Limits of Artificial Intelligence

According to this article, "[t]he current scramble to integrate 'neuro-symbolic' AI is less of a breakthrough and more of a desperate patch job, an attempt to hide the fact that modern machine learning still cannot truly reason".

Since machine learning has its limits, researchers and developers are now experimenting with systems that combine machine learning and rule-based inference. This is what they call "neuro-symbolic AI". But in my opinion, the author of the mentioned article is right that this is only like darning one's socks and not a real solution to the problem.

 

This figure (taken from an old article about Taylor polynoma written by me) illustrates the root cause of the problem that machine learning is increasingly reaching its limits. Polynomial functions can only approximate trigonometric and other complex functions to a certain extent. Outside a specific range, the results make no sense. And that explains why neural networks hallucinate.

Claus D. Volko 

Comments

Post a Comment

Popular posts from this blog

Summary of "The Inappropriately Excluded"

The late Michael Wells Ferguson was the author of the article "The Inappropriately Excluded", which I consider a classic and a must-read. Many highly intelligent people dream of reaching high positions in society, such as becoming university professors or winning the Nobel Prize. Ferguson shows that professors, doctors, judges, etc. are on average not that smart and that people with far above average intelligence usually do not get into these positions. The reason is not that these people have shortcomings but that highly intelligent people are systematically discriminated against. <<In the popular culture, IQ has become a point of contention. Many people credulously accept that the eminent have very high IQs and that people of ordinary accomplishment have ordinary IQs. For example, it was widely reported that Garry Kasparov has an IQ of 190. In truth, his IQ is verified to be 135.>> <<Over an extensive range of studies and with remarkable consisten...
Read more

A note about Mensa

The page about Mensa on German Wikipedia ( https://de.wikipedia.org/wiki/Mensa_in_Deutschland ) contains an interesting passage: The criterion for membership in Mensa is an intelligence quotient that is higher than that of 98% of the population. In the intelligence tests commonly used in Germany (mean 100, standard deviation 15), this corresponds to a score of 130 or higher. Mensa offers its own intelligence test for this purpose, which is evaluated by a psychologist. This test is approved for persons aged fourteen and older. External assessments, e.g. from registered psychologists, can also be accepted; for children under the age of fourteen, this is the only option. Almost half of new members submit external assessments. High IQ scores are often first achieved in tests independent of Mensa, which then draws attention to the topic of giftedness and Mensa. That means that almost half of the Mensa members have probably practiced dozens of intelligence tests until they once reached a sc...
Read more

Beyond IQ: Why We Should Measure the Intelligence of Expression

Abstract Traditional intelligence testing focuses almost exclusively on internal cognitive processing. Yet in real intellectual life—science, philosophy, leadership, and increasingly artificial intelligence—what ultimately matters is not merely having understanding, but expressing it. This article argues that intelligence assessment systematically neglects output quality and proposes a new complementary construct: the Externalization Quotient (EOQ) . 1. The Hidden Assumption Behind IQ Tests Modern intelligence tests are built on an implicit assumption: Intelligence is something that happens inside the mind. Input (perception, memory) and processing (reasoning, abstraction) are carefully operationalized and measured. Output—language, explanation, communication—is treated as a contaminant variable rather than an object of study. This design choice made sense historically. Psychometrics aimed to be: language-minimal culture-fair objectively scorable Expression, by...
Read more