The Racist Myth at The Core of The AI Hype

-

It is almost impossible to avoid AI hyped claims whenever you go online or try to watch the news. This forces one, whether you want to or not, to engage with such claims. However, the more I tried to contemplate the wild assertions and predictions of the AI spokespeople, the more I began to recognize a fundamental fallacy, or rather a myth, at the core of the hyped claims. This foundational myth is what the evolutionary biologist Stephen Jay Gould describes as the error of reification, i.e. the propensity to convert an abstract concept like intelligence, which is complex and multifaceted set of human capabilities, into a single hard entity that is defined and quantifiable1. This is so apparent in the term Artificial General Intelligence (AGI), which is at the core of all the AI promises and prophecies.

The term “general intelligence” traces its roots to Charles Spearman, who in the early 20th century applied the statistical method of factor analysis to certain test scores and derived a single factor, which he called g for “general intelligence”. With no real knowledge of neurophysiology or genetics (much of that had not been discovered yet), and based solely on a statistical analysis of unnormalized test scores (he couldn’t have possibly accounted for all the social factors that lead to divergent test scores), Spearman confidently claimed that his mathematical g is evidence for a single reality, a general energy within the brain that underlies all cognitive mental activity1. Even now, with exponentially more knowledge about the biology of the brain, the physical basis consciousness and how it relates to our diverse intellectual capabilities is still largely a mystery2. Nevertheless, the reified concept of “general intelligence” became very popular, not because it was a rigorous, mechanistically founded concept, but mainly because it gave an air of quantitative scientific legitimacy to the already widely held racist beliefs of group hierarchies.  Since its inception, it was utilized by countless racists from Cyril Burt to Charles Murray to argue for white supremacy and classify groups based on their average g scores as if it was a real physical entity. Moreover, as it is always the case with racist ideologies, the goal was not only to assert superiority, but also to justify dominance and abuse based on this claimed superiority, a separate, yet compounding slippery slope fallacy.

The AGI bros seem to have been entrenched up to their ears in these ideologies. The panic that a superhuman AGI would enslave or destroy us3 can be traced to Spearman’s flawed g concept, that more computational power will eventually produce machines with more “general intelligence”. Such machines will be capable and entitled to dominate anyone with a lower amount of it, as the white males have the same right over other groups with lower average g. This God-like AGI will also possess knowledge inaccessible to even the smartest human scientists because the limited biological brain cannot generate enough g to compete with the unlimited silicon-based power. Jeffery Hinton, the god father of AI, uses an analogy borrowed straight from the racist literature, which has always compared lower g groups to children, by comparing humans to a superintelligent AGI in the same manner:  

"Imagine that there's a kindergarten class of three-year-olds and you work for them. They're in charge, and you work for them. How long would it take you to get control? Basically, you'd say, 'Free candy for a week if you vote for me.' And they'd all say, 'Okay, you're in charge now.' When these things are much smarter than us, they'll be able to persuade us not to turn them off. Even if they can't do any physical actions. All they need to be able to do is talk to us."4

The fundamental influence of the error of reification on the AI luminaries and visionaries is unmistakable. However, it also hints to a related flawed way of thinking, which is believing in something called “emergence” or “emergent phenomena”. It is the belief that, within complex systems, new properties that cannot be reduced to the individual components of the system “emerge”. Emergence is the flip side of “irreducible complexity, and both are non-scientific concepts, one is down top, the other is top down.  The success of science, in contrast to other magical ways of thinking that claimed direct insight into higher truths, has depended on reducing the complexity of the outside world: matter to particles; organisms to organs, genes and proteins; animals and plants to kingdoms and species. By reducing and organizing complexity via isolation, controlled experimentation and mathematical representation, science gave us enormous powers to understand, predict and control the very complex world around us. Nothing is more non-scientific than giving up on this power to reduce complexity into defined entities, laws, and interactions. We have Newtonian physics, relativity, quantum mechanics, thermodynamics, evolutionary theory, the central dogma, and germ theory to understand much of the complexity that surrounds us. These reductionist theories enabled us to go to space, split the atom, cure diseases and double our life span. Reification and emergence, that somehow something like superintelligence will just spring out of complex systems, is going back to mysterious, prescientific and unproductive ways of thinking.

