media theory and ai

What are "operative portraits" according to Roland Meyer?

Digital images used for data processing, surveillance, and profiling, rather than static representations of individuality.

Which concept by Harun Farocki inspired Meyer's notion of operative portraits?

Operative images, which are not created to represent but to function within technical processes.

What historical transformation does Meyer associate with portraits?

A shift from representing individuality to becoming tools for linking, classifying, and processing data.

How did Lavater's silhouette machine contribute to physiognomy?

It created standardised facial profiles, reducing individuality to measurable and comparable data.

What was Alphonse Bertillon's key innovation in portraiture?

An anthropometric system combining photos and measurements for criminal identification and database organization.

How did Soviet avant-garde artists like Rodchenko redefine portraiture?

They emphasized collective identity and social interactions over static individual representation.

What is the significance of Andy Warhol's "Screen Tests" in the history of portraits?

They transformed portraiture into a performative act within a system of ranking and competition.

What company exemplifies modern operative portraits according to Meyer?

Clearview AI, which collects and processes facial images for surveillance and identification.

How do operative portraits function in today's digital culture?

They bridge visible images and invisible algorithmic processes, linking human vision with automated systems.

Why does Meyer argue that operative portraits are still considered images?

Because they remain visible and meaningful to human users while also serving algorithmic processes.

What parallels does Meyer draw between Lavater’s work and modern technology?

Both use visual data to classify and interpret identities, linking physical features to broader systems of analysis.

What is the "black hole" in Warhol's "Screen Tests"?

The camera, symbolizing the pressures of performativity and self-presentation.

How does Meyer view the cultural shift illustrated by operative portraits?

A movement from valuing individuality to prioritizing operational utility and data extraction.

What role does human labor play in training algorithms for operative portraits?

Human input is essential for annotating, interpreting, and preparing training data for machine learning.

What does Meyer identify as the hidden aspect of modern operative portraits?

Their integration into algorithmic "black boxes," obscuring how data is processed and used.

How do social media platforms reflect the principles of operative portraits?

By fostering performative individuality and ranking, similar to Warhol’s Factory dynamics.

What is the relationship between visibility and operativity in Meyer’s analysis?

Operative portraits must remain visible to humans to function as interfaces between culture and algorithms. Matteo Pasquinelli & Vladan Joler: What is the Nooscope, and what does it symbolize?

Matteo Pasquinelli & Vladan Joler: How does the Nooscope aim to challenge AI myths?

By secularizing AI, reframing it as a knowledge instrument, and demystifying its portrayal as an alien intelligence.

Matteo Pasquinelli & Vladan Joler: What is the "black box" problem in AI?

It refers to the opaque nature of machine learning systems, where their decision-making processes are not fully interpretable or transparent.

Matteo Pasquinelli & Vladan Joler: What does the term "knowledge extractivism" mean in the context of AI?

It describes the way AI technologies extract value from data, likened to historical forms of colonial resource exploitation.

Matteo Pasquinelli & Vladan Joler: How do bias and distortion occur in machine learning systems? (also neural network zoo image)

Through historical bias (pre-existing societal biases), dataset bias (selection and labeling), and algorithmic bias (amplification of biases by algorithms).

Matteo Pasquinelli & Vladan Joler: What role does the training dataset play in machine learning?

It serves as the foundational input for AI systems, encoding human labor and decisions into structured data, which heavily influences AI outputs.

Matteo Pasquinelli & Vladan Joler: What critique is made about the societal impacts of AI-driven classification?

AI classification perpetuates societal hierarchies, normalizes biases, and enforces disciplinary regimes through automated decision-making.

Matteo Pasquinelli & Vladan Joler: What is "dimensionality reduction," and why is it significant?

It reduces the complexity of high-dimensional data, making calculations more efficient but often introducing bias by oversimplifying data diversity.

Matteo Pasquinelli & Vladan Joler: What are adversarial attacks in the context of AI?

Techniques that exploit weaknesses in AI systems to produce incorrect outputs, revealing vulnerabilities in statistical models.

Matteo Pasquinelli & Vladan Joler: How does AI contribute to "statistical hallucination"?

By creating patterns or correlations based solely on training data, which may not accurately reflect real-world phenomena or causality.

Matteo Pasquinelli & Vladan Joler: What is the role of human labor in the AI assembly line?

Humans are integral at every stage, from dataset creation and labeling to algorithm supervision, exposing the myth of fully autonomous AI.

Matteo Pasquinelli & Vladan Joler: How is AI linked to colonial and capitalist systems?

