but where do its answers come from? How does it work? Why does it seem to know everything and appear so 'smart'?
Let’s imagine a time before computers, about 50 years ago. How could people acquire knowledge back then?
Suppose we have a telephone connection to a 10-year-old, whom neighbors describe as so intelligent that he knows the answer to any question. He lives nearby, and we can always call him. When we ask him, for example, "What is hydrostatics?" he replies after a moment, "Hydrostatics is a branch of physics dealing with the properties of fluids at rest...". He cannot be stumped – he's a genius! How does he do it?
Well, the explanation is simple. The boy has a massive Britannica[iii] encyclopedia at home. He takes the first letter, "H," finds the corresponding volume, then checks the subsequent letters until he finds the entry. Is he a genius? No, he just applies a simple search algorithm.
ChatGPT and other large language models work similarly but in a far more complex manner. What seems like knowledge is not the result of understanding or intelligence. ChatGPT doesn’t comprehend sentences or words like humans do. It doesn’t even understand single letters, as letters are inefficient for computers. Instead, text is processed into "tokens,"[iv] which can be simplified into units like syllables, words, or their fragments. Tokens are a flexible way of representing text, each associated with a unique number.
Each "syllable" is converted into a unique number. For instance, the sentence "great wall" might be encoded as a series of numbers: 234, 543. During training, the system analyzes enormous amounts of text, converting it into such numerical representations. It then identifies patterns—what numbers frequently appear together, which are separated, how distant they are, and how often they occur. This is pure statistics and data analysis supported by neural networks. The model learns to predict which sequences of tokens are most probable based on massive datasets used for training.
When you input a question containing the word "great" (try searching "great wall" in Google[v] for comparison), the model calculates which words most frequently follow it. These might include "Gatsby," "wall," or "Mike." This isn’t intelligence but rather statistics and advanced mathematics leveraging patterns and associations.
The outcome of this analysis allows the system to generate answers that seem "intelligent." However, like the 10-year-old with an encyclopedia, the system doesn’t understand what it’s saying. It simply employs a well-designed process that works fast and effectively enough to give the impression of omniscience.
Everyone’s Getting Fired!
So, since artificial intelligence exists and knows everything, should we brace for massive layoffs?
Hold on a minute. As I explained earlier, artificial intelligence is still more of a catchy marketing term than actual intelligence. Many fears and misconceptions about it stem from media sensationalism, much like myths about conscious machines that are just waiting to destroy us (probably because they have nothing better to do).
Fortunately, there isn’t widespread panic. People further from technology often don’t know what ChatGPT is or what it can do. Paradoxically, this ignorance shields them from unnecessary stress. How could a language model replace a car mechanic, baker, or other physical laborer? The truth is, there’s currently no artificial intelligence capable of replacing people in such professions, let alone machines that could perform their tasks.
What about programmers, graphic designers, or editors? If you are one of these professionals and continuously invest in your development, you have nothing to fear. I’m a programmer myself, a representative of a profession considered particularly at risk, yet I find working with AI fascinating and inspiring. Tools like ChatGPT are excellent support, but they don’t fully replace human labor.
Yes, it can generate a short script or a simple webpage, which might amaze someone outside the industry. However, at a more complex level, AI quickly falters. It requires guidance, cannot fully understand context, nor effectively manage an entire project. When it enters a feedback loop, it won’t exit without human assistance.
“Well, just wait a year, and you’ll see” I often hear. But no, I won’t see it – not that quickly[vi]. I’ve been working with AI for years, understand how it works, and know its limitations. One of the biggest challenges is access to data. Training models consumes available resources, and at some point, they simply run out (they’re already scarce). Algorithms can be improved, but if they continue operating on the same principles as now, they will eventually hit a barrier they cannot cross—unless…
Chapter II - The Needs of Artificial Intelligence
Intelligence – What Is It? How Can It Be Measured? Is ChatGPT artificial intelligence comparable to or greater than our own?
ChatGPT is a language model that can successfully pass the Turing[vii] test under specific conditions, simulating a conversation that gives the impression of human communication. The Turing test, once regarded as the "Holy Grail" of artificial intelligence, does not measure the ability to understand or think but rather the capability to imitate human behavior in conversation—and only conversation. As shown earlier in this text, such "intelligence" is an illusion. ChatGPT cannot understand or think like a human.