Additionally, the concept of emergence has been extensively utilized by historicists, those who believe that history unfolds according to fixed, discoverable laws or patterns that newly emerge from complex social interactions. This was highlighted by the eminent philosopher of science, Karl Popper, in his criticism of historicism:

“In the world described by physics nothing can happen that is truly and intrinsically new. A new engine may be invented, but we can always analyze it as a re-arrangement of elements which are anything but new. Newness in physics is merely the newness of arrangements or combinations. In direct opposition to this, [historicism insists that] social newness, like biological newness, is an intrinsic sort of newness… It is real newness, irreducible to the novelty of arrangements… Even if the ordinary methods of physics were applicable to society, they would never be applicable to its most important features: its division into periods, and the emergence of novelty. [Popper’s Italics]”5

This led historicists, most prominently Karl Marx, to claim that they have discovered truly emergent novel historical laws that enabled them to predict the inevitable future of societies5, which sounds eerily similar to modern day tech prophets who also claim that they know exactly what the not-yet-invented technology will be and how it will dramatically change societies. Similarly, they try to align everyone around these predictions, and we know how terribly this went for dialectical materialism.

Finally, to complete the mythical trilogy, a God-like superintelligence with miraculous emergence of powers needs its messiah, the savior and the prophet that will take the stray humans out of darkness and warn them of doomsday. Here comes the role of the comic-book reading, video game playing young white males. With superficial knowledge of science and history (mostly from comic books, movies and documentaries), huge egos and big envy of history’s famous figures, they enter this invisible competition with other historical figures projecting some kind of male insecurity in the need to prove that one is the best that has ever existed. For this to happen, all the previous human achievements must be small and insignificant compared to what this man will do, even if it hasn’t happened yet. This is why you hear them talk about AGI in biblical terms as some kind of “rapture” or “singularity”, a defining moment in the history of humanity, where everything after it is different than everything before, “the most beneficial technology ever created”6.  Then comes the promises and warnings, the paradise of free everything for everyone7 or the hell of AGI eternal enslavement of the human race3. They conveniently ignore that fact that we are already living in an age of extraordinary abundance and welfare, which came about due to classical reductionist science and liberal democratic policies without any AI help. Knowledge, music and entertainment are almost free, and we can cure, treat or prevent most human diseases (including many types of cancer). We can travel anywhere in the world within hours, communicate with anyone in the world instantly, and watch visual recordings of our recent past. We already have robots that clean our clothes and our dishes; they are called the washing machine and the dishwasher (imagine how inefficient they would be if they came in humanoid form). Of course, there are still problems of equity and distribution, and diseases haven’t been cured yet (e.g. neurodegenerative diseases). However, there is no reason to believe that continuing on the same path of classical scientific research and expanding human rights and welfare to more people won’t get us there. AI (or more accurately LLMs) can contribute to this path as previous technologies built and improved on their predecessors. It is another tool in the toolkit, not an emergent god. Acknowledging the tremendous advances we currently possess also demonstrates that, despite having much more freedom and information than our ancestors, we still fall for saviors and populists and are still deceived by their faulty logic and glorious prophecies. For better or worse, it seems that human nature doesn’t change much over time, but it’s all we have.  

 

References

  1. Gould, S. Jay. The Mismeasure of Man (Revised & Expanded). (2006).
  2. Lenharo, M. Decades-long bet on consciousness ends - and it’s philosopher 1, neuroscientist 0. Nature 619, 14–15 (2023).
  3. Yudkowsky, Eliezer. & Soares, Nate. If anyone builds it, everyone dies : why superhuman AI would kill us all. 259 (2025).
  4. Is AI Hiding Its Full Power? w/ Geoffrey Hinton (Transcript) – The Singju Post. https://singjupost.com/is-ai-hiding-its-full-power-w-geoffrey-hinton-transcript/.
  5. Popper, Karl. The Poverty of Historicism. 174 (2002).
  6. Demis Hassabis: The 100 Most Influential People in AI 2024. https://time.com/collections/time100-ai-2024/7012767/demis-hassabis/.
  7. Tesla boss Elon Musk says AI will create situation where no job is needed. https://www.cnbc.com/2023/11/02/tesla-boss-elon-musk-says-ai-will-create-situation-where-no-job-is-needed.html.

