書海Apsû
Chart Navigator Guide
Register as a voyager

ApsûVoyage of a Thousand Books

Chart of the Sea of Books

Drag or zoom the chart. Tap an island to go ashore.

← Back to the chart

← Back to the island · Lofoten
Rowan Tree
Finish the notes and leave a bottle to get the Rowan Tree
Register to collect this creature
The Alignment Problem
Original Title:The Alignment Problem
Author:Brian Christian ⓘ
The Alignment Problem

0:00
    The question is not only "how does a machine maximize its reward," but "how should we design the reward so that it does what we truly want."
    Reading Progress 80%
    ▶ Not in the mood to read? Here's the video summary :)
    ▶ Video summary coming soon
    Sail to Lofoten

    Who is this book for?

    • People who want to know why AI learns biases and how to correct them
    • Readers interested in social issues such as algorithmic fairness and AI transparency
    • Students of computing, psychology or social science who want to see how the fields meet
    • People who want to understand machine learning through stories rather than formulas

    Who is the author?

    • Brian Christian, American author who specializes in writing about computer science alongside philosophy and psychology for general readers
    • He co-authored Algorithms to Live By with the cognitive scientist Tom Griffiths
    • The book was published in 2020 as The Alignment Problem: Machine Learning and Human Values

    Other BooksAlgorithms to Live By (with Tom Griffiths);The Most Human Human

    Reading notes

    🔒 Arrive at Lofoten and visit the Rowan Tree to unlock this

    Sail to Lofoten
    The Book's Structure: Prophecy, Agency, Normativity

    The book is divided into three parts of three chapters each, moving from the basics to more advanced material.

    Part One, "Prophecy," deals with machine learning's predictive systems, with chapters on "Representation," "Fairness" and "Transparency"; Part Two, "Agency," deals with systems that act on their own, with "Reinforcement," "Shaping" and "Curiosity"; Part Three, "Normativity," deals with how machines learn what humans want, with "Imitation," "Inference" and "Uncertainty."

    Christian draws on numerous interviews with researchers and real cases to link technical problems with human psychology and behavioral science, so that readers without a technical background can follow.

    Representation: Who Is Missing from the Data
    Fairness: Mathematically Impossible to Have Both
    Transparency: The Black Box Is Not Inevitable
    Reinforcement: Learning from Reward
    Shaping and Curiosity
    Imitation and Inference
    Uncertainty and Corrigibility
    What Others Say

    Navigators' bottles Reader reviews

    No bottles have drifted to this book yet
    Cover is an original stylized design, not the publisher's actual cover art

    Leave a bottle

    Share your thoughts! You might meet voyagers who think just like you.

    I want to recommend this book to another voyager
    My rating Drag to choose 0.0 – 5.0
    4.0
    Sail to Lofoten to visit the Rowan Tree

    Register as a voyager to open the bottles.

    0 bottles

    ▇▇▇▇▇ ▇▇▇ ▇▇▇▇▇▇▇

    ▇▇▇▇ ▇▇▇▇▇▇ ▇▇▇

    Shang

    Shang

    Shang's books

    Voyager's Guide

    This world

    The Sea of Books has 16 real islands, one for each subject. Every building on an island is a book.

    What you can do now

    • Tap an island on the chart to go ashore
    • Tap a building to read the notes on that book
    • Meet Shang, the Navigator

    Coming soon

    • Chapter 2: register as a voyager, sail your ship, collect building badges, send message bottles
    • Chapter 4: private chats with other voyagers

    Symbols

    Change language

    Navigator's registry

    Before you set sail, we'll register your ship.

    Logbook

    Inbox

    Rankings

    Logbook

    Review bottles

    Shang

    書海 Apsû

    回海圖
    APSÛ // COMM●