Computer Science: Algorithm Definitions

Podcast Beta

Play an AI-generated podcast conversation about this lesson

Questions and Answers

What is a core concern about the 'Black Box Effect' in algorithms?

not exactly know how it works

What are the 3 types of Algorithmic bias mentioned in the content?

Algorithmic bias (correct)

Historical bias (correct)

Data set bias (correct)

Correlation is the same as Causation in Machine Learning.

False

Ghost work involves unpaid workers contributing to building, maintaining, and testing ___ systems.

AI Signup and view all the answers

Match the following labor-related terms with their descriptions:

Annotation = Often underpaid and seen as temporary, involving labeling emotions in videos, classifying spam, and correcting AI errors Ghost work = Unpaid workers contributing to building, maintaining, and testing AI systems Amazon labor conditions = Presents a positive image to the public but workers voice concerns about physical strain and scheduling issues Signup and view all the answers

What type of learning involves feeding a model a lot of data, but fine-tuning is needed?

Supervised learning Signup and view all the answers

What is the purpose of ghost work in algorithm training?

Human reinforcement Signup and view all the answers

Algorithmic management aims to replace labor entirely.

False Signup and view all the answers

____ is used by big tech companies to conceal their activities.

Branding Signup and view all the answers

What is the definition of an algorithm in computer science?

A set of instructions for solving a problem or completing a task following a planned sequential order. Signup and view all the answers

Match the following terms with their descriptions:

Surveillance = Constant assessment Standarization = Flexibility Amazon Mechanical Turk = Cheap labor Exploitation = Outsourcing Signup and view all the answers

What is the calculation problem related to?

Algorithm that distributes knowledge Signup and view all the answers

What are the 5 properties of an algorithm?

Definiteness Signup and view all the answers

What happens if machine learning is given biased or lacking data?

Bias amplification Signup and view all the answers

What are the environmental concerns related to artificial intelligence?

All of the above Signup and view all the answers

Machine learning can only solve problems that we already know how to solve.

False Signup and view all the answers

Rule-based learning involves telling the computer a set of rules, then the computer will complete the __________.

solution Signup and view all the answers

Match the following machine learning terms with their descriptions:

Supervised machine learning = Data has to be labeled and expensive Unsupervised machine learning = Model finds patterns without labeled data Machine vision = Related to machine learning, real-world may not fit rules Rule-based learning = Computer completes based on predefined rules Signup and view all the answers

What is representational harm in algorithms?

Stigmatization or stereotyping of certain people or groups. Signup and view all the answers

What is historical bias in algorithms?

All of the above. Signup and view all the answers

Evaluation bias occurs when the benchmark data used for a particular task represents the use population.

False Signup and view all the answers

____ bias arises when there is a mismatch between the problem a model is intended to solve and the way in which it is actually used.

Deployment Signup and view all the answers

Match the following bias types with their descriptions:

Technical Bias = Arises from technical considerations Preexisting Bias = Embodies existing biases independently Emergent Bias = Arises in real-world use context Signup and view all the answers

What type of blurriness is unacceptable?

Fabrication blurriness Signup and view all the answers

What is crucial for curation and documentation of datasets?

Budget for dataset curation Signup and view all the answers

Large language models do basic logic and are trained to understand logical thought.

False Signup and view all the answers

String prediction models predict the likelihood of a token given its ______ context.

preceding Signup and view all the answers

Match the following critical theorists with their viewpoints:

Critical theorist = Argue that educational AI systems aim to present visions of a particular type of learner and prepare them for society Call for a more radical approach to education = Rethinking the societal structures and roles of education for a more just and equitable world Signup and view all the answers

What issue does the critique of algorithmic fairness focus on?

ignoring broader social and historical contexts of injustice Signup and view all the answers

What type of systems model students' socio-emotional states to influence desired behaviors?

Biopolitical systems Signup and view all the answers

Proctoring software recognizes darker skin.

False Signup and view all the answers

Techno-solutionism suggests that social problems can be solved by __________.

technology Signup and view all the answers

Match the following areas of concern related to biopolitical algorithms:

Detecting affective state/behaviors = Designed to measure behavioral norms including ones in edu Surveillance = Proctoring students At-risk prediction = Has impact on one's identity Signup and view all the answers

What is an example of a technology that enhances security through isolation and compartmentalization, discussed in the text?

Qubes Signup and view all the answers

What does Qubes emphasize in relation to continuous innovation and collective strategies to combat pervasive data surveillance?

the need for continuous innovation and collective strategies Signup and view all the answers

Which term best describes computer enthusiasts driven by an inquisitive passion for tech systems and committed to an ethical version of info freedom?

