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
Machine Learning

Machine Learning

A model can learn patterns from examples, but its assumptions decide what it will miss.
by Ethem Alpaydin 2016 230 pages
3.62
1k+ ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Replace hand-written rules with examples when the rules are unknowable

From code to evidence. Machine learning is a way to build programs when people can specify the desired result but cannot write the rules that produce it. Instead of coding every spam signal, for instance, developers provide examples of spam and legitimate messages. A learning algorithm adjusts a model, a flexible template with changeable parameters, to fit those examples. The trained model can then classify messages it has not seen before.

Data becomes a resource. Cheap storage, connected computers, smartphones, and digital services have created vast records of purchases, images, speech, and movement. A supermarket can use transaction histories to detect recurring buying patterns and make predictions. But more data is not automatically useful: a phone book does not reveal a reliable rule for predicting a new person’s phone number from their name.

Analysis

The shift is not simply from human work to automation. It is from specifying procedures to specifying evidence and goals. That is powerful when expertise is difficult to articulate, as with recognizing faces or filtering evolving spam. It also changes who shapes a system: whoever chooses the examples, labels, and success measure quietly defines what the machine is taught to value. The practical question is therefore not only whether a model learns, but whose experience its training examples represent.

A model learns only what the data's regularities let it generalize

Beyond memorization. Generalization means how well a model trained on examples predicts cases it has not encountered. It is possible because similar inputs often have similar outputs: cars with roughly comparable attributes tend to have roughly comparable prices. A learner that merely stores every training example has memorized, not learned.

Assumptions fill the gaps. A short number sequence can fit many rules. The longer Fibonacci sequence points toward a simple pattern, while a version with a couple of altered terms could be explained either by a complicated exact rule or by a simpler rule plus errors. Choosing among possibilities requires an inductive bias, the assumptions a learning method uses to prefer some explanations over others. No method can guarantee that a pattern in past data will persist when the world changes.

Analysis

Generalization is a form of disciplined induction: evidence supports a rule, but never logically proves that the rule will hold in every new case. The book’s emphasis on regularity makes learning intelligible, while also pointing to a central risk: a model can perform well on familiar conditions and fail when those conditions shift. Scientific models face the same problem. Their value depends not just on fit, but on whether the assumptions that connect observations to prediction remain credible.

Choose the model's assumptions before polishing its parameters

Structure comes first. A model states what kind of relationship is possible; learning adjusts its parameters to fit the data. A linear model, for example, estimates a car’s price by adding weighted contributions from attributes such as mileage. It is simple, interpretable, and often useful, but may miss relationships that are not additive. The book notes that a typical vehicle can lose about 15 percent of its value each year, a pattern better represented by compounding than by a fixed yearly dollar reduction.

Fit the task's scale. One global model may be too crude if sports cars and sedans depreciate differently. At the other extreme, nearest-neighbor estimation predicts from similar past cases, such as averaging the prices of the three most similar cars. More flexible models can capture more detail, but can also mistake noise for a real pattern.

Analysis

This is a useful antidote to treating model training as a button that turns data into truth. Parameter tuning cannot repair a badly chosen representation or an assumption that excludes the real relationship. Simplicity matters because it makes explanations easier to inspect and can reduce the chance of fitting accidental quirks. Yet simple models are not automatically fair or accurate. Model choice is a substantive judgment about which differences matter, and that judgment should be made in light of the task and its consequences.

Make decisions by weighing error costs, not probability alone

Errors have unequal prices. A classifier might estimate whether a patient has a disease, but the most probable answer is not always the best decision. Missing a serious illness can be far worse than ordering an unnecessary follow-up test. The system may therefore recommend further investigation even when the estimated disease probability is below one half. Expected value combines probabilities with the gains or losses attached to each possible outcome.

Some choices resist arithmetic. A novel in the book describes a mother rejecting reassurance that a snakebite fatality rate is only a few percent: a small risk still matters greatly when the stakes concern her only child. In cases where either mistake is costly, a system can abstain and send the case to a human expert. A decision rule needs to account for the cost of being wrong, not just predictive accuracy.

Analysis

This distinction separates prediction from decision. A model estimates what may happen; a person or institution decides what consequences are acceptable. Decision theory clarifies why a threshold should change with the cost of false positives and false negatives. But assigning those costs is not a neutral technical step. In medicine, finance, and public services, the people who bear the harms may not be the people who set the threshold. An abstention option can preserve human judgment, provided the human reviewer has time and authority to act.

