Artificial Super Intelligence (ASI) can never be. There will never be an entity in existence that is “better” than a human in every possible way. I make this claim with complete confidence. This blog today is my evidence to support why I am saying this.
I came to this realization recently as I was considering another problem: how can humans be compared to other humans? That is, in terms that often lead to issues of racism and sexism, how is it reasonable to consider one individual human to be better, in a general or overall sense, than another individual human?
I believe that no individual is better (or worse) than any other individual, in a general sense. That is, I can compare two individuals using a test or examination, and I might be able to suggest that one is superior within the very narrow realm of the test or examination. When I was in school, I would be graded in my classes, and the mark I received in some classes was greater than the mark received by other students, suggesting that perhaps within the realm of that one class, I was superior to those other students. But even then, it gets tricky to affirm such a statement.
Instructors in a class teach certain material based on their own bias (and the bias of the curriculum and school system). This means that the specific material covered by the class is already incomplete. This makes sense as it is not feasible for an instructor to cover absolutely everything there is to cover about a particular topic in the limited amount of time they have. Further to this, toward the end of such a class, the instructor will go through a process by which they will assess the students to determine whether the students had acquired the desired knowledge and skills for that class. The manner of testing and examination is also highly biased and limited in its ability to accurately assess the knowledge of the student. For example, a multiple choice exam question could have been answered correctly by the student because they understood the question, or they could simply have been lucky by utilizing the 25% probability that they could guess the answer.
It is not my intention right now to delve completely into the challenges of educating individuals in various topics. My intent here is simply to show that the process is wrought with difficulties and challenges that make the process of assessment incomplete and potentially inaccurate as well. Even if I receive a higher grade in a class than another person, to suggest that I am somehow better than that other person in that subject is already highly suspect.
The best one can do, on my view, is suggest that I did better on the specific exam or test than the other individual. If on that very specific final exam in that very specific class I happened to get the higher grade, I might say with reasonable accuracy and confidence that I am superior to the other individual at that very specific exam. Outside of that, any claims I make about being better are simply committing the fallacy of a faulty generalization.
Taking all of this to heart, it seems obvious to me that individuals can only be compared in a very, very narrow sense to one another. I can be compared to another in only very, very specific domains. To generalize that I am somehow better than another in an overall or general sense would be infeasible. Not only would I need to go through absolutely every possible specific domain (like somehow performing calculus on people), but I would need to take into account the fact that my proficiencies within all these very specific domains actually changes over time as well. Every moment that passes, I learn something new and improve upon myself. And similiarly, I also forget things and become less proficient over time as well. So all these specific domains of intelligence or skill are constantly in a state of flux.
To sum all this up, unless I was able to take a snapshot of myself in this very specific moment, capturing with complete accuracy every minute detail of myself, and then also doing the same with another individual, and then also performing the step by step comparison of each and every specific domain, performing a sort of calculus in summing the values of all comparisons, might I be able to finally assess my superiority to that other individual (or lack thereof) within that extremely narrow moment in time. This is not at all feasible. This would require a certain level of omniscience, in addtion to incredibly fast computational ability.
It is for these reasons that I also believe stereotyping and predjudice, such as sexism and racism, is entirely unfounded and logically incorrect. Individuals cannot be compared to each other in these general ways, nevermind the clear and obvious issue with suggesting that the possession of a penis somehow imbues the bearer with superior intellectual capacities. (As an aside, as one who possesses such a thing, from my own first hand experiences, I would actually argue the opposite anyway.)
You may have noticed that I tried to focus my discussion thus far not necessarily on the fact that I am comparing humans to other humans, but instead individuals to other individuals. This was quite intentional. I have discussions with others about my pet rabbit and comparing his facilities to my own. As with my preivous post, discussing different ways of experiencing the world, to compare myself to my rabbit has other significant challenges as well.
The most significant challenge is that there will be certain domains that I possess that my rabbit does not, and vice versa. There will be points I may wish to compare that have no analog on the side of the other individual being compared against. My rabbit does not have an opposable thumb; any domain of comparison I wish to examine will automatically fail if it is related to having a thumb. To make matters even more interesting, to suggest that the having a thumb is somehow itself superior (or inferior) is to make a value assessment which itself adds a level of subjectivity that is like throwing a monkey wrench into the works. Who can say, in a purely objective sense, whether having a thumb is better or not? As with everything else I have been talking about, there likely will be certain very specific situations where having a thumb is advantageous, and other very specific situations where having a thumb is disavantageous. And again, we are in this situation of trying to perform crazy calculus and summing up results, just like earlier.
This is where I now bring into this discussion artificial intelligence (AI). I do not even need to raise the issue of a machine consciousness, as it is unnecessary to this discussion. Any AI system we wish to examine will be sufficient to refer to as an individual for the purposes of comparison. And like my pet rabbit, those AIs will have various features that can be compared to my own features, and various features that have no analog to me. (And I have features with no analog to the AI as well.)
This applies across the board! That is, there will never be an AI that possesses every single feature that I have, and specifically nothing more. Because if such a system existed, we would not call it an AI at all; it would simply be another human. To be able to compare to such a degree, it would need to possess the same sorts of features as I do presently, including a carbon-hydrogen based structural matrix, because this is what I possess. If we want to place AI into the category of machines, with silicon based structures, then there is already a first feature that we are unable to compare properly.
The point I want to make is that AI are different to humans. I do not think anyone would even want to dispute this. Like apples and schoolbuses, these are things that are very, very different. To compare an apple to a schoolbus is nonsensical already; what are we trying to compare?
Thus, to suggest we can compare AIs to humans in any logical or reasonable sense is highly problematic to begin with. The terms Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI), and ASI are themselves highly problematic already. At best, I can suggest that an ANI is focused on a singular task, an AGI is focused on all tasks (in some sort of general sense), and an ASI is also focused on all tasks (in some general sense). But even that is problematic. When Google is working on self driving vehicles, the AI system being utilized is referred to as an ANI, and yet driving a vehicle is not really a singular task at all.
To play chess requires a variety of skills. There is no one thing that makes a chess player a chess player, let along a good chess player. And even the best chess AIs do still lose occasionally. They are not perfect, just as humans are not perfect either. Given a particular moment or instance, they may shine, but in another instance they may fail. In a term used in AI programming, these can sometimes be referred to as “hallucinations.”
The term ANI is intended, I think, to refer to a system that focuses in a narrowish domain. Driving a vehicle or playing chess are complex tasks, filled with many subtasks, but are still rather narrow when compared to the seemingly infinite number of other domains that could be considered. But then, if we acknowledge that there are a seemingly infinite number of domains that could be considered, under what conditions could we refer to a human as being a “general intelligence” either? No human engages with or practices absolutely every single domain possible, so why do we think an AI might either? The term AGI is already highly problematic.
An ASI is simply an AGI which it has been decided that after performing the crazy calculus on the entire range of domains results in it being decided that the AI is better than a human in all cases. This makes no sense at all.
The point of my blog today is simply to say that there will never be an ASI, because the very idea of an ASI makes no logical or reasonable sense. The term itself is meaningless. But saying all of this isn’t to suggest that an AI couldn’t possibly take over the world or destroy all of humankind. It does not require a “super intelligence” to come along to absolutely monkey wrench all of human existence. The systems we have in place presently are sufficient to cause us great harm already. In fact, the very simple evidence I can present to you that this is all possible is the mere fact you are reading this blog.