AI perpetuates epistemic colonialism and cognitive capitalism by exploiting global labor networks and privatizing public data resources.

Matteo Pasquinelli & Vladan Joler: What does the term "automation is a myth" imply?

That AI relies heavily on human labor, contradicting the narrative of complete automation and independence.

Matteo Pasquinelli & Vladan Joler: What is the critique of "creative AI" in the article?

AI-generated art is limited to imitating existing patterns and cannot produce genuinely novel styles, making its creativity statistical rather than innovative.

Matteo Pasquinelli & Vladan Joler: What are Generative Adversarial Networks (GANs)?

Neural networks designed for generative tasks, producing highly realistic outputs by combining classification and generation in a feedback loop.

Matteo Pasquinelli & Vladan Joler: What is the main objective of the Nooscope diagram?

To illustrate the mechanisms and limitations of AI, highlighting both its operational components and inherent biases.

Matteo Pasquinelli & Vladan Joler: How does the Nooscope relate to historical scientific instruments?

It likens AI to instruments like microscopes and telescopes, which enhance human perception but also introduce distortions.

Matteo Pasquinelli & Vladan Joler: What are the three primary components of machine learning systems outlined in the Nooscope?

Training dataset (data input), learning algorithm (processing tool), and statistical model (output representation).

Matteo Pasquinelli & Vladan Joler: Why do the authors describe AI as a "baroque" and "spurious" architecture?

Because AI systems are composed of ad hoc methods, approximations, and adaptations rather than a unified, rational design.

Matteo Pasquinelli & Vladan Joler: What is dataset bias, and how does it occur?

Dataset bias arises during the collection, preparation, and labeling of data, embedding human prejudices and outdated taxonomies into AI systems.

Matteo Pasquinelli & Vladan Joler: How do the authors link AI to "epistemic colonialism"?

By showing how AI systems extract and exploit global knowledge, often reinforcing power imbalances and historical inequalities.

Matteo Pasquinelli & Vladan Joler: What is the role of historical bias in AI systems?

It refers to pre-existing societal biases that are amplified when AI integrates them as "neutral" components of its models.

Matteo Pasquinelli & Vladan Joler: What is the "black box rhetoric" criticized by the authors?

The oversimplified narrative that AI systems are inherently inscrutable and beyond human understanding, which obscures their flaws and biases.

Matteo Pasquinelli & Vladan Joler: How does "knowledge extractivism" relate to cognitive capitalism?

AI monetizes vast amounts of data, turning human activity and knowledge into commodities for profit within a capitalist framework.

Matteo Pasquinelli & Vladan Joler: What is the significance of "pattern recognition" in AI?

It is the foundational mechanism of machine learning, where systems identify statistical patterns in data to classify or predict outcomes.

Matteo Pasquinelli & Vladan Joler: How does "dimensionality reduction" impact AI?

It simplifies data by reducing complexity, which can save resources but risks oversimplifying and erasing important nuances.

Matteo Pasquinelli & Vladan Joler: What critique do the authors make about "statistical models" in AI?

Statistical models are approximations that often compress and distort reality, limiting AI’s ability to handle novel or anomalous data.

Matteo Pasquinelli & Vladan Joler: What is the "curse of dimensionality," and how is it addressed?

It refers to the exponential increase in computational complexity with more data dimensions, addressed by techniques like dimensionality reduction.

Matteo Pasquinelli & Vladan Joler: How do adversarial attacks challenge AI systems?

By exploiting weaknesses in AI models, adversarial attacks manipulate inputs to produce incorrect outputs, exposing vulnerabilities.

Matteo Pasquinelli & Vladan Joler: What are "deep fakes," and how are they created?

Synthetic media created using Generative Adversarial Networks (GANs) to produce realistic but fabricated images or videos.

Matteo Pasquinelli & Vladan Joler: What limitations do the authors identify in machine learning’s creative potential?

AI can only recombine and improvise within the boundaries of its training data, lacking true innovation or creativity.

Matteo Pasquinelli & Vladan Joler: How does AI enforce "data-centric rationality"?

By normalizing the use of data to measure, classify, and control social behaviors, often in ways that perpetuate biases and inequalities.

Matteo Pasquinelli & Vladan Joler: What is "automation bias," and how does it manifest?

The tendency to overtrust automated systems and their outputs, even when they are flawed or biased.

Matteo Pasquinelli & Vladan Joler: How is the labor behind AI systems described in the article?

As "ghost work," where human labor is hidden in tasks like data labeling, training, and supervision to create the illusion of AI autonomy.