Teach recognition to preserve identity and discard irrelevant variation

Separate signal from appearance. Handwriting recognition cannot store every possible version of a letter, so a learner must infer what different examples share. Face recognition poses a harder version of the same problem: identity should remain recognizable despite changes in pose, lighting, hairstyle, or facial expression. A generative model describes observations through hidden causes, such as identity and appearance, then estimates which causes best explain the image.

Define what should stay constant. Invariance means treating a change as irrelevant to the task. A character recognizer should generally ignore size or slant, but not rotation in every direction: turning one letter can make it resemble another. The right features depend on the goal. A system identifying a person may ignore expression, while a system reading emotion must pay attention to it.

Analysis

The deeper lesson is that recognition is not the extraction of every available detail. It is selective attention shaped by purpose. This connects machine perception to measurement in science: a thermometer ignores color because color is irrelevant to temperature, while a face system cannot assume that every variation is harmless. The design challenge is to distinguish nuisance variation from meaningful variation. That distinction deserves particular scrutiny when the system’s categories influence access, identity checks, or judgments about people.

Use deep networks to learn layered features, not just outputs

Abstraction builds in stages. A neural network contains connected processing units whose connection weights are adjusted during learning. In a multilayer network, early layers can combine image pixels into edges, later layers into shapes, and still later layers into objects such as letters or faces. Deep learning refers to methods that train networks with multiple levels of representation, allowing useful features to be learned rather than specified entirely by hand.

Architecture still matters. A convolutional network connects units to local patches, reflecting the fact that nearby pixels often belong together. That structure reduces the number of connections the system must learn. The perceptron, an early neural model, could not solve the simple exclusive-or problem on its own; training multilayer networks with backpropagation helped revive neural-network research. Deep systems can learn powerful representations, but the book stresses that their capabilities remain limited to particular tasks and settings.

Analysis

Layered representations offer a compelling bridge between raw measurement and meaningful categories. They also explain why deep learning can reduce the burden of manually designing features. Yet learning representations does not mean learning without assumptions: the network architecture, training examples, and objective all shape what becomes visible to the model. A system may recognize a pattern reliably without providing a human-readable account of why. In high-stakes applications, predictive strength and interpretability are separate properties, not interchangeable signs of understanding.

Let unlabeled data expose groups and hidden causes

Find structure without answers. Unsupervised learning looks for patterns in inputs without a supervisor supplying the correct output. Clustering groups similar examples; a company might use it to segment customers and tailor services to different groups. Such analysis can reveal a grouping that experts did not anticipate, though the result depends on how similarity is defined.

Infer the factors behind behavior. Recommendation systems can model customer purchases through hidden factors rather than relying only on direct item-to-item associations. A latent factor might reflect the presence of a baby at home, which could help explain purchases across several baby-care products. Matrix decomposition represents customers and products through a smaller set of such factors. The same idea can organize movie ratings or discover topics in documents, even when most users have rated only a small fraction of available movies.

Analysis

Unsupervised methods are valuable because labels are expensive and can impose categories that obscure unexpected structure. A cluster, however, is evidence of similarity under a chosen representation, not proof that its members share a real-world cause. Likewise, a latent factor may predict purchases without corresponding neatly to an identifiable human trait. Treating these constructs as useful summaries rather than discovered essences makes recommendations easier to question and less likely to turn statistical convenience into a fixed story about a person.

Reward learning must explore before exploiting its best guess

Feedback arrives late. In reinforcement learning, an agent takes actions in an environment and receives rewards or penalties. Unlike a supervised teacher, a critic does not identify the correct action in advance. A chess program or maze-solving robot may receive meaningful feedback only after many moves, creating the credit assignment problem: which earlier actions helped produce the final outcome?

Learn value along the path. Temporal difference learning estimates the value of intermediate actions by backing up future rewards. If reaching a goal earns 100 points, an action one step earlier might receive an internal value of 90 when future rewards are discounted by 0.9. The agent must also balance exploration, trying uncertain actions, with exploitation, choosing actions already believed to work. TD-Gammon reached master-level play after 1.5 million games against itself.

Analysis

Reinforcement learning makes the source of training data explicit: the agent generates it by acting. That can discover strategies no teacher supplied, but it also makes exploration costly when actions affect people, equipment, or money. The reward is a proxy for the real objective, and an agent can optimize the proxy while missing the purpose behind it. Simulated environments and careful reward design therefore matter, especially when a successful game-playing method is transferred to settings where mistakes have physical or social consequences.

Audit learned decisions before trusting them with people or machines

Accuracy is not enough. Data mining includes preparing data, fitting a model, testing it on examples withheld from training, and interpreting the result. A decision tree may expose readable rules, while a neural network can be difficult to inspect. Testing on unseen examples estimates future performance, but it cannot guarantee success for underrepresented cases or a changing environment.

