Start free trial
Trending New Genres Authors
EnglishEnglish
EspañolSpanish
简体中文Chinese
繁體中文Chinese (Traditional)
FrançaisFrench
DeutschGerman
日本語Japanese
PortuguêsPortuguese
ItalianoItalian
한국어Korean
РусскийRussian
NederlandsDutch
العربيةArabic
PolskiPolish
हिन्दीHindi
Tiếng ViệtVietnamese
SvenskaSwedish
ΕλληνικάGreek
TürkçeTurkish
ไทยThai
ČeštinaCzech
RomânăRomanian
MagyarHungarian
УкраїнськаUkrainian
IndonesiaIndonesian
DanskDanish
SuomiFinnish
БългарскиBulgarian
עבריתHebrew
NorskNorwegian
HrvatskiCroatian
CatalàCatalan
SlovenčinaSlovak
LietuviųLithuanian
SlovenščinaSlovenian
СрпскиSerbian
EestiEstonian
LatviešuLatvian
فارسیPersian
മലയാളംMalayalam
தமிழ்Tamil
اردوUrdu
Searching...
SoBrief
Unmasking AI

Unmasking AI

My Mission to Protect What Is Human in a World of Machines
A face-tracking failure sends a researcher from MIT into the fight over AI's human costs.
by Joy Buolamwini 2023 317 pages
4.10
1k+ ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Treat AI as a mirror of its makers, not a neutral judge.

A mask revealed the system. At MIT, Joy Buolamwini built a playful mirror that used face-tracking software to follow her reflection. The program detected a face drawn on her palm and detected a white mask, but failed to detect her dark-skinned face. The failure was not merely a glitch: it exposed whose faces the technology had been built to recognize.

Bias can be designed in. Buolamwini calls this influence the coded gaze: the priorities and prejudices of technology makers carried into systems, sometimes without conscious intent. AI now helps decide who gets hired, housed, monitored, or medical care, so a technical blind spot can become a real barrier.

Analysis

What's striking is how a small, repeatable demonstration turns an abstract claim into something people can inspect. This resembles a basic principle in experimental science: make the conditions visible so others can test the result. The mask test does not prove every vision system fails in the same way, and a single example cannot identify the precise cause. Its force is different. It challenges the assumption that a product working for one user works for everyone, and makes the burden of testing harder to evade.

Audit the benchmark before trusting the accuracy score.

A gold standard can be lopsided. AI researchers judge systems against benchmark datasets, but familiar benchmarks can quietly favor the people most represented in them. A study of the Labeled Faces in the Wild dataset found 77.5 percent male-labeled faces and 83.5 percent faces labeled white. Even a government dataset designed for geographic diversity, IJB-A, was 75.4 percent male and 79.6 percent lighter-skinned.

The omissions matter. In IJB-A, lighter-skinned men made up 59.4 percent of the data, while women of color made up 4.4 percent. A system could fail on every woman of color in that benchmark and still score 95.6 percent overall. Buolamwini calls the inherited social exclusions in data power shadows.

Analysis

A benchmark is not just a measuring stick. It is a claim about which cases count as representative. This connects to measurement theory: a score is only meaningful for the population and conditions it actually samples. The example also shows why aggregate accuracy can conceal unequal error. Still, greater demographic representation alone cannot guarantee fairness. The labels, collection process, intended use, and consequences of mistakes all shape what a benchmark can legitimately establish.

Treat labels as human decisions, not truths machines discover.

Ground truth has authors. In machine learning, ground truth means the labels treated as correct examples during training or testing. But people choose the categories and assign the labels. Buolamwini found that labeling perceived gender from photographs relied on appearance, clothing, names, and pronouns, while the usual male-or-female options erased people outside that binary. A model trained on those labels can reproduce the assumptions embedded in them.

Classification has a history. Race categories differ across countries and eras, and skin-color scales also carry assumptions. Buolamwini used the Fitzpatrick skin phototype scale, developed to describe skin response to sunlight, while acknowledging its limitations and lighter-skin bias. The categories helped expose differences, but were never a neutral map of humanity.

Analysis

This is a useful distinction between observing a person and assigning a category to them. Measurement can clarify patterns, yet the act of measurement can also harden a disputed category into an apparent fact. That tension appears throughout public statistics, medicine, and law. A classification may be necessary to document unequal treatment, but its limits should be stated, and people should not be forced to fit a category merely because a model needs one.

Measure performance where identities intersect, not one category at a time.

