🤖 AI, Biases, and Behavioral Ethics
Algorithmic biases, epistemic bubbles, digital polarization, and the human space in the age of automation. When machines inherit our prejudices and amplify our cognitive flaws.
Modules
Topics
Duration
Level
Learning Path Map
Algorithmic Biases
When AI inherits our biases. COMPAS, facial recognition, and automated hiring.
Epistemic Bubbles
Filter bubbles, digital polarization, and how recommendation algorithms distort reality.
Ethics and Human Space
Bruzzo, erosion of cognitive agency, and what remains human when we delegate everything to the machine.
Detailed content
Algorithmic Biases: When AI Inherits Our Prejudices
ProPublica, MIT Media Lab, Amazon · 6 topics
What is
AI systems trained on historical data that reflect racial, gender, and class discrimination. The algorithm doesn’t “invent” bias: it encodes it mathematically based on patterns present in the training data.
Why learn
Automated decisions affect millions of people. Understanding the source of bias is the first step toward mitigating it.
Key concepts
Data bias, selection bias, algorithmic discrimination, fairness in ML.
What is
A ProPublica investigation (2016) revealed that the COMPAS system, used in the U.S. justice system, classified Black people as high risk for reoffending almost 2x more often than white people, even when controlling for objective criminal factors.
Why learn
The case shows how historical social biases translate directly into automated judicial decisions with real consequences for people’s freedom.
Key concepts
Risk assessment, differential false positives, predictive fairness, algorithmic accountability.
What is
Joy Buolamwini (MIT Media Lab, 2018) showed that facial recognition systems had an error rate of up to 34.7% for Black women, compared with less than 1% for white men. Cause: training datasets dominated by white male faces.
Why learn
Facially biased systems are used in police surveillance, border control, and device unlocking, with a disproportionate impact on minorities.
Key concepts
Gender Shades, representation bias, dataset bias, algorithmic intersectionality.
What is
In 2018, Amazon scrapped an AI résumé screening system that systematically penalized women applying for engineering jobs. The system learned from historical hiring data, where engineers were predominantly men.
Why learn
Shows how ML optimizes metrics based on data. If hiring data favors men, the algorithm learns that men are "better candidates".
Key concepts
Gender bias, feedback loop, blind optimization, proxy discrimination.
What is
Removing protected variables (race, gender) from the data isn’t enough. Other correlated variables (ZIP code, educational history, name) act as proxies and reproduce discrimination through indirect paths.
Why learn
The naive solution (“just remove the race variable”) fails because the entire social structure is correlated.
Key concepts
Proxy variables, digital redlining, structural correlation, fairness through unawareness.
What is
There are multiple mathematical definitions of "fairness" in AI (demographic parity, equality of opportunity, calibration). Chouldechova’s Impossibility Theorem (2017) showed that some of them are mathematically incompatible.
Why learn
There is no neutral technical solution. Every fairness choice involves ethical and political trade-offs that humans need to decide.
Key concepts
Demographic parity, equalized odds, calibration, impossibility theorem, ethical trade-offs.
Epistemic Bubbles and Digital Polarization
Eli Pariser, Facebook documents · 6 topics
What is
Recommendation algorithms maximize engagement. Engagement is maximized by intense emotion: fear, anger, outrage. The content that holds attention the most isn’t the most informative, but the most emotionally provocative.
Why learn
The platforms billions of people use every day are optimized to exploit emotional biases, not to inform.
Key concepts
Engagement metrics, attention economy, outrage optimization, digital dopamine.
What is
Algorithmic personalization creates a unique information environment for each user, progressively filtering out divergent content. The world that appears in your feed isn't the world: it's a projection that confirms your preexisting views.
Why learn
Billions of people believe they are seeing "reality" when in fact they are seeing a version edited algorithmically to maximize engagement.
Key concepts
Filter bubbles, personalization, algorithmic curation, illusion of informational diversity.
What is
Research on YouTube and leaked internal Facebook documents revealed that algorithms progressively steer users toward more extreme content because extremism generates more engagement. Dieting leads to anorexia; a moderate politician leads to extremism.
