Natural Intelligence (NI) is the Barrier for Artifical Intelligence (AI)
By Sebastian Edinger
This text is predicated on the precarious assumption that AI will remain dependent on human intelligence for the foreseeable future, i.e., for at least the next 25 years, meaning it will not be capable of maintaining itself independently (let alone evolving). Self-maintenance in the strict sense would entail that AI was able to autonomously maintain and control the supply chains of its production and maintenance.
While AI remains far from production-technical autonomy due to the immense complexity of supply chains, both graduated autonomy—that is, the ability to self-repair, including through organizing the supply of necessary parts—and effective resistance against its own abolition appear possible within the next 15 years.
Some will consider this premise—conceptualizing AI development in 25 years as fundamentally (rather than merely marginally) dependent on human achievement—unnecessary and soon to be anachronistic. However, we must reckon with two fundamental possibilities—the autonomy and self-sustainability of AI—rather than wearing the prophet's mantle. I maintain an informal wager with a friend: What will prevail—AI or dysgenics? Will we find through AI a means to reverse dysgenics once more or bypass it by employing means of genetic enhancement, or will dysgenics progress too rapidly, leaving AI as a meaningless triumph, functionally irrelevant in a functional zoo? About two years ago, I placed my bet on dysgenics, though I now see it as an open race.
Above all, one must consider as a possible success scenario for AI that it does not prevent dysgenics but rather shapes it—that is, as Aravind Srinivas, CEO of Perplexity.ai, prophesies, runs our lives.
What does this mean? If AI runs our lives, it fundamentally means that it will lock people into itself, take over and micro-manage their daily lives, if necessarily in a paternalistic manner, make them dependent upon it, force companies to lock themselves into it as well if they wish to maintain relevance and market reach. What Adorno termed the "administered world" (verwaltete Welt) will become literally true at the micro-level; being subject to obsessive helicopter parenting in comparison will look living the life of Robinson Crusoe. We would no longer be genuine humans then, but we would cling to a teleological self-deception that could not be simultaneously more comical and tragic: We would permit AI to manage our lives in the name of efficiency, regarding it as an optimizing addition to our existence, while conceptualizing ourselves as the telos of a process whose own necessarily empty telos renders almost everyone superfluous.
The Disturbing Natural Intelligence Aspect
Should Western societies experience full-scale dysgenic regression—where IQs drop to an average of 90, with Germany advancing at an almost incomprehensible speed toward such a cognitively catastrophic state (as explored by Rindermann 2018 & 2024)—one might perversely interpret this as a form of grotesque relief, an AI takeover would constitute a grotesque relief. In the unlikely case of demographic recovery (see the fertility patterns among the best-educated on page 237), this would amount to an administered zoologization of humanity.
But where do we stand presently, and why does natural intelligence constitute a barrier not only to the potential advancement of artificial intelligence in the coming 10-15 years but, maybe even more significantly, to its utilization? Is AI not already something that benefits merely 5% of society and can be employed responsibly and—pun intended—intelligently by only a minute minority?
Let us examine some disturbing data, in particular the general OECD statistics from PISA, where we observe, in the presentation of mean values and development across 81 countries, a uniformly horrifying picture.
In mathematics, the situation appears thus:
Reading, the most fundamental of all cultural techniques, is on the path to becoming an esoteric capability of ominous outsiders.
Furthermore, the great prospects in the natural sciences have been continuously diminishing since 2014 (not that surprising the Zuckerberg is intending to delegate a huge bulk of Meta’s programming activity to AI, right? And 25% of Google’s code is now being AI-generated. Microsoft is also ambitious in this regard).
Similarly, particularly with regard to the United States, an "eternal pattern" emerges that is either surrounded by an awkward silence or by the Corybantic din of denial's roar. If this pattern is not immediately addressed and corresponding political measures are not implemented to counteract its unyielding and alarming constancy, the West will soon face collapse regarding both education and societal development.
We observe a nation that, should the elimination of Affirmative Action and establishment of a serious meritocracy be earnestly pursued, is heading toward a condition where—if one deals honestly with the data, actual results, and their inherent prerequisites—two distinct educational systems would be necessary. As long as demographic Americanization, a project of hopefully departing globalists, is not reversed in Europe, this applies to Europe as well. Shouting "racist!" changes nothing, and one ardently hopes that the kindergarten of moralization will now close throughout the former West in the course of its re-Westernization.