Averages hide the sharpest failures. In Gender Shades, Buolamwini and Timnit Gebru tested gender-classification systems from IBM, Microsoft, and Face++ on 1,270 images of parliamentarians. Overall accuracy looked respectable: 87.9 percent for IBM, 90 percent for Face++, and 93.7 percent for Microsoft. Yet every company performed worst on darker-skinned women.

The gaps were substantial. Microsoft reached 100 percent accuracy for lighter-skinned men but 79.2 percent for darker-skinned women. Face++ scored 65.5 percent on darker-skinned women, and IBM scored 65.3 percent. IBM's gap between lighter-skinned men and darker-skinned women reached 34.4 percentage points. Looking at gender or skin tone separately would have concealed that intersectional pattern.

Analysis

The study's central methodological contribution is not simply that some groups receive worse results. It shows that the choice of comparison can determine whether a disparity is visible at all. This echoes intersectional analysis in law and social science, where overlapping forms of disadvantage can disappear inside broad group averages. The findings concern specific products and a specific classification task. They do not by themselves prove that every facial-recognition task has identical error patterns, which is why auditing each system and use matters.

Use counter-demonstrations to make technical failures impossible to ignore.

Show the failure, not just the score. Buolamwini used a counter-demo, a demonstration that challenges a technology's promotional story by making its limitations visible. Her spoken-word film AI, Ain't I a Woman? paired test results with images of admired Black women. Systems variously misgendered Serena Williams, failed to detect Oprah Winfrey, and labeled an image of Sojourner Truth as a gentleman.

Art can carry evidence. Buolamwini calls an evocative audit a way of showing technical failure so people can witness its human meaning. The approach has precedents in Sojourner Truth's and Frederick Douglass's use of dignified photographs against racist portrayals. Public exposure of ImageNet Roulette's degrading labels also helped prompt removal of hundreds of thousands of images.

Analysis

Numbers are good at establishing patterns; images and stories can make the stakes legible to people who would never read a research paper. Combining the two is a form of translation, not a retreat from rigor. The caution is that people shown in a demonstration can become props in someone else's argument, especially when their images were not collected with consent. Responsible counter-demonstrations therefore need to explain both the system's failure and the ethical choices involved in displaying it.

Ask whether a system should be used before making it more accurate.

Accuracy is not the whole harm. Facial recognition can misidentify someone, but a flawless system could still enable pervasive surveillance. In Brooklyn, tenants at a large, mostly low-income building challenged a landlord's plan to replace keyless entry with facial recognition. The vendor claimed its system was free of demographic bias, yet had published no independent validation for the tenants' circumstances. Buolamwini helped tenants question accuracy, consent, security, and police access.

Context changes performance. A system tested in corporate offices may not work reliably for elderly residents, children, or people in different conditions. Buolamwini calls this context collapse: assuming a system proven in one setting will work in another. The option to refuse deployment, rather than patch the tool, must remain available.

Analysis

This reframes the debate from whether a technology can work to whether its use is justified, for whom, and under what protections. That is especially important when people cannot meaningfully opt out, as with housing or essential services. The same caution applies to well-intended projects: Buolamwini's mobile survey work in Ethiopia ran into missing internet access and Amharic keyboard support. Local participation is not a courtesy; it is a way to discover requirements that distant designers may not know.

An upload can travel far. Photos shared for family albums, social media, or creative portfolios have been repurposed to build facial-recognition and image-generation systems. Users of Lensa's avatar feature received sexualized images they had not intended to create, while artists found their work inside training datasets without their explicit agreement. AI systems therefore depend on a chain of data, companies, and human labor, not just a single app.

Deletion must reach the model. A faceprint is a digital representation derived from facial images. Deleting the original photo may leave both that representation and the trained model intact. Buolamwini calls for deep data deletion: removing the source data, derived data, and models trained on data obtained without consent. She also highlights workers who process content for AI, often for very low pay.

Analysis

The lifecycle view is stronger than a narrow privacy notice at the point of upload. It asks what information persists, who can reuse it, and whether a person can meaningfully withdraw permission. In practice, removing the influence of one training example from a complex model can be technically difficult to verify. That makes prevention and auditable data records especially important. Individual opt-out tools can help, but they should not replace duties on the organizations that collect and commercialize data.

Turn audits into coalitions that can make powerful institutions answer.

Evidence needs an audience. After Gender Shades, IBM engaged with the findings and improved a later model, while Amazon publicly attacked a follow-up audit of its Rekognition system. Buolamwini and her collaborators received support from civil-rights organizations and 75 researchers. In 2019, a federal study by NIST found that some facial-recognition algorithms produced false-positive rates 10 to 100 times higher for Asian and African American faces than for white faces.