Why learn
Radicalization is not accidental: it is a predictable consequence of optimizing for engagement.
Key concepts
Rabbit hole, radicalization pipeline, Frances Haugen, Facebook Papers, autoplay.
What is
Three amplified biases: confirmation (the algorithm shows you what you already believe), availability (rare and emotional events dominate the feed), and the bandwagon effect (viral = credible, simply because it is viral).
Why learn
Cognitive biases that have always existed now operate at industrial scale, amplified by algorithms that turn them into permanent infrastructure.
Key concepts
Digital confirmation bias, availability cascade, algorithmic social proof, Bruzzo: bias as infrastructure.
What is
Mitigation proposals: independent algorithmic audits, transparency in recommendation criteria, defaults that promote information diversity, and digital literacy as a required curricular component.
Why learn
A diagnosis without action is sterile. There are viable ways to mitigate the effects of algorithmic polarization.
Key concepts
Algorithm auditing, explainability, informational nudges, media literacy, platform regulation.
What is
While regulation arrives (or doesn’t), individual defense depends on digital literacy: understanding how algorithms work, actively seeking out different sources, and recognizing when emotions are being exploited.
Why learn
This is the direct practical application: knowing that a bubble exists is different from knowing how to get out of it.
Key concepts
Digital critical thinking, information diet, deliberate serendipity, algorithmic metacognition.
Ethics, Technology, and Human Space
Marcus Bruzzo, "Seremos Dados" · 6 topics
What is
Marcus Bruzzo ("We Will Be Data") argues that the central question is not whether AI is superior or inferior to humans. The question is: what happens to human space when we delegate cognitive, creative, and decision-making functions to automated systems?
Why learn
Shifts the debate from "Will AI replace us?" to "What are we choosing to lose?"
Key concepts
Human space, cognitive delegation, automation of creative functions, We Will Be Data.
What is
Outsourcing memory to smartphones, navigation to GPS, and writing to AI. Neuroplasticity means that unexercised abilities atrophy. This isn’t a metaphor: the brain physically loses connections it doesn’t use.
Why learn
Connects directly with the neuroplasticity of Track 3: the same malleability that enables learning also enables atrophy.
Key concepts
Cognitive offloading, cognitive atrophy, the Google effect, GPS and the hippocampus, reverse neuroplasticity.
What is
Billions of people using the same text, image, and code generation systems produce homogenized culture. The opposite of the imaginative surplus that Bruzzo considers essential to being human.
Why learn
Cultural diversity isn't an aesthetic luxury: it's a condition for adaptability, innovation, and civilizational resilience.
Key concepts
Cultural homogenization, algorithmic monoculture, imagined surplus, diversity as resilience.
What is
Critical decisions (credit, health, justice, employment) delegated to systems that no one can explain or challenge. Less transparency, less review, less accountability. Power shifts from auditable institutions to corporate black boxes.
Why learn
Democracy presupposes the possibility of contestation. Opaque systems erode the ability to question decisions that affect people's lives.
Key concepts
Black box, accountability gap, right to explanation, algorithmic opacity, faceless power.
What is
For Bruzzo, what remains human is the useless. Art that doesn’t optimize. Love that doesn’t maximize utility. Play without a productive purpose. Conversations that exist just for the sake of existing. Philosophical resistance: valuing what cannot be reduced to data.
Why learn
This is the philosophical answer that brings the whole course together: from System 1 to the algorithm, what defines us is not efficiency.
Key concepts
Uselessness as value, resistance to algorithmic utilitarianism, art as exception, love as irreducible.
What is
Integration of the entire course: System 1, which makes quick decisions; the biases that distort; social conformity that amplifies; dopamine that hooks you; the economy that manipulates; philosophy that questions; and AI, which inherits all of this on a planetary scale.
Why learn
The integrated view is the ultimate goal: understanding the complete system, not isolated fragments.
Key concepts
Systems thinking, technological-cognitive convergence, the future of human behavior, shared responsibility.