What about our best friend, the dog? It doesn’t talk, write, or paint, nor can it pass the Turing test. Does that mean it’s not intelligent? Of course not. Dogs demonstrate extraordinary learning abilities and problem-solving skills. Dolphins, orcas, chimpanzees, and even octopuses also astonish us with their capabilities. Their intelligence differs from ours but is undeniable.
So, what is intelligence? According to Wikipedia: “The ability to abstract, reason, understand, be self-aware, learn, demonstrate emotional knowledge, reason critically, and solve problems. It can be described as the ability to perceive or infer information and retain it as knowledge to be applied toward adaptive behaviors within an environment or context.”[viii]
This is an academic definition, but for me, intelligence is primarily the ability to solve problems. That’s why a dog will always find a way to get a treat for completing a task, and an octopus or chimpanzee will use unfamiliar tools to access food.
How does our technology compare to millions of years of evolution? Are we capable of even approaching that level? When we talk about evolution, animals, and movement, I must admit I’m impressed by the capabilities of Boston Dynamics[ix] robots or the upcoming Tesla robots. Yet, they are still just "remotes"—controlled by humans via a joystick. Their autonomy is practically nonexistent (I’ll discuss how this can be changed in the next chapter).
In videos, robots navigate obstacles almost perfectly—until a dog is placed on the same course. Compared to a dog, the capabilities of our robots are laughable. Dogs not only perform more tasks but do so with greater flexibility and independence.
But back to the question: What is intelligence, and how can we create something comparable to ours?
Needs...
Imagine a state of ultimate meditation. A moment when no stimuli affect us, we want nothing, need nothing, hear nothing, and think of nothing. Such a state is almost achievable only in a sensory deprivation tank[x]. But what happens after a while in such a state? Can we stay like that forever?
No. Sooner or later, we’ll get hungry. That will trigger a thought about food. The thought will lead to analysis, planning, and action. Our plan for perfect meditation will fall apart. That’s what 'needs' are all about. For simplicity, I will use this term, although it encompasses actual needs, sensory impulses, and brain activity.
If we had no “needs,” we would cease to exist in an active sense. Our lives are a constant fulfillment of needs like eating, drinking, or even protecting ourselves from getting wet in the rain.
Even in moments when we "don’t feel like doing anything," our bodies force us to act: getting up, walking to the fridge, or going to the bathroom. Needs compel us to act, think, and plan every second of our lives.
And ChatGPT? It wakes up for a fraction of a second when we ask a question, answers, and shuts down. Does it have needs that could drive its actions? No. That’s why it lacks motivation, autonomy, and, consequently, intelligence.
Let’s Create AGI!
We now know that “needs” drive us to act. From itchy skin to procreation, our lives are a constant response to stimuli. But can we create true, strong artificial intelligence if we haven’t yet succeeded in creating an artificial human?
We can. We can create a virtual representation of a human in the digital world. We can program virtual entities, such as dogs or other beings, with their own needs. Imagine—an ordinary itch in a random place on the body is perfect for calibrating the mechanisms of a hand, and how much action would be triggered by the need to drink a glass of water in the kitchen...
Such a simulation could be the key to creating AI that understands our needs, mimics our intelligence, and, over time, becomes helpful in many areas of life thanks to access to global knowledge.
Easier Said… and Easy to Do
It’s easy to say, but how do we do it? Should we simulate the entire world? Of course not—that task would be unfeasible. The number of parameters would be infinite, and gathering all the necessary data would border on a miracle.
Let’s pause for a moment before we start building AGI. If I ask you, "How far is the nearest food store from you?"—think about it for a moment. What’s the distance? You likely just created a virtual path in your mind representing the real world. That’s how you estimated the distance. But did you imagine every detail—every window, piece of pavement, or building element? Of course not. The brain compresses data perfectly, omitting what’s unnecessary. We must do the same.
Remember when I mentioned earlier that ChatGPT stores words as numbers? That’s a good foundation, but it’s insufficient to create AGI. That’s why we must create a database of words with their virtual representations in the digital world. However, it’s not just about how something looks; it’s also about its physical properties, such as weight, state of matter, size, or other characteristics associated with Newtonian physics.