Hackers Signup and view all the answers

Counter-Al technology uses compartmentalization like Qubes to address concerns about _________________.

surveillance Signup and view all the answers

Counterdata collection is when institutions effectively collect and utilize data for analysis purposes.

False Signup and view all the answers

Study Notes

Algorithm Definition

An algorithm is a set of instructions for solving a problem or completing a task in a planned sequential order.
It is a set of instructions telling a computer what to do.

Programs and Software

A program is an algorithm + data structure.
Algorithm alone does not make software; it needs to be computable.

Properties of Algorithm

5 properties of an algorithm: finiteness, definiteness, input, output, and effectiveness.

Machine Learning

The notion that we can program a computer to learn itself.

Foucault's 3 Levels of Power

Strategic games between liberties; Power through the algorithm, Power as a form of political, social, and economic domination.

Datafication

The practices and relations through which data is constituted and made legible.
Data is fundamental to Al, but often overlooked.

Data Positivism

Data's role, but critics argue that it overlooks biases and errors in data-driven systems.

Al Discourse

A prominent discussion in Al discourses is that it will change everything, reduce inequality, enhance productivity, and make things more accessible.

The Age of Al

Promises of technical and economic efficiency, as well as fairness, are an important component in the discourses of "New elites".

Al Concerns

Environmental cost/effects of computing material and labor conditions.
Bias and discrimination.
Epistemological challenges.

Critical Al Studies

Al has become an empty word; it's not artificial intelligence, but rather an ideology.
The fact that we think it's a technology is part of its ideology.

Enchanted Determinism

Data-driven systems are often perceived as both magically unknowable and absolute in their prediction patterns.

Algorithmic Organization

The way an algorithm organizes and orders the world is never neutral.

Demystifying Al

Artificial intelligence equals machine learning algorithms.

Rule-Based Learning

You tell the computer a set of rules, then the computer will complete; however, you need to already have the rules/solution.

Machine Learning

Can solve things that we don't know how to do.

Machine Vision

Very closely related to machine learning/pattern recognition; one problem is that the real world may not fit your rules.

Supervised Machine Learning

The first age of machine learning; the data has to be labeled, which is expensive.

Unsupervised Machine Learning

Just give data to the model and let it find the pattern; requires too much data.

Al Shift

From automation of logic reasoning to automation of pattern recognition.

What Algorithms Can Do?

They can classify and predict, but they also create.

Algorithmic Power

Epistemic challenges; the way algorithms produce knowledge implies concerns for agency, responsibility, and accountability.

Black Box Effect

Dealing with algorithms raises questions of responsibility and accountability.
A core concern about this tech is that it's very good at recognizing patterns, but we don't exactly know how it works.

Algorithmic Bias

These technologies are not neutral; they are biased over certain groups of people and knowledge.
3 types of algorithmic bias: Historical bias, Data set bias, and Algorithmic bias.### Correlation and Causality
Our knowledge and thinking are based on causality, but we are starting to think that causality is not the only way to understand the world, especially with machine learning and prediction without causality.

Worker Conditions and Labor Automation

Worker conditions are not good, with many experiencing stress and physical strain to keep up with picking rate quotas (e.g., Amazon).
Workers are considered less valued or trusted compared to robotic systems, and are often treated like robots.
Human-Al collaboration often lacks fair negotiation, and is akin to 19th-century labor dynamics.

Ghost Work and Exploitation

Ghost work involves unpaid workers contributing to building, maintaining, and testing AI systems.
AI functionability relies on workers performing repetitive digital tasks, but they receive little credit and fair compensation.
Content moderators perform crucial tasks, but are poorly paid and can suffer psychological trauma.

Facade of Automation and Exploitation

Some systems present a facade of automation while relying heavily on human labor behind the scenes.
Workers face exploitation due to interchangeable roles and low compensation.
AI development relies on the exploitation of human labor across the supply chain.

Amazon Labor and AI Development

Amazon presents a positive image to the public with incentives and slogans, but workers voice concerns about physical strain and scheduling issues.
AI and algorithms intensify worker exploitation across various sectors.

Annotator Labor and Exploitation

Annotation is critical but often seen as temporary or unimportant by engineers focused on AI model building.
Annotators perform tasks like labeling emotions in videos, classifying spam, and correcting AI errors, often underpaid.