Matteo Pasquinelli & Vladan Joler: What is the "regeneration of the old" in machine learning?

AI’s reliance on past patterns and data, which limits its ability to recognize or create genuinely new or disruptive phenomena.

Matteo Pasquinelli & Vladan Joler: What societal implications arise from AI-driven classification systems?

These systems often reinforce existing hierarchies and discrimination, transforming social norms into computational rules.

Matteo Pasquinelli & Vladan Joler: How does "data sovereignty" relate to AI ethics?

It addresses the need for individuals and communities to control how their data is collected, used, and shared by AI systems.

Matteo Pasquinelli & Vladan Joler: What is the "new Machinery Question" posed by the authors?

A call for public education and critical awareness about AI technologies, akin to debates during the industrial revolution about mechanization and labor.

Matteo Pasquinelli & Vladan Joler: How do adversarial examples highlight the gap between human and machine perception?

They reveal how AI misinterprets inputs that are clear to humans, emphasizing the limitations of machine "intelligence."

Matteo Pasquinelli & Vladan Joler: How does the concept of "heteromation" challenge automation narratives?

By showing that AI depends on distributed human labor for tasks that cannot be fully automated, undermining the myth of complete AI autonomy.

Matteo Pasquinelli & Vladan Joler: What role does "training data" play in perpetuating AI biases?

Training data encodes human judgments and societal structures, embedding pre-existing biases into AI models.

Matteo Pasquinelli & Vladan Joler: How do GANs (Generative Adversarial Networks) work?

GANs pit a generator network against a discriminator network to produce and refine realistic outputs, such as images or audio.

Matteo Pasquinelli & Vladan Joler: What is "statistical hallucination" in AI?

When AI creates outputs based on patterns in the data that may not correspond to real-world phenomena, leading to distorted or fabricated results.

Galdeano et al.: What is developmental learning in the context of social robots?

Developmental learning is an unsupervised learning strategy where robots incrementally build their understanding of the world through interaction, inspired by human cognitive development.

Galdeano et al.: What is the main claim regarding developmental learning in social robots?

Developmental learning can enable social robots to autonomously learn suitable behaviors over time and take an active role in building human-robot interactions.

Galdeano et al.: What is the goal of the BEHAVIORS.AI project?

To design and develop developmental learning algorithms that make human-robot interactions more natural, instinctive, and adaptive.

Galdeano et al.: How does developmental learning address the "symbol grounding problem"?

By enabling robots to construct sensorimotor transformation knowledge through interaction rather than relying on predefined representations of reality.

Galdeano et al.: What are the key properties of the developmental learning approach used in this research?

Active perception, intrinsic motivation, hierarchical learning of interactions, and balancing exploration with exploitation.

Galdeano et al.: What is an "intended interaction" in developmental learning?

An interaction that the robot plans to perform, including the action and the expected result.

Galdeano et al.: What is an "enacted interaction"?

An interaction that the robot has actually performed, including the action and the actual result, which may differ from the expected result.

Galdeano et al.: What role does "valence" play in the decision-making process of a robot?

Valence is the internal value a robot assigns to an interaction, reflecting its perceived cost or benefit.

Galdeano et al.: What is "proclivity," and how does it influence a robot's behavior?

Proclivity is a weighted value combining valence and other factors, guiding the robot's decision on which interaction to perform next.

Galdeano et al.: How do composite interactions function in the developmental learning model?

Composite interactions are higher-level representations combining multiple primitive interactions, allowing the robot to build more complex behaviors incrementally.

Galdeano et al.: How does the robot balance exploration and exploitation?

By alternating between choosing actions with high proclivity for reliable results (exploitation) and trying new actions to learn novel interactions (exploration).

Galdeano et al.: What experiment was conducted using the Nao robot?

The Nao robot learned sequences of actions based on user preferences, such as reacting to head touches, to demonstrate developmental learning in a social context.

Galdeano et al.: What challenges were identified in applying developmental learning to physical robots?

Timing mismatches between actions and feedback, environmental noise, and slower execution compared to simulations.

Galdeano et al.: What is the purpose of the visualization tools developed in this research?

To help researchers observe, analyze, and modify the robot's learning process in real-time, facilitating experimentation and debugging.

Galdeano et al.: What features were added in the second version of the visualization tool?

Enhanced displays for complex composite interactions, control over the robot's algorithms, and adjustable execution rates and preferences.

Galdeano et al.: What benefits do multi-robot implementations provide for developmental learning research?