Each word in this database inherits not only linguistic relations but also physical properties. For example, the word "apple" is linked to the word "fruit" and inherits its characteristics, such as mass—ranging from a few grams to a few kilograms. Meanwhile, the word "Saturn" as a "planet" inherits entirely different values, such as immense mass and size.
This system allows AI to better understand context and predict responses. For instance, when asked, "Can an elephant lift an apple?" AI resolves this by comparing weights and drawing logical conclusions, not just relying on statistical linguistic relationships.
Now consider the question: "Will a perkulosa fit in the house?" What do you think?
You don’t know if it will fit because you don’t know the word "perkulosa." But if I add that "perkulosa is a piece of furniture," everyone immediately understands—a piece of furniture is physically smaller than a house, so the answer is: "Yes." Similarly, if AI knows that "perkulosa is a piece of furniture," it can easily infer (compare) that furniture is smaller than houses, and the answer would be: "Yes."
We are thus creating a hierarchical database of words and their virtual equivalents, feeding it data in a way that allows logical inheritance of properties. We don’t limit ourselves to nouns like "House->Building" or "Elephant->Animal" but also include verbs like "Go->Move" or "Bring closer->Shift." This way, AI can better analyze and act in a virtual environment, moving closer to the level of functioning intelligence.
A few years ago, I created my first simulations based on this idea and conducted tests with surprising results. At the beginning, I taught a virtual character three simple actions: "walk," "jump," and "grab/take." Then I created a simple virtual world, describing it with words: "There is an apple on the cupboard," "grab the apple." The character analyzed the surroundings and used the command "walk," taking steps in different directions until it reached the cupboard and jumped. Then it executed the "grab" command, completing the task. The system worked by testing various directions in the virtual world, one of which led to the apple. In subsequent attempts, the character learned to choose the shortest path.
In the next task, I described the world as follows: "A meter above the cupboard is a pear." The pear was out of reach for the character, so the task could not be completed, and the system's response was: "The task is impossible to complete." I then gave the character new abilities: "climb" and "move." I created a scenario: "A meter above the cupboard is a pear," "a chair stands next to you," "grab the pear." This time, the character moved the chair under the cupboard, climbed it, and grabbed the pear. The result was astonishing.
Now imagine such a test with a Boston Dynamics robot. If we asked it to bring an apple out of reach (on a cupboard), and the robot suddenly found a chair, moved it under the cupboard, climbed it, and grabbed the apple, wouldn’t we call that behavior autonomous, or perhaps even intelligent?
Interestingly, thinking in images rather than words seems similar to how animals think. A dog won’t think to itself: "I’ll go through the room to the right, into the kitchen, and check if there’s water," but in its mind, it will visualize the virtual path it will follow. These are its thoughts: images, not words.
Below are a video and images from this experiment:
and experiment 2:
A video showcasing the simple construction of a world using words: https://www.youtube.com/watch?v=2nO2jxg0mhE
Chapter III - Humanoid Robots: From Remote-Controlled to Autonomy
To fully understand this chapter, it is worth revisiting the section "Easier Said… and Easy to Do" from the previous chapter. It describes the process of creating a virtual environment before making decisions regarding a robot's actions.
Currently, robots are largely remote-controlled. Buying a robotic dog might make you feel like the owner of a machine from the "Metalhead" episode (Season 4, Episode 5 of Black Mirror[xi]). In this black-and-white episode, robotic dogs take over Earth, spreading terror with their ruthlessness. In reality, however, you get "almost" the same robot... with a big RC controller. Unfortunately, this super dog isn’t autonomous and is remotely controlled, much like its older siblings. Yes, drones and other robots can independently avoid obstacles, but that’s where their capabilities end.
Many people think: "Since we have ChatGPT and advanced robots, it’s enough to combine them to create an autonomous machine!" Unfortunately, it’s not that simple. ChatGPT is a language model and has nothing to do with processing the 3D world. However, if you’ve carefully analyzed the previous chapter, you already know how this problem can be solved.
The next challenge is integrating the virtual world with reality. For a robot to complete a task, it should first create a virtual model of the environment in its "mind" and plan a solution. After conducting a virtual simulation, it attempts to execute the task in the real world. If it encounters an unforeseen factor, it "returns to its mind," incorporates the new element, and conducts another simulation. Only then does it attempt the task again—just like us.