Organizing can change deployment. Research informed tenant resistance, public campaigns, congressional testimony, and local restrictions on police use of facial recognition. The lesson is not that an audit automatically changes policy. Researchers, affected communities, journalists, advocates, and lawmakers each have different leverage, and coordinated pressure can make a technical finding consequential.

Analysis

Independent evaluation matters most when it is paired with routes to remedy. A company can publish a better model and still deploy it in a harmful setting; a law can prohibit a use but lack enforcement. The strongest accountability structure therefore combines external testing, transparency about limitations, public oversight, and meaningful redress for people harmed. The book also makes the professional risk visible: criticism of a powerful vendor can threaten careers, so institutions need to protect independent research rather than merely praise it when convenient.

Make voice, choice, and redress the test of algorithmic justice.

People must shape the rules. Buolamwini defines algorithmic justice as people having a say in decisions that affect them, receiving remedies when systems cause harm, and refusing standards of fairness that ignore social history. The 2022 White House Blueprint for an AI Bill of Rights and the EU AI Act debate offered steps toward safety, privacy, and protection against discrimination, though guidance alone does not guarantee enforcement.

Justice also requires sustainable work. After exhaustion nearly led her to leave her PhD, Buolamwini paused, then returned and completed it. Her story resists the idea that service requires self-destruction. AI governance must also address immediate harms, such as wrongful arrests and denied services, rather than letting speculative future risks eclipse people already affected.

Analysis

The book's strongest claim is that AI is a political and institutional choice as much as an engineering achievement. Its proposals gain force when they include participation, accountability, and the ability to stop a system, not only technical refinements. There is a productive tension between preparing for future dangers and addressing documented harms now. Those priorities need not compete, but budgets and public attention are finite, so governance should explain how both kinds of risk are being weighed.

Analysis

Unmasking AI works as memoir, technical primer, research history, and organizing manual. Its narrative engine is a reversal: a young engineer enters technology hoping to escape social conflict, then discovers that the machines she builds inherit choices about whose faces, data, and experiences count. The white-mask experiment makes that reversal memorable, while the progression from classroom prototype to benchmark research, corporate confrontation, tenant organizing, and policymaking shows how technical critique can become public action.

The book's lasting intellectual contribution is its insistence that performance depends on more than model architecture. Defaults, training data, benchmark design, labels, deployment conditions, and institutional power all shape outcomes. Gender Shades makes the point empirically: respectable aggregate accuracy coexisted with severe disparities for darker-skinned women. The book also refuses a simplistic repair narrative. More inclusive data may reduce exclusion from useful systems, but it may also strengthen surveillance. The responsible question is not merely how to improve a tool, but whether it should exist in a particular setting.

The argument is strongest when it distinguishes specific task failures from broad claims about AI, and when it connects technical evidence to people who bear the consequences. Its own research choices reveal a genuine tension: measuring disparities can depend on contested labels and images collected without individual consent. That tension does not invalidate the audit. It shows why transparency about methods and limits is part of trustworthy evidence.

The book's political vision is plural rather than technocratic. Researchers can test systems, but affected communities contribute knowledge about context; artists make harms legible; advocates and lawmakers can restrict uses or require remedies. The result is not a promise that AI can be made neutral. It is a case for keeping decisions contestable, protecting people from imposed systems, and treating refusal as a legitimate outcome of responsible design.

Summary in 30 Seconds
AI systems inherit the biases of their data, labels, benchmarks, and deployment settings. Aggregate accuracy can conceal severe disparities: one facial analysis study scored 93.7 percent overall yet only 79.2 percent for darker-skinned women. Better accuracy cannot resolve harms from surveillance, weak consent, or misuse. Responsible governance requires independent audits, affected communities' participation, meaningful refusal, and remedies for people harmed.
Contains spoilers

Last updated:

Report Issue

Review Summary

4.10 out of 5
Average of 1k+ ratings from Goodreads and Amazon.

Unmasking AI is praised for its accessible exploration of AI bias and ethics. Readers appreciate Buolamwini's personal journey and insights into the tech world. Many find the book informative and thought-provoking, highlighting the importance of addressing algorithmic bias. Some reviewers note the memoir-like style and wish for more technical depth. Overall, the book is seen as a vital contribution to understanding AI's societal impact, though opinions vary on its balance of personal narrative and technical content.

