Masco
This everyday children's toy could be hacked and turned into a listening device.
Hello Barbie
Masco uses Hello Barbie to demonstrate how seemingly harmless household technologies can become surveillance devices.
This term describes AI designed for specific tasks rather than fictional humanlike intelligence.
Narrow AI
Narrow AI performs particular tasks using computational methods. Broussard distinguishes it from general AI, which she discusses as an imagined form of humanlike intelligence.
This philosopher wrote Discipline and Punish and developed the analysis of panopticism.
Michael Foucault
Foucault examined how surveillance and discipline allow institutions to influence people's behavior.
This shared concept connects Foucault's Panopticon with Masco's digital monitoring systems.
Surveilance
Both authors explore how observation becomes a form of power. Foucault focuses on disciplinary institutions, while Masco examines digital surveillance.
True or false: AI systems can identify patterns in data without understanding those patterns the way humans do.
True
AI can use mathematical models to recognize patterns and generate predictions without human like understanding.
This former NSA contractor exposed widespread government surveillance through everyday communications technology.
Edwards Snowden
Snowden revealed how governments could use digital technologies to monitor people's communications, illustrating the enormous scale of modern surveillance.
This term refers to the information used to teach a machine-learning model to recognize patterns.
Training Data
Machine-learning models analyze training data to identify patterns and make predictions. These data can reflect existing social inequalities.
This thinker designed the Panopticon, a circular prison with a central observation tower.
Jeremy Bentham
Bentham proposed the Panopticon as a prison design that would allow guards to observe prisoners without prisoners knowing when they were being watched.
Masco's discussion of unequal facial recognition and Broussard's analysis of machine bias both challenge the assumption that technology is this.
Neutral or unbiased
Both authors demonstrate that technology can reproduce existing social inequalities rather than operate objectively.
This everyday feature uses algorithms to decide which videos, posts, or songs you might enjoy next.
Recommendation system
Recommendation systems analyze user activity to suggest personalized content, such as TikTok videos, YouTube recommendations, or Spotify songs.
Masco uses this social media company to show how a free service can profit from collecting and analyzing user information.
Facebook collects information about users' interests, locations, and interactions. Companies can use this information for targeted advertising, raising concerns about privacy.
In this type of machine learning, a computer identifies patterns without being given explicit labels for the desired outputs.
Unsupervised learning
Unsupervised learning identifies patterns or structures within data without labeled target answers.
In the Panopticon, prisoners change their behavior because they cannot determine this.
Whether they are currently being watched
Prisoners cannot know when surveillance is occurring, so they may behave as though they are constantly being observed.
A student acts differently during an online exam because monitoring software might be recording them. This is an example of which Foucault concept?
Panopticism
The possibility of being watched encourages the student to regulate their own behavior, even without knowing whether someone is actively observing.
True or false: Giving an AI system more data automatically makes its predictions fairer.
False
More data does not necessarily eliminate bias. If training data reflects discrimination, AI may reproduce those same inequalities.
This military surveillance system was designed to use drones to monitor large areas from above.
ARGUS-IS
ARGUS-IS was designed to collect detailed surveillance footage across large geographic areas. Masco discusses how military technology expands the ability to monitor populations.
Broussard uses employment history and this numerical measure in her example of a bank deciding whether to approve a loan.
Credit score
Broussard uses credit scores and employment history to demonstrate how mathematical models can influence decisions about financial opportunities.
Foucault uses this historical epidemic to demonstrate strict spatial separation, inspection, and recordkeeping as forms of disciplinary power.
The plague
Foucault describes how plague-stricken towns used quarantine, surveillance, and detailed records to control populations.
A company uses past customer behavior to predict future purchases. This illustrates Masco's predictive surveillance and Broussard's discussion of this computational technique.
Machine learning
Machine learning uses historical data to identify patterns and make predictions. Both authors examine how these systems affect individuals and society.
This term describes AI-generated video, audio, or imagery that convincingly imitates a real person or event.
Deepfake
Deepfakes can create realistic-looking or realistic-sounding media, raising concerns about misinformation, identity, and consent.
Unlike older surveillance, modern predictive surveillance seeks to anticipate and influence this.
Future human behavior
Modern surveillance uses algorithms and large datasets to predict people's future actions rather than simply recording their past behavior.
Broussard argues that discriminatory outcomes in technology are not merely accidental errors but can reflect this deeper problem.
Structural bias built into technological systems
Algorithms can reproduce existing racism, sexism, and other inequalities through their design, training data, and implementation.
This concept describes how individuals regulate their own behavior because they believe they may be observed.
Internalized surveillance or self-discipline
Foucault argues that people can internalize surveillance and begin monitoring their own behavior without needing constant direct supervision.
A predictive policing system disproportionately targets an already over-policed neighborhood. Which two major concerns from the readings does this combine?
Predictive surveillance and algorithmic bias
Predictive surveillance attempts to anticipate future behavior, while algorithmic bias can reinforce inequalities already present in historical data.
This term describes when a generative AI system produces a confident-sounding answer that is actually false.
AI hallucination
AI systems can generate convincing but inaccurate information, highlighting the importance of checking their outputs.