DATA 0080 Midterm 2 Study Guide

What does data mean in Latin?

Things given.

What did data originally mean in English?

Accepted facts or premises used as the starting point of a logical or theological argument.

How did the meaning of data change in the eighteenth century?

It shifted from accepted premises to empirical evidence collected through measurement.

Why does the history of the word data matter?

It helps explain why data can seem like undeniable truth, even though people choose how to collect and interpret it.

What are the three metaphors for data?

Data as a resource, data as capital, and data as a verb.

Data as a resource

The idea that data is something we extract from the world, process, and use.

What does data is the new oil suggest?

Data is a valuable resource that can be extracted and processed to make it useful.

How can collecting data be exploitative?

Organizations can take information from people and gain value from it while the people receive little benefit or control.

What does the resource metaphor help us understand?

The extraction, processing, and storage of data, including possible exploitation.

Data as capital

The idea that data is an asset that can produce ongoing value.

How does Google illustrate data as capital?

It uses information about users to generate value, such as through targeted advertising, without necessarily selling the underlying data.

Data imperative

Collecting and keeping as much data as possible because it might become useful or valuable later.

How does the capital metaphor explain the data imperative?

Organizations treat stored data as an asset that could generate future value, even without an immediate use.

How can personal data provide capital for an individual?

A good credit history or useful online reputation can help someone gain better terms or opportunities.

Why might people willingly share personal data?

They may expect social benefits, useful services, or opportunities in return.

Data as a verb

The idea that data is something people do: they turn parts of the world into information by counting, recording, or categorizing them.

What does it mean to data something?

To choose an aspect of the world and turn it into a record that can be analyzed.

Example of data as a verb

Recording a baby's dirty diapers turns everyday activity into data about patterns that may relate to health.

Why is data as a verb a useful correction to the resource metaphor?

It shows that data is created through choices rather than simply found in nature, and that new data can continually be generated.

Abstraction

A simplified representation of reality that includes some details and leaves others out.

Why is data an abstraction rather than reality itself?

A record captures selected features of an event or person, not everything about them.

How do the three data metaphors differ?

Resource emphasizes extraction and processing; capital emphasizes value; verb emphasizes the choices that create data.

Corpus

A large, structured collection of material such as text, images, or videos used for analysis or model training.

Corpora

The plural of corpus.

Why are corpora important to machine learning?

They provide examples that models can use to learn patterns.

Why are text corpora important to large language models?

They supply large collections of language from which the models learn patterns.

How do text corpora help linguistics research?

Researchers count and compare words, sounds, and language patterns to test claims about language use.

How are text corpora collected?

Researchers gather or sample material from chosen sources, genres, time periods, or recorded conversations.

What can text reveal beyond its literal content?

Patterns related to who produced it, whom it was intended for, and the context in which it was produced.

Brown Corpus

A carefully designed collection of about one million words from 500 samples of edited American English published in 1961.

Why was the Brown Corpus important?

Its documented sampling made systematic language research, reuse, and comparison possible.

What kind of language did the Brown Corpus represent?

Edited American English published in a particular year, rather than all language or everyday conversation.

Brown Corpus versus modern text corpora

Brown was smaller and carefully documented. Modern web corpora are much larger but can have less transparent or controlled sampling.

Does a larger corpus automatically represent a population better?

No. More material does not fix missing groups, narrow sources, or a mismatch with the research question.

Three ways corpus research can be biased

Researchers collect what is available instead of what they need; the sample fails to represent the intended population; or claims go beyond what the corpus supports.

What does biased for what claim mean?

A corpus should be evaluated in relation to the specific claim it is being used to support.

Example of a claim a corpus cannot support

Using edited newspaper articles alone to claim how everyone speaks in everyday conversations.

What should you ask when evaluating a corpus?

Who produced the material, what was included or excluded, and whether it represents the population relevant to the claim.

What technological change helped spark modern privacy concerns?

Portable cameras made it easier to photograph people without their active participation.

Right to be let alone

A rights-based view of privacy associated with the influential Warren and Brandeis article.

What helped lead to the Privacy Act of 1974?

Concerns about government databases and abuses of government information, including those associated with Watergate.

Privacy Act of 1974 blind spot

Its central protections focused on federal government records rather than comprehensively regulating private companies.

Why was the Privacy Act's blind spot poorly timed?

Computers, credit cards, databases, and the internet soon made private companies' collection of personal data much easier.

How did technological change support the data imperative?

Lower collection and storage costs encouraged companies to keep information for possible future use.

What privacy developments since 2018 does the guide emphasize?

California's Consumer Privacy Act and Europe's GDPR, which expanded protections concerning private-sector data practices.

Data minimization

Collecting only the personal information needed for a specific purpose.

Data minimization versus the data imperative

Minimization says collect only what is needed; the data imperative says collect broadly in case information becomes useful later.

Is data minimization the same as deletion on request?

No. Minimization limits collection and retention to what is needed; deletion on request is a separate protection concerning existing records.

Data privacy

Who collects personal information, how it is used, and who is allowed to access or share it.

Data security

Protecting data from unauthorized access, loss, alteration, or exposure.

Data privacy versus data security

Privacy concerns appropriate collection and use; security concerns protecting the information.

Can data be secure but still raise privacy concerns?