Your rating:
4.56
853 ratings
Want to read the full book?

Glossary

Coded gaze

Bias embedded in technology

The coded gaze describes how the priorities, preferences, and prejudices of people with power to shape technology can be built into technical systems. The effects do not require deliberate discrimination. The term directs attention to who designs a system, what assumptions guide it, and whose lives or characteristics the technology fails to represent.

Power shadows

Social hierarchy reflected in data

Power shadows are traces of existing social exclusion and unequal authority that appear in datasets. For example, data gathered from public officials may overrepresent men because men hold more political office, while media visibility and colonial histories can skew representation by skin tone. These patterns can then be carried into systems trained or evaluated on the data.

Pilot Parliaments Benchmark

Parliamentary face-image benchmark

The Pilot Parliaments Benchmark is a dataset Buolamwini assembled from images of 1,270 parliamentarians across selected countries. It was designed to test commercial gender-classification systems across gender and skin-type groups. Its results exposed disparities hidden by aggregate accuracy, while its construction also raised questions about consent, labeling, and the limits of the categories used.

Gender Shades

Intersectional audit of classifiers

Gender Shades is Buolamwini and Timnit Gebru's study of commercial systems that classified gender from face images. Using the Pilot Parliaments Benchmark, the researchers compared performance across gender and skin-type groups, including their intersections. The study found the worst performance on darker-skinned women, demonstrating how overall accuracy can conceal unequal error rates.

Evocative audits

Audits that make harms visible

Evocative audits use demonstrations and specific experiences to help an audience witness how an AI system can fail and understand the human significance of that failure. They complement quantitative audits, which can identify broader patterns but may not communicate lived impact. Their cultural references may not resonate equally with every audience, and they must avoid exposing people without consent.

Counter-demo

Demonstration challenging tech claims

A counter-demo shows a technology failing in ways that challenge promotional claims or assumptions of neutrality. Buolamwini's white-mask demonstration, for instance, contrasted a system's detection of a mask with its failure to detect her face. Counter-demos make technical limitations visible, but are most persuasive when paired with careful explanation of what the example does and does not establish.

Excoded

People harmed by algorithmic systems

The excoded are individuals or communities harmed by algorithmic systems, including people denied services, misidentified as suspects, or subjected to surveillance. The term emphasizes that harms are experienced by people, not merely recorded as technical errors. It also names a constituency whose knowledge and participation should shape decisions about AI.

Deep data deletion

Removal through trained models

Deep data deletion means removing not only uploaded information, but also derived data and AI models trained using information collected without consent. Deleting an image alone may leave a faceprint or a model that still reflects the image's contribution. The concept calls for data rules that account for downstream uses and retained model effects.

Algorithmic justice

Voice, remedy, and fair power

Algorithmic justice is Buolamwini's vision for governing systems that make or shape decisions about people's lives. It includes a meaningful voice and choice in how those systems are used, accountability and redress when they cause harm, fairness attentive to social history, and a technology sector that reflects the societies it affects. It permits refusing or stopping harmful systems.

AI functionality fallacy

Assuming a system works as intended

The AI functionality fallacy is the mistaken assumption that a system performs its advertised or designed task as expected. A product can be marketed as detecting faces, classifying attributes, or answering questions while failing in real-world conditions or producing unsupported outputs. Independent testing is necessary because a stated function is not proof of reliable performance.

FAQ

What's Unmasking AI about?

  • Exploration of AI and Society: Unmasking AI by Joy Buolamwini examines the intersection of artificial intelligence and social justice, focusing on how AI can perpetuate biases and discrimination.
  • Personal Narrative: The book combines Buolamwini's personal experiences with rigorous research, detailing her journey as a researcher and activist in the field of algorithmic justice.
  • Coded Gaze Concept: A central theme is the "coded gaze," which refers to how the biases of technology creators can be embedded in AI systems, leading to harmful outcomes.
  • Call to Action: Buolamwini emphasizes the need for collective action to ensure AI serves humanity equitably, advocating for algorithmic justice and diverse representation in tech development.

Why should I read Unmasking AI?

  • Timely and Relevant Topic: As AI technologies become more integrated into daily life, understanding their implications is crucial. Buolamwini's insights provide a necessary critique of these systems.
  • Personal and Engaging Narrative: The book combines storytelling with research, making complex topics accessible. Buolamwini's journey offers inspiration and motivation for readers.
  • Empowerment Through Knowledge: Reading this book equips individuals with the knowledge to question and challenge the status quo in technology, encouraging critical thinking about societal structures.
  • Insightful Perspective: It provides a unique viewpoint from a leading voice in algorithmic justice, essential for anyone interested in technology's impact on society.