Protect people and preserve choice. Removing names from a dataset may not make it anonymous; combinations of details such as birth date and postal code can still identify someone. The book calls for transparency about collection and use, data security, and the ability to inspect, correct, or remove personal data. It also warns that recommendations based only on past preferences can narrow exposure to new music, films, books, and opinions.

Analysis

The book’s ethical concerns belong inside the technical workflow, not in a policy appendix after deployment. Data selection defines whose behavior is legible, validation reveals only some kinds of failure, and anonymity can be undermined by linking records. Recommendation systems also shape the preferences they appear merely to predict. The practical standard should be broader than accuracy: examine who benefits, who carries error, what information is retained, and whether people can contest or escape a system’s decisions.

Analysis

Alpaydın’s central argument is that machine learning is not magic added to computing. It is a change in how a computational task is specified. When explicit rules are unavailable, examples, a model structure, and a performance criterion take their place. The method succeeds because the world contains repeatable patterns, but the model’s assumptions determine which patterns can be seen and which will be ignored. This makes the book’s strongest contribution conceptual: it connects learning to estimation, scientific model-building, perception, and action instead of treating algorithms as isolated tricks.

The progression from supervised prediction to recognition, unsupervised discovery, and reinforcement learning also reveals distinct kinds of evidence. Labels tell a learner what outcome to imitate. Unlabeled observations can expose structure but do not establish what that structure means. Rewards guide action, yet arrive too late to identify the responsible choice directly. These are not interchangeable forms of learning. Their differences matter whenever a system’s prediction is mistaken for an explanation, or a rewarded behavior is mistaken for an ethically sound one.

The account is especially valuable in its insistence that generalization is the objective, not reproduction of training data. Its limitation is that the ideal of a relatively simple hidden pattern can understate how unstable and socially produced data may be. A historical transaction record reflects both customer behavior and the systems, prices, and choices that shaped what was recorded. Modern practice therefore benefits from adding causal questions: would the predicted relationship survive if policy or context changed, and could acting on it alter the very data used to justify it?

The book’s closing concerns follow directly from its technical foundations. Once models make decisions at scale, privacy, transparency, distribution of errors, and the possibility of human review are part of system design. The most useful reading is neither uncritical enthusiasm nor fear of autonomous machines. It is a demand to match model ambition to evidence, test assumptions under change, and treat the people represented in data as participants with interests, not merely as useful observations.

Summary in 30 Seconds
Machine learning builds models from examples when explicit rules are unavailable, but success depends on patterns that persist beyond the training data. A model's structure and assumptions shape what it can learn, while greater flexibility can also fit noise. Good decisions weigh the costs of different errors, not probability alone. Learning may use labeled examples, unlabeled data, or delayed rewards. Testing, privacy safeguards, transparency, and human review matter when predictions affect people.
Contains spoilers

Last updated:

Report Issue

Review Summary

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

Machine Learning receives mixed reviews, with an average rating of 3.62/5. Readers appreciate its accessible introduction to machine learning concepts, though some find it too basic or dry. The book is praised for balancing technical details with high-level explanations, making it suitable for non-experts seeking an overview. However, some criticize its lack of practical examples and outdated information. While valuable for understanding fundamental concepts, it may not satisfy those looking for in-depth technical knowledge or hands-on application.

Your rating:
4.25
176 ratings
Want to read the full book?

Glossary

Machine learning

Programs fitted from examples

A computational approach that adjusts a model using data so it can perform a task, such as classifying messages or estimating prices. It is especially useful when people can define the desired input and output but cannot write reliable rules connecting them. Learning aims to make useful predictions beyond the examples used to train the model.

Generalization

Prediction on unseen cases

The ability of a model trained on a sample of examples to make accurate predictions for new instances not used during training. Generalization depends on the suitability of the model, the quantity and quality of training data, and the stability of patterns between past and future cases. Memorizing the training examples alone is not generalization.

Inductive bias

Assumptions guiding learning

The assumptions a learning algorithm uses to choose among possible models when the available data does not uniquely determine an answer. For instance, favoring a smooth relationship between nearby inputs and outputs is one possible bias. These assumptions make generalization possible, but they can also lead a learner away from the right explanation when they do not fit the task.

Generative model

Model of data-producing causes

A model that represents how observed data may arise from underlying, often hidden factors. In character recognition, the visible image may depend on both the character’s identity and its appearance, such as font or writing style. Learning these dependencies can help explain observations and infer likely hidden causes from them.