Yes. A well-protected database can still contain excessive information or be used for purposes people did not agree to.

Regulation

Rules and other mechanisms that shape or constrain behavior.

Why does regulation matter for data ethics?

Ethical principles need mechanisms that put them into practice and hold people or organizations accountable.

Lessig's four modalities of regulation

Legal rules, social norms, markets, and architecture.

Legal modality

Formal rules and penalties that influence behavior, including laws and institutional policies.

Example of legal regulation of data

FERPA rules concerning education records, or a university policy restricting recording.

Why can a university policy count as legal regulation in this course?

The course uses legal broadly to include formal institutional rules, not only government laws.

Norms modality

Social expectations, approval, disapproval, and peer pressure that influence behavior.

Example of norms regulating data sharing

Students protesting or criticizing a university for sharing surveillance information with ICE.

Market modality

Economic incentives, competition, and business interests that shape behavior.

Example of market regulation in the guide

Apple's privacy policies can support its hardware business and distinguish it from competitors that depend on user data.

Architecture modality

The way technology or the physical environment is designed to enable or restrict actions.

Example of architecture regulating data sharing

End-to-end encryption restricts a service provider's ability to read or disclose message contents.

How could all four modalities discourage speeding?

Legal: tickets. Norms: passenger disapproval. Market: insurance costs. Architecture: speed bumps.

Deontological argument for data privacy

People have rights to privacy and autonomy, so data collectors have duties to respect them even when intrusive collection might be useful.

Utilitarian argument about data privacy

Weigh the benefits of data collection against its harms and choose the action with the best overall consequences.

Deontology versus utilitarianism

Deontology asks whether rights and duties are respected; utilitarianism asks whether overall benefits outweigh harms.

Examples of benefits in a utilitarian privacy argument

Useful services such as Gmail and Google Maps, or safety benefits from appropriate data use.

Examples of harms in a utilitarian privacy argument

Personal exposure, stalking, discrimination, manipulation, or misuse of sensitive information.

Personally identifiable information; PII

Information that identifies a person directly or can identify them when combined with other information.

Database join

Combining records from different datasets using shared information or fields.

Why are joins important for privacy?

Combining datasets can reveal identities or sensitive facts that were hidden in each dataset alone.

Anonymization

Removing or changing identifying information to make records harder to link to people.

Re-identification

Linking supposedly anonymous data back to a particular person.

What did Latanya Sweeney's example demonstrate?

Anonymized medical records could be linked to voter records using ZIP code, birth date, and gender to identify the Massachusetts governor's records.

Does removing names guarantee anonymity?

No. Other details, combinations of attributes, or distinctive patterns can identify people.

Why is location data difficult to anonymize?

A person's home, workplace, and repeated travel routes can form a distinctive identifying pattern.

Why are images and likenesses difficult to anonymize?

Appearance and movement can allow people to be recognized or tracked even without names.

Duke MTMC study

A multi-target, multi-camera study that recorded campus footage to improve matching people or objects across camera views.

What was the goal of Duke MTMC?

Re-identification: matching a person or object that leaves one camera view and appears in another.

Is Duke MTMC re-identification the same as facial recognition?

No. Matching someone across camera views can use appearance or movement without identifying a named person from a face.

How was the Duke MTMC dataset used after publication?

It became a machine-learning benchmark and was used in research by companies involved in Uyghur surveillance in China.

What happened to the Duke MTMC dataset?

Duke withdrew it in 2019 amid ethical concerns.

Why did the professor express sympathy for the Duke researchers?

They had IRB approval and posted notices about recording, yet the data later supported uses that raised serious ethical concerns.

Institutional Review Board; IRB

An ethics review body that evaluates research involving human participants.

What does the Duke case teach about research ethics?

Ethical risks include what happens after publication, not only the original research purpose or initial approval.

Contextual integrity

The idea that privacy depends on information flowing appropriately within its context, rather than simply on whether information is secret.

Contextual integrity example

Being seen by people while crossing campus is different from having that movement recorded and made searchable for distant audiences.

Why can publicly visible information still raise privacy concerns?

Collecting, storing, linking, and sharing it can change its audience, purpose, and consequences.

Surveillance

Focused, systematic, routine attention to personal details for purposes such as influence, management, protection, or direction.

Surveillance example

Using connected cameras to track vehicle movements across different locations.

Privacy-security tradeoff

Surveillance can provide safety benefits while reducing privacy and creating risks of misuse or excessive monitoring.

What does security mean in the camera-surveillance debate?

Usually public or personal safety. This differs from data security, which means protecting information.

How would a utilitarian evaluate surveillance?

Compare the evidence of safety benefits with privacy harms, misuse risks, and other consequences for everyone affected.

Flock

A private company that provides automated license-plate-reading cameras and tools for searching vehicle information.

Automated license-plate reader; ALPR

A system that detects and reads vehicle license plates from images.

How can Flock cameras help law enforcement?

They can help locate vehicles and investigate crimes by making vehicle observations searchable.

Why is Flock controversial?

It records movements of people not suspected of crimes, enables tracking across locations, and raises concerns about access, secondary uses, and oversight.

How has ICE accessed Flock data in the course examples?

Through local law-enforcement partners, illustrating that indirect sharing can expand access beyond the original purpose.