What are the key takeaways of Unmasking AI?

  • AI Reflects Human Bias: AI systems are not neutral; they reflect the biases of their creators, leading to real-world consequences for marginalized groups.
  • Importance of Diverse Representation: Buolamwini argues for the necessity of diverse voices in technology development to ensure AI serves all communities fairly.
  • Call to Action: The book urges readers to engage in conversations about AI and its societal impacts, advocating for algorithmic justice and shaping the future of technology.
  • Collective Action for Change: Buolamwini emphasizes the need for a united front against harmful AI practices, encouraging participation in movements for algorithmic justice.

What are the best quotes from Unmasking AI and what do they mean?

  • "The coded gaze describes the ways...": This quote highlights how biases can be embedded in AI systems, leading to discrimination against marginalized groups.
  • "We cannot have racial justice...": It underscores the urgent need for ethical considerations in AI development, particularly in sensitive areas like law enforcement.
  • "AI will not solve poverty...": Buolamwini emphasizes that technological solutions cannot address systemic societal issues without addressing underlying social and economic structures.
  • "Your silence will not protect you.": This quote underscores the importance of speaking out against injustices, particularly in the context of AI and technology.

How does Joy Buolamwini define the "coded gaze" in Unmasking AI?

  • Definition of Coded Gaze: The "coded gaze" is how technology reflects the biases of its creators, leading to harmful outcomes for marginalized communities.
  • Examples of Coded Gaze: Buolamwini shares personal anecdotes, such as her experience with facial recognition technology that failed to recognize her dark-skinned face.
  • Broader Implications: The concept extends beyond race and gender, encompassing various forms of discrimination, including ableism and ageism.
  • Call for Critical Examination: Buolamwini calls for a critical examination of technology to ensure it serves all communities equitably.

What is algorithmic bias, as defined in Unmasking AI?

  • Definition of Algorithmic Bias: It refers to the systematic and unfair discrimination that occurs when AI systems produce different outcomes for different demographic groups.
  • Real-World Implications: Buolamwini's research shows how algorithmic bias can lead to misidentification and wrongful arrests, particularly for people of color.
  • Need for Accountability: Addressing algorithmic bias requires transparency and accountability from tech companies, with rigorous testing and auditing of AI systems.
  • Impact on Society: The book provides examples of how these biases manifest in technologies like facial recognition, affecting marginalized communities disproportionately.

What are the implications of AI technologies on marginalized communities as discussed in Unmasking AI?

  • Discrimination in AI Systems: AI technologies, particularly facial recognition, can lead to discrimination against marginalized groups, often misidentifying or not recognizing individuals.
  • Real-World Consequences: Instances of AI failures have resulted in wrongful arrests and other serious repercussions for individuals from marginalized communities.
  • Need for Ethical AI: Buolamwini advocates for ethical AI systems that prioritize the needs and rights of all individuals, emphasizing inclusive data and diverse representation.
  • Urgent Accountability: The book underscores the urgent need for accountability in AI development to prevent discrimination and harm.

How does Unmasking AI address the issue of data collection and representation?

  • Critique of Data Practices: Buolamwini critiques data collection practices that prioritize convenience over inclusivity, often lacking representation of diverse populations.
  • Importance of Diverse Datasets: The book emphasizes the need for datasets that accurately reflect societal diversity to prevent biased outcomes in AI systems.
  • Call for Transparency: Buolamwini urges tech companies to disclose data sourcing and demographics, essential for building trust and accountability in AI technologies.
  • Impact on AI Development: Without diverse representation in data, AI systems will continue to perpetuate existing inequalities, highlighting the need for change.

How does Joy Buolamwini's work contribute to the fight for algorithmic justice?

  • Founding the Algorithmic Justice League: Buolamwini established the organization to raise awareness about AI bias and advocate for equitable technology.
  • Research and Activism: Her work combines academic research with grassroots activism, making her a prominent figure in the movement for algorithmic justice.
  • Influencing Policy: Buolamwini's advocacy has led to significant discussions around AI regulation, including calls for moratoriums on facial recognition use by law enforcement.
  • Community Engagement: She emphasizes the importance of community engagement in addressing AI harms, ensuring that affected voices are heard.

What is the significance of the "Gender Shades" project mentioned in Unmasking AI?

