Why AI systems don’t learn and what to do about it

Dupoux, E., LeCun, Y., & Malik, J. (2026).

ArXiv.org. https://arxiv.org/abs/2603.15381

Introduction

We critically examine the limitations of current AI models in achieving autonomous learning and

propose a learning architecture inspired by human and animal cognition. The proposed framework

integrates learning from observation (System A) and learning from active behavior (System B) while

flexibly switching between these learning modes as a function of internally generated meta-control

signals (System M). We discuss how this could be built by taking inspiration on how organisms adapt

to real-world, dynamic environments across evolutionary and developmental timescales.

The article is here.

Here are some thoughts:

This paper draws heavily on cognitive science and developmental psychology in ways that should resonate with practicing psychologists. The authors lean on foundational developmental psychology, including Piaget, Vygotsky, infant perceptual learning, critical periods, and social learning theory, as the blueprint for next-generation AI. For psychologists, this is a meaningful acknowledgment that decades of careful empirical work on human cognition is not just descriptively interesting but architecturally prescriptive for building intelligent systems.

By cataloguing what current AI cannot do, the paper implicitly maps the distinctive features of human cognition: flexible switching between learning modes, active data selection, embodied grounding, and lifelong adaptation. For clinical or educational psychologists, this reinforces the irreplaceable value of understanding genuine human learning. The ethical sections of the paper are also directly clinically relevant, as the authors raise concerns about anthropomorphization, over-trust in AI agents, and the possibility that AI systems processing somatic-like signals may have uncertain moral status. These are questions psychologists will increasingly face as clients interact with AI systems in therapeutic and educational contexts. Perhaps most importantly, the paper suggests that the gap between AI and human intelligence is not primarily about raw computation but about the architecture of learning itself, which has been psychology's domain all along.

The fallacy of obviousness

Teppo Felin
aeon.co
Originally posted July 5, 2018

Here is an excerpt:

The alternative interpretation says that what people are looking for – rather than what people are merely looking at – determines what is obvious. Obviousness is not self-evident. Or as Sherlock Holmes said: ‘There is nothing more deceptive than an obvious fact.’ This isn’t an argument against facts or for ‘alternative facts’, or anything of the sort. It’s an argument about what qualifies as obvious, why and how. See, obviousness depends on what is deemed to be relevant for a particular question or task at hand. Rather than passively accounting for or recording everything directly in front of us, humans – and other organisms for that matter – instead actively look for things. The implication (contrary to psychophysics) is that mind-to-world processes drive perception rather than world-to-mind processes. The gorilla experiment itself can be reinterpreted to support this view of perception, showing that what we see depends on our expectations and questions – what we are looking for, what question we are trying to answer.

At first glance that might seem like a rather mundane interpretation, particularly when compared with the startling claim that humans are ‘blind to the obvious’. But it’s more radical than it might seem. This interpretation of the gorilla experiment puts humans centre-stage in perception, rather than relegating them to passively recording their surroundings and environments. It says that what we see is not so much a function of what is directly in front of us (Kahneman’s natural assessments), or what one is in camera-like fashion recording or passively looking at, but rather determined by what we have in our minds, for example, by the questions we have in mind. People miss the gorilla not because they are blind, but because they were prompted – in this case, by the scientists themselves – to pay attention to something else. The question – ‘How many basketball passes’ (just like any question: ‘Where are my keys?’) – primes us to see certain aspects of a visual scene, at the expense of any number of other things.

The info is here.

Social observation increases deontological judgments in moral dilemmas

Minwoo Leea, Sunhae Sul, Hackjin Kim
Evolution and Human Behavior
Available online 18 June 2018

Abstract

A concern for positive reputation is one of the core motivations underlying various social behaviors in humans. The present study investigated how experimentally induced reputation concern modulates judgments in moral dilemmas. In a mixed-design experiment, participants were randomly assigned to the observed vs. the control group and responded to a series of trolley-type moral dilemmas either in the presence or absence of observers, respectively. While no significant baseline difference in personality traits and moral decision style were found across two groups of participants, our analyses revealed that social observation promoted deontological judgments especially for moral dilemmas involving direct physical harm (i.e., the personal moral dilemmas), yet with an overall decrease in decision confidence and significant prolongation of reaction time. Moreover, participants in the observed group, but not in the control group, showed the increased sensitivities towards warmth vs. competence traits words in the lexical decision task performed after the moral dilemma task. Our findings suggest that reputation concern, once triggered by the presence of potentially judgmental others, could activate a culturally dominant norm of warmth in various social contexts. This could, in turn, induce a series of goal-directed processes for self-presentation of warmth, leading to increased deontological judgments in moral dilemmas. The results of the present study provide insights into the reputational consequences of moral decisions that merit further exploration.

The article is here.

The problem with p-values

David Colquhoun

aeon.co

Originally published October 11, 2016

Here is an excerpt:

What matters to a scientific observer is how often you’ll be wrong if you claim that an effect is real, rather than being merely random. That’s a question of induction, so it’s hard. In the early 20th century, it became the custom to avoid induction, by changing the question into one that used only deductive reasoning. In the 1920s, the statistician Ronald Fisher did this by advocating tests of statistical significance. These are wholly deductive and so sidestep the philosophical problems of induction.

Tests of statistical significance proceed by calculating the probability of making our observations (or the more extreme ones) if there were no real effect. This isn’t an assertion that there is no real effect, but rather a calculation of what would be expected if there were no real effect. The postulate that there is no real effect is called the null hypothesis, and the probability is called the p-value. Clearly the smaller the p-value, the less plausible the null hypothesis, so the more likely it is that there is, in fact, a real effect. All you have to do is to decide how small the p-value must be before you declare that you’ve made a discovery. But that turns out to be very difficult.

The article is here.

The Liar's 'Tell': Is Paul Ekman stretching the truth?

By Christopher Shea
The Chronicle
Originally published October 10, 2014

Here is an excerpt:

But Ekman’s lie-detection work has recently taken some hard blows. He has long had academic critics (unmentioned in Blink) who say he has not proved that his behavior-based lie-detection techniques actually work. In November 2013, the Government Accountability Office took things up a notch by recommending that Congress cut the funding of the TSA program. The watchdog agency argued that neither scholarship in general nor specific analyses of SPOT offered any proof that malign intent could be divined by looking at body language or facial cues.

Plenty of academics share this negative view of SPOT. "I really don’t think the current program at TSA is doing anything to protect us," says Charles R. Honts, a professor of psychology at Boise State University, who has consulted with the Department of Defense on behavioral observation.

The entire article is here.