Computer skills – a rarety
But what about the specifics? Let us examine computer skills. Jacob Nielsen, in his essay The Distribution of Users' Computer Skills: Worse Than You Think discusses four different levels of computer proficiency. In total, 215,942 people were tested, with at least 5,000 participants in most countries, across 33 countries. The sample size suggests that the results are reliable, although methodological errors can diminish their reliability. The results appear as follows (average numbers are provided):
26% are incapable of using computers at all and have never attempted any of the tasks by which competency measurement was conducted. 14% of the adult population in the USA is at "Below Level 1". What must one be capable of at Level 1? "An example of task at this level is 'Delete this email message' in an email app." 29% of all users are at Level 1. We are already at 69% of the total pool. Level 2 includes tasks such as: "An example of level-2 task is 'You want to find a sustainability-related document that was sent to you by John Smith in October last year.'" Here, another 26% is added to Below-Level 1 and Level 1, so we are already at 95%.
Only 5% remain who can fulfill the following requirement profile:
"At this level, tasks typically require the use of both generic and more specific technology applications. Some navigation across pages and applications is required to solve the problem. The use of tools (e.g. a sort function) is required to make progress towards the solution. The task may involve multiple steps and operators. The goal of the problem may have to be defined by the respondent, and the criteria to be met may or may not be explicit. There are typically high monitoring demands. Unexpected outcomes and impasses are likely to occur. The task may require evaluating the relevance and reliability of information in order to discard distractors. Integration and inferential reasoning may be needed to a large extent."
In light of the description, the task is rather simple: "You want to know what percentage of the emails sent by John Smith last month were about sustainability."
This situation assessment is an average picture based on data from 33 countries, and 5 percent is a magic number, as 125 is the IQ benchmark for the top 5 percent (Western Europeans and White Americans). In a study on reading proficiency among white Americans, extensively presented by Linda Gottfredson in her remarkable essay Why g Matters: The Complexity of Everyday Life, merely 4% achieved the highest of five levels. This finding illustrates the profound implications of cognitive ability distribution in everyday tasks, even among demographically homogeneous populations. The supposedly advanced countries only marginally improve this picture, as Nielsen points out:
"In the United States, only 5% of the population has these computer skills. In Australia and the UK 6% are at this level; in Canada and across Northern Europe the number increases to 7%; Singapore and Japan are even better with a level-3 percentage of 8%."
When one considers that the quality of usage of the "old Google" was already considerably stratified along the intellectual capabilities of users, one must ask what implications this holds for AI, which can hardly be used at an advanced level (i.e., as a work and research tool) by even 5% of people. AI is a tool for a tiny minority, as only a tiny minority possesses the cognitive resources and correspondingly sophisticated interest when it comes to exploring AI at all.
For those who are depressed now, I have an even more substantial addendum.
Jeff Atwood, a programmer with extensive professional experience, including in hiring, who knows other experienced programmers with similar experiences, summarically stated the following in his text Why Can't Programmers.. Program?:
"But I am disturbed and appalled that any so-called programmer would apply for a job without being able to write the simplest of programs. That's a slap in the face to anyone who writes software for a living."
Dan Kegel, whom Atwood quotes in his essay, even says:
"A surprisingly large fraction of applicants, even those with masters' degrees and PhDs in computer science, fail during interviews when asked to carry out basic programming tasks."
And now, let us not forget that, through whatever madness, the Bachelor's degree is considered a real academic degree (before the Bologna Reform, today's Bachelor's was considered merely an "intermediate examination" and only the Master's was considered a university degree). That the Bachelor's is what in martial arts is termed a make-believe belt is shown by Longbing Cao using the specific example of Data Science:
"As some surveys show, some 70% of qualified data scientists hold at least a Master's degree in data science related disciplines." (Longbing Cao: Data Science Thinking. The Next Scientific, Technological and Economic Revolution. Cham: Springer, 2018, p. 343)
Serious expertise requires both longer-term concentrated and broader work in a field. In a military analogy: Hitler was warned against invading the Soviet Union because the Soviets possessed the decisive advantage of "defense in depth" (Tiefenrüstung) on their own territory, which could neither be improvised at a distance nor adequately countered. (General Georg Thomas was right.) And intellectual defense in depth results from the combination of talent and persistent effort, the latter understood as the combination of: time (in this, Malcolm Gladwell is right) and grinding continuity in working your way through continuously more complicated problems. – An academic analogy: It is no different in philosophy: People who know only Analytic Philosophy or only certain disciplines in philosophy (they exist in abundance) can now easily obtain professorships, but they are a plague of ignorance.