Deep learning

Learning layered representations

Methods for training neural networks with multiple layers that transform raw inputs into increasingly abstract features. In image recognition, layers may progress from pixels to edges and shapes, then to objects or characters. Deep learning can learn useful features from examples, reducing the need to specify every feature manually, though architecture and training data still shape what is learned.

Unsupervised learning

Learning structure without labels

A type of machine learning in which examples have no supplied target output or class labels. The learner searches the inputs for regularities, such as groups of similar customers or recurring topics in documents. It can reveal patterns not anticipated by an expert, but the discovered groupings depend on how the data is represented and similarity is measured.

Reinforcement learning

Learning actions from rewards

A learning approach in which an agent interacts with an environment, takes actions, and receives rewards or penalties. Feedback may be delayed until a sequence of actions is complete, so the agent must estimate which earlier actions contributed to the outcome. It learns through its own attempts, balancing exploration of actions with use of actions already believed to work.

Temporal difference learning

Backing up future reward

A family of reinforcement learning methods that updates an estimate of an action’s value using information about the next state and its estimated value. This allows credit for a future reward to be assigned incrementally to earlier actions, rather than waiting to evaluate an entire sequence. Q learning is one example discussed in the book.

Dataquake

Surge of digitally generated data

Alpaydın’s term for the rapid growth of data produced as computers, networks, online services, smartphones, and smart devices become part of everyday life. The resulting records create opportunities for data analysis and machine learning, because recurring patterns may be extracted from observations too numerous for people to examine manually.

FAQ

What's "Machine Learning: The New AI" by Ethem Alpaydın about?

  • Overview of Machine Learning: The book provides a comprehensive introduction to machine learning, explaining how it enables computers to learn from data and improve their performance over time.
  • Historical Context: It traces the evolution of machine learning from its early days to its current status as a key component of artificial intelligence.
  • Applications and Implications: The book discusses various applications of machine learning, from facial recognition to autonomous vehicles, and explores its implications for the future of technology and society.
  • Educational Resource: It serves as an educational resource for those new to the field, offering insights into the fundamental concepts and methodologies of machine learning.

Why should I read "Machine Learning: The New AI" by Ethem Alpaydın?

  • Comprehensive Introduction: The book is an excellent starting point for anyone interested in understanding the basics of machine learning and its role in artificial intelligence.
  • Expert Insights: Written by a leading expert in the field, it provides authoritative insights into the principles and practices of machine learning.
  • Real-World Applications: It offers numerous examples of how machine learning is applied in various industries, making the concepts more relatable and easier to understand.
  • Future Implications: The book discusses the potential future developments in machine learning, helping readers understand its long-term impact on technology and society.

What are the key takeaways of "Machine Learning: The New AI"?

  • Learning from Data: Machine learning involves creating algorithms that allow computers to learn from and make predictions based on data.
  • Supervised vs. Unsupervised Learning: The book explains the difference between supervised learning, where models are trained on labeled data, and unsupervised learning, which involves finding patterns in unlabeled data.
  • Neural Networks and Deep Learning: It covers the basics of neural networks and the concept of deep learning, which involves training models with multiple layers of abstraction.
  • Ethical Considerations: The book highlights the ethical and privacy concerns associated with machine learning, emphasizing the need for responsible use of technology.

What are the best quotes from "Machine Learning: The New AI" and what do they mean?

  • "A quiet revolution has been taking place in computer science for the last two decades." This quote highlights the transformative impact of machine learning on the field of computer science.
  • "Machine learning is not just the commercial application of methods to extract information from data; learning is also a requisite of intelligence." It underscores the importance of machine learning in developing intelligent systems.
  • "The aim of this book is to give the reader an overall idea about what machine learning is, the basics of some important learning algorithms, and a set of example applications." This quote encapsulates the book's purpose as an educational resource for understanding machine learning.

How does Ethem Alpaydın explain the concept of supervised learning in "Machine Learning: The New AI"?

  • Definition and Process: Supervised learning is described as a process where models are trained on labeled data, meaning the input data is paired with the correct output.
  • Examples Provided: The book provides examples such as credit scoring and image recognition to illustrate how supervised learning is applied in real-world scenarios.
  • Model Training: It explains how models learn to make predictions by minimizing the difference between their predictions and the actual outcomes in the training data.
  • Importance of Labeled Data: The book emphasizes the necessity of having a large and diverse set of labeled data to train effective supervised learning models.

What is unsupervised learning according to "Machine Learning: The New AI"?