 

Comments

Post a Comment

Popular posts from this blog

Heresy, Orthodoxy & Time

-
Image
Following up on yesterday's discussions about lecanemab , some points to address. Maybe I am old-fashioned, but I believe that science should be about well-defined entities and mechanisms. I also believe that scientific theories should not be self-contradictory and must make sense within the known laws of physics, chemistry, and biology. From this perspective, I do not think that the two major hypotheses used to describe the amyloid phenomenon (The Amyloid Cascade Hypothesis (ACH) & The Prion Hypothesis (PrH)) fulfill these criteria. I also think that this is the reason behind much of the confusion and the scarcity of clinical success in the field. The ACH & the PrH were unorthodox hypotheses when they were developed in the 80s & 90s. The ACH boldly stated, in contrast to conventional biological and physicochemical wisdom at the time, that the endogenous proteins that form aggregates, even if evolutionary conserved & highly expressed in their native state, are j...
Read more

What is biological information?

-
Image
Biological information is a well-defined concept. It denotes the specific amino acid  sequence   of a protein that corresponds to a specific base  sequence  in DNA. This concept can be represented simply as: Biological information = amino acid sequence = DNA sequence x the genetic code, where 3 specific bases = 1 specific amino acid. This simple definition of biological information is the core principle at the foundation of modern molecular biology since Francis Crick introduced it in his seminal paper “On Protein Synthesis” published in 1958 1 . In a subsection appropriately titled " The essence of the problem ”, Crick lays out the definition of biological information: “ By information I mean the specification of the amino acid sequence of the protein”.  He called this the  Sequence Hypothesis  which states “ that the specificity of a piece of nucleic acid is expressed solely by the sequence of its bases, and that this sequence is a (simple) code for ...
Read more

Amyloid Toxicity: Myths, Contradictions & Failures

-
Image
For over a hundred years, since Alois Alzheimer published his famous plaque and tangle pictures, it has been taken for granted that these plaques and tangles are actively killing the neurons by being toxic. There was no proof needed of why or how they are toxic, the details and mechanisms were to come eventually over time to prove what we already knew all along, these structures are toxic. Observations and experiments were not performed to test the toxic gain -of-function ( GOF) theory of amyloid pathology, but just to verify what we already know, they are toxic. Here, and exactly here, is where the pathology (as a scientific discipline) got its own disease; confirmation bias. No amount of counterevidence was ever enough to challenge GOF. Any counterevidence was either accommodated somehow or simply dismissed, leading to a group of myths and contradictory arguments that are recited almost religiously in papers, reviews, conferences, and media interactions to defend GOF. A smokescreen o...
Read more

Different Colored Eggs in One Basket is Still Risky

-
Image
A n interesting paper was recently published by Dr. John Hardy, the father of the amyloid cascade hypothesis, introducing a hypothesis for explaining late-onset neurodegenerative diseases (NDDs) based on the failure of degradation of amyloidogenic proteins. This adds to the list of other interesting hypotheses aiming to explain the etiology and pathophysiology of NDDs, especially the prevalent sporadic forms. The list includes the inflammation hypothesis, the antimicrobial hypothesis, the innate immune hypothesis, and the autoimmune hypothesis. All are interesting and each has its own merits; however, they all have one common problem. They are all gain-of-toxic-function (GOF) hypotheses . They all assume that either amyloids are toxic or that they trigger other toxic cascades that ultimately kill neurons. At the same ti me, they all ignore the very plausible possibility that these peptides and proteins have original functions that are lost due to their ...
Read more

The "Protein-Only" Hypothesis: The Emperor Without Clothes

-
Image
The protein-only hypothesis has dominated the field of amyloid pathologies for more than two decades now, particularly after the Nobel Prize that was awarded to Dr. Stanley Prusiner in 1997, which acknowledged prions as "a new biological principle of infection". The history of the theory of “protein-only conformational replication” goes something like this:  Brain cells are dying, we went looking for a virus, we couldn't find a virus, instead, we found a protein, only! But, we still assumed it behaves like a virus in two main aspects: 1. It kills the cells →  origin of the gain-of-toxic-function hypothesis (GOF). 2. It self-reproduces without nucleic acid →  origin of the prion/amyloid strain hypothesis. Despite being around for about 40 years (since initial publication in 1982), the theory of protein-only conformational replication still lacks a mechanistic description of its two main viral-like assumptions: 1. Amyloid fibrils are no...
Read more

Why LOF was lost?