Yet not only do we have a competency and talent problem, we also have a severe problem regarding our demands on ourselves: Degrees do not generate competencies. When standards are loosened (in effect, loosening them means abandoning them), one has lots of IT professionals plus an IT skilled worker shortage. Degrees have no weight of their own; the institutions that award them stand for what the certificate states. If the institutions fail to adhere to rigorous standards, the certificates aren't worth more than used toilet paper.
AI could prove an invaluable tool if and as long as it remains the tool, rather than us becoming its pets. The non-pets will, however—we must face this bitter truth—be a small minority who must prove themselves humanly capable of not wielding AI as a weapon against the rest, even if merely as an essentially harmless domestication weapon. Without demographic restoration, current AI development will be largely worthless, but here on a geopolitical as well as on a national level: Only few will be able to derive adequate benefit from it, and we should damn well stop lying to ourselves about it like foolish children who define lofty ideals but are incapable of grasping and acknowledging elementary realities.
But the next generation will likely not be prepared for accomodating to what their predecessors have inherited to them. To conclude with a list of examples and a graphic that must give us hope that AI will save the young AI saves the young, instead of them decisively developing AI further:
This graphic is taken from the article Poor student achievement and near-zero accountability: An indictment of Illinois’ public education system – Wirepoints Special Report by Ted Dabrowski and John Klingner, where it is also summarily stated (my emphasis): :
"For sure, plenty of Illinois’ 860 school districts achieved better student outcomes in 2019. But a cursory glance at state report card data shows higher performers were the minority. Just 89 districts had at least 60 percent of students who could read at grade level. In 168 districts, by comparison, less than 25 percent of students could read at grade level. Overall, less than 40 percent of all students in Illinois were proficient in either reading or math."
Have in mind also:
– „In 26 school districts, no low-income third-grade students were proficient in reading.“ (https://www.illinoispolicy.org/low-3rd-grade-literacy-is-warning-for-future-learning-earning-potential/)
– "Not a single fifth grader at Martin Luther King Jr. School, in the shadow of the Fruit Belt neighborhood, tested at a proficient reading level in 2022.“ (https://www.investigativepost.org/2023/02/22/buffalos-abysmal-reading-scores/)
– „Last year, almost 60% of California’s third-graders, the students most deeply impacted by distance learning and other Covid disruptions, could not read at grade level.“ (https://edsource.org/2023/amid-californias-mounting-literacy-crisis-state-names-new-literacy-directors/687430)
– "According to the National Assessment of Educational Progress (NAEP), a sector of the U.S. Department of Education, 84 percent of Black students lack proficiency in mathematics and 85 percent of Black students lack proficiency in reading skills." (https://thehill.com/opinion/education/579750-many-of-americas-black-youths-cannot-read-or-do-math-and-that-imperils-us/)
– Oregon has already capitulated: „Oregon high school students won’t have to prove basic mastery of reading, writing or math to graduate from high school until at least 2029, the state Board of Education decided unanimously on Thursday, extending the pause on the controversial graduation requirement that began in 2020.“
If you want, you can easily find dozens more such examples, which you have to express in numbers, only to remain incredulous about the reality they reveal. Let's hope that AI gives us a soft landing, or at least a controlled decline we can come back from (if we start preparing for it now) The problem is at least as much societal in nature (anti-intellectualism and frivolous hedonistic complacency, ubiquitous disregard for cognitive ambition in partner „selection“, if you even want to call it „selection“ with a straight face) as it is a pressing political problem (politically, all of this would have to be rigorously countered if zoologization is to be stopped).
References:
Rindermann, Heiner. Cognitive Capitalism: Human Capital and the Wellbeing of Nations. Cambridge, UK: Cambridge University Press, 2018.