Leisure Society and Mass Unemployment

AI is automating human labor at a speed that's never been done before, leading to concerns about mass unemployment.
There are two views: a utopian view, where AI brings increased leisure, quality of life, safety, and security; and a dystopian view, where AI leads to economic inequality, loss of purpose, surveillance, and control.

Hidden Labor and Asymmetric Distribution

There are three types of hidden labor: data annotation, human reinforcement, and content moderation.
The benefits and disadvantages of technology are not equally distributed.

Algorithmic Management and Surveillance

Algorithmic management is used to manage labor, not replace it.
Surveillance, standardization, and flexibility are used to control labor.
Researchers buy labor without knowing the conditions or pay, leading to exploitation of cheap labor in other countries.

Reconfiguration of Societal Structures

AI transforms societal structures, power dynamics, labor, governance, ethics, and cultural practices.
There is a need to reshape the economy in response to profound changes by AI and automation.

Bias in Machine Learning

Machine learning data can be biased, but it's not only the data's fault.
The machine learning pipeline involves a series of choices, and bias can arise at any stage.
Bias can be allocative, representational, historical, or deployment-based.

Fairness and Bias

Fairness algorithms modify the modeling to satisfy particular notions of fairness.
Data sets are used to train computers, but bias can arise from technical constraints or pre-existing biases.
Emergent bias arises in a context of use with real users.
To remedy bias, we need to be able to identify or diagnose bias in a system and develop methods to avoid and correct it.### Data Sets and Bias
Data sets consist of a collection of images labeled and sorted into categories, but lack of representation can lead to underperformance due to bias.
Algorithmic bias can occur due to interconnected aggregation of biases, and researchers' decisions on model development.
Bias in data labeling can promote harmful stereotypes and representations.
Bias cannot be simply fixed due to lack of social context, imperfect processes, and no clear definition of fairness.

Model Cards and Fairness

Model cards provide essential information about an AI model, including its development, intended use cases, and details about its performance.
Fairness is about justice and technical solutions, and investigation of categorization and classification of people.
Language models have biases towards languages that are already well-represented, leaving many languages underrepresented.

Fighting Bias in AI

Researcher Joy Buoamwini fought against algorithmic bias, highlighting how facial recognition algorithms failed to recognize Black women's faces.
Hallucinations in AI models occur due to the gap between statistical production of language and the real world.

Environmental Effects of Large Language Models

Large language models require substantial energy and result in high CO2 emissions and financial costs.
These effects disproportionately affect marginalized communities.

Mitigation Strategies

Prioritize environmental impact assessments in model research.
Budget research for dataset curation and documentation to minimize bias.
Intersectional bias in LM's can encode more bias against marginalized identities.

Curation and Documentation

Crucial for curation and documentation are datasets, and minimizing bias.
Lovelace effect: users perceiving computing systems as original and creative, but focusing on human reactions, not machine capabilities.

Foundation Models

Foundation models are a type of machine/deep learning model that starts with a large dataset and finds patterns.
They are pre-trained and require a whole ecosystem, and need tons of data.

Dangers of Large Language Models

Environmental cost: data centers need huge computer power and result in carbon emissions.
Unfathomable training data: size of datasets and refusal of company transparency make it difficult to critically assess.

Language Models and Understanding

Attribution of models: trained on the most probable way to repeat, but lack strong semantic understanding.
Logical thought: models do not understand logical thought, only trained on statistical prediction.

Pandemic and Education

Shift to remote online learning during the pandemic worsened existing systematic inequalities in education.
Al became prominent, raising concerns about bias and fairness.

Critical Theorist

Argue that education AI systems aim to present visions of a particular type of learner, but reproduce inequitable societal structures.
Call for a more radical approach to rethink what is possible and remake the world in a more just and equitable way.

Surveillance in Schools

Surveillance tech in classrooms is widely accepted, but students rarely know and consent to it.
Monitoring emails and online use at school gives false positive rates for LGBTQ-related terms, claiming it's hate speech.

Biopolitical AI Systems

Model students' socio-emotional states to influence desired behaviors.
Algorithmic controversies in the UK: using algorithms to predict A-levels and giving lower grades to kids in lower-class schools.

Early Dreams of Machine Teaching

Pressy's MC teaching machine (1920s): reward structure.
B F Skimmer's adaptive teaching machine (1950s): emphasizes the closeness of the student and machine.

Studying That Suits You

Use AI to generate personalized quizzes and flashcards to suit your learning preferences.

Description

Learn about the definition of an algorithm in computer science, including its role in solving problems and completing tasks.