-
Image
Why LOF was lost? When a protein aggregates, two things happen, one is certain and the other is uncertain. -  The certain consequence is that the protein loses its function, because any protein needs its native conformation and solubility to function  →  loss-of-function (LOF) - The uncertain consequence is that the resulting aggregates become particularly more toxic  →  gain-of-function (GOF) While the two mirror-image possibilities are scientifically valid, the majority of the field studying amyloids, especially within neurodegenerative diseases, chose to focus on the uncertain consequence (GOF) and almost completely ignore the certain one (LOF). A friend of mine asked me why? Why something as obvious as LOF became almost unthinkable and is only mentioned shyly as a heretic belief on the outskirts of the literature? Here, I will try to discuss the historical factors that I think contributed to this huge asymmetry in amyloid science, and how this uni...
Read more

Genetics (only) won’t save us

-
Image
In November 2020, a news feature in Nature discussed “ A fringe theory [that] links microbes in the brain with the onset of dementia ” (Figure 1). While it was nice to see that this “fringe” theory is finally being considered worthy of discussion by top journals, the article also highlighted the mainstream view, the one which kept this theory on the fringes for a long time. “ But I don’t think it is going to be provable ” the article quotes Dr. John Hardy, the father of the amyloid cascade hypothesis, “ and I don’t think there is much left that needs to be explained about Alzheimer’s beyond the genetics ”, he added. This statement is not only inaccurate, but it also reflects the core framework that has dominated the field of neurodegeneration for the past four decades. This framework emphasizes that nothing more than genetics is needed to understand these diseases, and if we just keep sequencing more people and conducting more genetic studies, the problem will eventually be solved. ...
Read more

Life & death between the double helix & the cross-β sheet

-
Image
The DNA double helix is probably the most famous structure in the world. Since its discovery in 1953; it became an icon for the molecular basis of life, featured in arts and architecture as a symbol of science's triumph and beauty. However, it is less known that another double-stranded structure is involved in devastating diseases such as Alzheimer's, Parkinson's, and mad cow diseases; the cross-β structure . This structure forms the basis of all amyloids and prions.  Both the double helix and the  cross-β  structures are said to have the capacity to replicate and transmit biological information.  But how similar/different are these two double-stranded structures? Why one of them encodes life while the other causes death? The  double-helix/ cross-β  is the relevant comparison when we want to study the question of whether amyloids/prions can self-replicate and what are their functions in biology and disease. We should not be discussing the amyloid supers...
Read more

Cores & Protective Belts in Amyloid Thinking

-
Image
When is it time to give up a scientific theory and embrace a new one?? Three philosophers of science had opinions about this. Karl Popper laid the foundation of the modern view of science that a theory should have testable and falsifiable predictions. For example: if we have a theory claiming that all swans are white, then the observation of one black swan is enough to falsify our white swan theory. This observation should force us to adopt a more general theory that explains the phenomena of swans being black or white.   Another philosopher of science, Thomas Kuhn, objected. He argued that looking at the history of science; this is not how things always worked. Scientists do not abandon a theory at the first negative test. Instead, science progresses in phases. The phase of “normal science”, where negative results are not rejected, but attempts are made to incorporate them within the old theoretical paradigm. Then comes the phase of “scientific revolution”, where a “par...
Read more

Debate Follow-Up: Can Toxic Gain-of-Function (GOF) in Amyloid Pathologies Be Defended?

-
Image
We recently had a very interesting  debate  around the mechanisms of toxicity in neurodegenerative diseases (NDDs) and whether it is mainly a gain-of-function (GOF) or loss-of-function (LOF) toxicity due to amyloid aggregation. Before the debate, the inspiring patient advocate Marina Noordegraaf ( @Sparks4PD ) asked me if the GOF and LOF proponents could switch sides during the debate and try to defend the other point of view. I apologized to her saying that it would be very difficult for me to defend GOF since the most common GOF defenses are counterfactual. Here I will list some of the common GOF defenses that we usually get whenever we discuss LOF in papers, reviews, seminars, or on Twitter.   1.   K nock out/down a nimals  of the amyloidogenic proteins show no phenotype, thus not supporting LOF This is just not true . While there is never going to be a perfect animal model that accurately reflects all aspects of the human pathology (and this is true...
Read more