They allow researchers to test algorithms across different robots and environments, demonstrating the generalizability of developmental learning methods.

Galdeano et al.: What future research areas are identified for developmental learning in social robots?

Addressing scalability, noise robustness, memory optimization, emotional intelligence, multimodal interactions, and evaluating user experience impacts.

Galdeano et al.: How does the developmental learning approach contribute to lifelong learning in robots?

By enabling continuous adaptation and incremental learning from interactions, allowing robots to evolve their behaviors over time.

Galdeano et al.: What question does the paper raise about evaluating developmental learning algorithms?

How to assess algorithm performance and user experience improvements when no predefined goal exists for learning outcomes.

Galdeano et al.: What is the central argument of the paper?

Developmental learning, inspired by human cognitive development, is a promising framework for enabling social robots to learn behaviors autonomously, making interactions more natural and adaptable.

Galdeano et al.: Why is developmental learning considered suitable for social robots?

It addresses challenges like life-long learning, unsupervised adaptation without prior knowledge, and incremental learning from limited data.

Galdeano et al.: What quote supports the importance of developmental learning in robotics?

"Developmental learning appears to be an appropriate approach to develop a form of ‘interactional intelligence’ for social robots."

Galdeano et al.: How does the robot initiate the learning process in developmental learning?

The robot begins by experimenting with actions, perceiving their consequences, and using the results to refine its internal models.

Galdeano et al.: What is the purpose of "hierarchical aggregated schemas"?

To enable robots to abstract and combine primitive interactions into higher-level composite behaviors for complex tasks.

Galdeano et al.: What is the relationship between intrinsic motivation and developmental learning?

Intrinsic motivation drives robots to explore and learn behaviors autonomously, even in dynamic environments with limited prior knowledge.

Galdeano et al.: How is "active perception" defined in the paper?

"The agent reacts to its active perception of the environment... by attempting a new experiment based on its internal evaluation of its perception."

Galdeano et al.: What are the benefits of balancing exploration and exploitation in robotic learning?

It improves learning efficiency by allowing the robot to refine known behaviors while discovering new possibilities, enhancing adaptability.

Galdeano et al.: What is a key challenge in scaling developmental learning?

"Memory consumption... is very costly as it records all interactions from the start," which limits scalability.

Galdeano et al.: How do the authors connect timing to social interactions?

Timing is critical in matching the robot’s actions with human feedback, ensuring smoother and more natural interactions.

Galdeano et al.: What is a direct quote on the role of composite interactions?

"Composite interactions can be seen as higher-level representations of the abilities of the agent."

Galdeano et al.: What problem does the visualization tool aim to solve?

It allows researchers to observe and modify the robot's learning process in real-time, speeding up development and testing.

Galdeano et al.: What features of developmental learning make it distinct from traditional AI approaches?

Incremental learning,
intrinsic motivation,
unsupervised adaptation, and
the ability to learn sensorimotor contingencies through interaction.
Increment, intrinsic, unsupervised adaptation and sensorimotor contigencies

Galdeano et al.: How do the authors describe developmental learning's impact on user experience?

"Could we estimate how developmental learning mechanisms implemented in social robot applications affect the user’s experience?"

Galdeano et al.: What quote reflects the potential of developmental AI for autonomy?

"The developmental paradigm is intrinsically defined to provide autonomy, lifelong learning, and adaptability to an agent."

Galdeano et al.: How is emotional intelligence envisioned in the context of developmental learning?

Through shared representations of interactional contexts between humans and robots, potentially enhancing empathy in social interactions.

Galdeano et al.: What multimodality challenge is discussed in the paper?

Combining verbal and non-verbal cues in the robot’s interaction model to improve contextual understanding.

Galdeano et al.: What role does the concept of "symbol grounding" play in this framework?

Developmental learning addresses the symbol grounding problem by enabling robots to build sensorimotor knowledge rather than static representations of reality.

Galdeano et al.: How do the authors evaluate developmental learning systems?

They emphasize the need for tools to measure technical performance and user experience impact, given the absence of predefined learning goals.

Galdeano et al.: What is a direct quote on the scalability of developmental learning?

"Some major questions still have to be answered... scalability issues and robustness to noisy environments."

Galdeano et al.: What conclusion do the authors draw about the role of humans in developmental learning?

Humans act as dynamic environments, providing feedback that shapes the robot’s learned behaviors and interaction patterns.

Galdeano et al.: What are two future goals outlined for this research?

Integrating emotional intelligence into robot interactions and improving memory optimization for scalability.