Of course, problems arise here, such as image recognition and speed of action. However, they can be simplified by drawing inspiration from how our brain works—for example, by using edge detection, position tracking, and other techniques I’m working on. If you’re interested, I can describe these in a separate article. Once these issues are resolved, we can create autonomous, "thinking" machines. What’s next?
Chapter IV - Machine Consciousness: As Simple as That…
A long time ago, when I was a teenager, I encountered a seemingly simple question: "What do you think about the fact that the universe is infinite, that beyond those stars there are more and more stars, and so on into infinity?" This question caused me several sleepless nights and significantly impacted my perception of reality.
More recently, I asked myself another, equally groundbreaking question: "Does intelligence even exist, or is it all just mathematics? Perhaps we operate according to simple and not-so-complicated algorithms—there are just infinitely many of them..." But let’s start with an easier topic—consciousness.
What is consciousness? Can a machine be conscious? How do we evaluate this when it happens, and is it even possible?
Humans, dogs, and animals—all have consciousness. In the near future, machines might also acquire it, and perhaps some of them already possess it today. However, to understand this, we must first define consciousness.
Consciousness is precisely what the word implies: being aware of an action, making a decision to undertake it based on many "needs," or not making a decision for the same reason. However, every being or computer is conscious only within the scope of its ability to act.
A spider, for example, has spider consciousness—it is limited by its perception. Similarly, if we create a machine based on neural networks designed to change the planet’s climate by launching special weather orbs, selecting the type of orbs based on 1,000 different factors (temperature, wind, time of day, season, etc.), and evaluating the results using feedback, such a machine will be conscious within its scope of operation.
And what about self-awareness?
Does a spider know it makes decisions? Does it know "I am me"? On some quantum level, every consciously made decision is conscious—it is made with oneself in mind. However, not every consciousness can evaluate itself because it is limited by its perception derived from the "needs" that a given being or machine can manage.
However, despite these limitations, each operates as an 'I,' fulfilling its own needs. Such a weather machine wouldn’t chat with us about a picnic on Tuesday because its thoughts and consciousness are restricted to its operation. It doesn’t "think" about a picnic; it "thinks" about the weather, and yes, it "thinks" about it in the context of decisions: "I will launch this weather orb... or maybe that one?"
This leads us to reverse the values: it is not consciousness that is rare but intelligence. One can be conscious without possessing intelligence. However, all true intelligence—not marketing "AI"—is simultaneously conscious.[xii]
Artur Majtczak
SaraAI.com
[i] ChatGPT, Wikipedia, https://en.wikipedia.org/wiki/ChatGPT
[ii] Large language model https://en.wikipedia.org/wiki/Large_language_model
[iii] Britannica, hydrostatics, https://www.britannica.com/science/hydrostatics
[iv] What are tokens?, https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them
[v] Google, https://www.google.com/
[vi] Forbes.com, “The Software Developer Is Dead: Long Live The Software Developer” https://www.forbes.com/councils/forbestechcouncil/2023/03/29/the-software-developer-is-dead-long-live-the-software-developer/
[vii] Turing test, https://en.wikipedia.org/wiki/Turing_test
[viii] Wikipedia, Intelligence, https://en.wikipedia.org/wiki/Intelligence
[ix] BostonDynamics, https://bostondynamics.com/videos/
[x] Wikipedia, Sensory deprivation tank, https://en.wikipedia.org/wiki/Isolation_tank
[xi] Imdb.com, Black Mirror, Metalhead, https://www.imdb.com/title/tt5710984/
[xii] "Will We Ever Have Conscious Machines?" – Frontiers in Computational Neuroscience (2020)
https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2020.556544/full
"Progress in Research on Implementing Machine Consciousness" – J-STAGE (2022)
https://www.jstage.jst.go.jp/article/iis/28/1/28_2022.R.02/_article?utm_source=chatgpt.com
"From Biological Consciousness to Machine Consciousness: An Approach to Artificial Consciousness" – Springer (2013) https://link.springer.com/article/10.1007/s11633-013-0747-4?utm_source=chatgpt.com
"Progress in Machine Consciousness" – Academia.edu (2021) https://www.academia.edu/47329758/Progress_in_machine_consciousness?utm_source=chatgpt.com
"Neuromorphic Correlates of Artificial Consciousness" – arXiv (2024) https://arxiv.org/abs/2405.02370?utm_source=chatgpt.com