  • Research on Facial Recognition: The "Gender Shades" project examined facial recognition systems' performance across demographic groups, revealing significant accuracy disparities.
  • Impact on Industry Standards: The findings influenced major tech companies to reconsider practices and improve algorithms, prompting discussions about AI ethics.
  • Raising Awareness: The project catalyzed broader conversations about representation and bias in AI, highlighting the need for diverse datasets and inclusive practices.
  • Influence on Policy: It has served as a foundation for advocating policy changes to address algorithmic bias and ensure ethical AI development.

How does Unmasking AI propose solutions for algorithmic bias?

  • Advocacy for Algorithmic Justice: Buolamwini calls for ethical considerations in AI development, encouraging conversations about societal impacts and equitable solutions.
  • Diverse Representation in Tech: The book highlights the importance of including diverse voices in technology development to ensure fair AI systems.
  • Policy Recommendations: Buolamwini suggests collaboration between policymakers and tech companies to create regulations addressing algorithmic bias.
  • Accountability and Transparency: Emphasizes the need for accountability and transparency in AI systems to prevent discrimination and harm.

How can individuals contribute to the movement for algorithmic justice?

  • Educate Yourself and Others: Buolamwini encourages learning about AI and its societal impacts, sharing knowledge to build an informed community.
  • Advocate for Change: Participate in advocacy efforts, such as supporting legislation addressing AI bias and discrimination, and engage with local organizations.
  • Challenge Harmful Practices: Question and challenge AI technologies that perpetuate harm, holding companies accountable for a just technological landscape.
  • Empowerment Through Action: Individuals can contribute to a more equitable future by actively participating in movements for algorithmic justice.

About the Author

Dr. Joy Buolamwini is a computer scientist, researcher, and activist known for her work on algorithmic bias in AI. She founded the Algorithmic Justice League to combat discrimination in AI systems. Buolamwini gained recognition for her research on facial recognition technologies and their biases against women and people of color. She has presented her findings to policymakers and tech companies, advocating for more inclusive and ethical AI development. Buolamwini is also known as the "Poet of Code" for integrating poetry into her technical work. Her efforts have significantly influenced discussions on AI ethics and fairness.

Download PDF

To save this Unmasking AI summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.31 MB     Pages: 21

Download EPUB

To read this Unmasking AI summary on your e-reader device or app, download the free EPUB. The .epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.
Download EPUB
File size: 1.44 MB     Pages: 26
Want to read the full book?
Follow
Listen8 mins
Home
Swipe
Library
Get App
Try Full Access for 3 Days
Listen, bookmark, and more
Compare Features Free Pro
📖 Read Summaries
Read unlimited summaries. Free users get 3 per month
🎧 Listen to Summaries
Listen to unlimited summaries in 40 languages
—
❤️ Unlimited Bookmarks
Free users are limited to 4
—
📜 Unlimited History
Free users are limited to 4
—
📥 Unlimited Downloads
Free users are limited to 1
—
Risk-Free Timeline
Today: Get Instant Access
Listen to full summaries of 26,000+ books. That's 12,000+ hours of audio!
Day 2: Trial Reminder
We'll send you a notification that your trial is ending soon.
Day 3: Your subscription begins
You'll be charged on Nov 25,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
1M+ readers
Trustpilot Rating
TrustPilot
4.6 Excellent
This site is a total game-changer. I've been flying through book summaries like never before. Highly, highly recommend.
— Dave G
Worth my money and time, and really well made. I've never seen this quality of summaries on other websites. Very helpful!
— Em
Highly recommended!! Fantastic service. Perfect for those that want a little more than a teaser but not all the intricate details of a full audio book.
— Greg M
Save 62%
Yearly
$119.88 $44.99/year/yr
$3.75/mo
Monthly
$9.99/mo
Start a 3-Day Free Trial
3 days free, then $44.99/year. Cancel anytime.
Unlock a world of fiction & nonfiction books
26,000+ books for the price of 2 books
Read any book in 10 minutes
Discover new books like Tinder
Request any book if it's not summarized
Read more books than anyone you know
#1 app for book lovers
Lifelike & immersive summaries
30-day money-back guarantee
Download summaries in EPUBs or PDFs
Cancel anytime in a few clicks

Why one summary?

Scanner
Find a barcode to scan

⭐
⭐
⭐
We have a special gift for you
Open
38% off
DISCOUNT FOR YOU
⭐
⭐
⭐
$79.99
$49.99/year
Only $4.16 per month
Continue
2 taps to start, super easy to cancel
Settings
General
Widget
Loading...