  • Definition and Purpose: Unsupervised learning involves finding patterns or structures in data without pre-existing labels, aiming to discover hidden relationships.
  • Clustering and Association: The book discusses methods like clustering, which groups similar data points, and association, which finds rules that describe large portions of the data.
  • Applications: Examples include customer segmentation and market basket analysis, where unsupervised learning helps identify natural groupings or associations in data.
  • Exploratory Nature: It highlights the exploratory nature of unsupervised learning, which can reveal insights that were not previously considered.

How does "Machine Learning: The New AI" describe neural networks and deep learning?

  • Neural Networks Basics: The book explains neural networks as models composed of interconnected nodes (neurons) that process information in layers.
  • Deep Learning Concept: Deep learning is described as using neural networks with many layers to learn complex patterns in data, allowing for high-level abstractions.
  • Applications and Successes: It discusses the success of deep learning in fields like image and speech recognition, where it has achieved state-of-the-art results.
  • Training Challenges: The book also addresses the challenges of training deep networks, such as the need for large datasets and significant computational resources.

What role does probability play in machine learning according to Ethem Alpaydın?

  • Modeling Uncertainty: Probability is used to model uncertainty in data and predictions, allowing for more robust decision-making.
  • Bayesian Methods: The book covers Bayesian methods, which incorporate prior knowledge into the learning process to improve model predictions.
  • Randomness in Data: It explains how probability helps account for randomness and noise in data, which can affect the accuracy of machine learning models.
  • Decision Making: Probability is crucial in decision-making processes, such as calculating expected values to choose the best course of action.

What ethical considerations are discussed in "Machine Learning: The New AI"?

  • Data Privacy: The book emphasizes the importance of protecting individuals' privacy when collecting and analyzing data for machine learning.
  • Bias and Fairness: It discusses the potential for bias in machine learning models and the need for fairness in their development and deployment.
  • Transparency and Accountability: The book advocates for transparency in how machine learning models are trained and used, ensuring accountability for their outcomes.
  • Impact on Society: It highlights the broader societal implications of machine learning, including its potential to disrupt industries and affect employment.

How does "Machine Learning: The New AI" address the future of machine learning?

  • Continued Growth: The book predicts continued growth and integration of machine learning into various aspects of technology and daily life.
  • Advancements in Algorithms: It anticipates advancements in algorithms that will enable more efficient and accurate learning from data.
  • Interdisciplinary Impact: The book foresees machine learning impacting a wide range of fields, from healthcare to finance, by providing new insights and capabilities.
  • Ethical and Regulatory Challenges: It acknowledges the ethical and regulatory challenges that will need to be addressed as machine learning becomes more pervasive.

What is the significance of reinforcement learning in "Machine Learning: The New AI"?

  • Learning from Interaction: Reinforcement learning is described as a method where agents learn by interacting with their environment and receiving feedback in the form of rewards.
  • Applications in Gaming: The book highlights its use in gaming, where reinforcement learning has been used to develop agents that can play complex games like Go and chess.
  • Delayed Rewards: It explains the concept of delayed rewards, where the agent must learn to associate actions with outcomes that occur much later.
  • Exploration vs. Exploitation: The book discusses the balance between exploration (trying new actions) and exploitation (using known actions) to maximize rewards.

How does Ethem Alpaydın explain the concept of clustering in "Machine Learning: The New AI"?

  • Grouping Similar Data: Clustering is described as the process of grouping similar data points together based on their attributes.
  • Unsupervised Learning Method: It is an unsupervised learning method, meaning it does not rely on labeled data to find patterns.
  • Applications in Segmentation: The book provides examples such as customer segmentation, where clustering helps identify distinct groups within a dataset.
  • Exploratory Analysis: Clustering is used for exploratory data analysis, revealing natural groupings that may not have been previously considered.

About the Author

Ethem Alpaydin is a distinguished Turkish computer scientist specializing in machine learning. He received his education from Bogazici University and École Polytechnique Fédérale de Lausanne. Alpaydin has held academic positions at Bogazici University and conducted research at prestigious institutions worldwide. His contributions to the field have earned him numerous awards and recognition. Alpaydin is best known for his book "Introduction to Machine Learning," published by MIT Press in multiple editions and translated into several languages. He has served on editorial boards of prominent journals and is a member of various scientific organizations, including The Science Academy, Turkey, and IEEE.

Download PDF

To save this Machine Learning summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.41 MB     Pages: 21

Download EPUB

To read this Machine Learning 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: 1.43 MB     Pages: 27
Want to read the full book?
Follow
Listen
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 24,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
600,000+ 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...