Society of Minds

• Why this course exists — and what Northeastern means by AI readiness
• Marvin Minsky's Society of Mind, and where this course gets its name
• How the course runs: what you will build, what we read, what I expect
• A live experiment — this class's own feelings about AI, analysed in real time
• Phrase of the day: self-selection bias — and how I disclose my own use of AI

• Getting everyone into Claude Desktop — Northeastern sign-in, and a Project set up for Homework 1
• Three ways to use Claude: in a browser, as a desktop app, and Claude Code
• A search engine finds, a chatbot writes — and why that difference matters
• An experiment: the same question asked both ways, then compared
• Citation failure in three levels — Phrase of the day: citation fabrication

• Phrase of the day: “2023 Thinking” — confident opinions built on early encounters
• Early encounters: qualitative reasoning, common sense, and Cyc's ten million rules
• What changed — research, data analysis, and coding as everyday work
• Vibe coding: building a data dashboard by describing what you want, not how
• Real projects, from Kyoto's cherry blossoms and the phases of Venus to ten million variable stars

• Discussion: Everybody's Lost Their Minds — Jan Schaumann's counterpoint to the AI hype
• An early encounter with AI: the man hanging from a rope over shark-infested waters
• What design actually means, and what makes game design a design problem
• Is a human in the loop necessary — or, as the reading argues, a rubber stamp?

• How AI image generation changed in three years, and what that rate implies
• Can a language model be funny? Sarcasm, timing, and what humor needs
• Cyc, common sense, and the ten million rules written by hand
• AI in Action: what AI could and could not do in 2025 — and what it can infer about you
• Research built this way: rocksinspace.org, thecosmicweb.org, and arcade games then and now

• An introduction to Claude Code — and how to use it for Homework 2
• Starting a session: environment, project folder, model, and permission mode
• Watching Claude work — and reading the diff before you accept it
• Arranging the workspace, and running more than one session at a time
• Starter prompts to copy, and what a good correction sounds like

• Word of the day: data science — getting defensible answers out of data
• Big data was a change in circumstances, not a technique
• Databases: unglamorous, almost invisible, and everything runs on them
• Machine learning turns the problem around — rules out, not rules in
• Data mining: patterns nobody asked for, and the ones that are only chance

• What forty job roles add up to — which work is automated, which is quietly rewritten, which is barely touched
• Walking the AI in the Workplace pages on the course site, sector by sector
• The explainer video — a synthesis of the whole class, in one sitting
• Which citations turned out not to exist, and how their authors found out
• Where this goes next: the foundations of AI — logic versus connectionism, and the Turing Test

• Installing Anaconda — giving Claude Code a Python toolbox
• Phenology, not phrenology: reading the climate in the blossoms
• Tracking down real data, and checking two sources agree before splicing them
• From a first plot to a publication-ready figure, one sentence at a time
• The result: a two-page site with an interactive explorer — and your turn

Deep learning and computer vision — from optical character recognition to facial recognition and self-driving cars. What a network is actually looking at, and why that is hard to find out.

How companies use machine learning on customer data: churn prediction, market segmentation, and the recommendation systems behind Netflix, Amazon, and Spotify.

Learning techniques drawn from evolution by natural selection. Populations, mutation, selection pressure — and watching a computer solve a problem it was never told how to solve.

Optimization with many competing objectives, from manufacturing to medicine to scheduling. What it means for a decision to be good when the goals disagree.

Chatbots from 1966 to now. How a system trained only to predict the next word ends up holding a conversation.

Tokens, context, attention, and why these systems confabulate. What is actually happening when you type into a chat box.

We play each other's arcade games and critique them — as humans. AI can generate a game in ninety seconds; only a person can tell you it isn't fun.

False positives, false negatives, and what accuracy actually measures. Real failures in legal and hiring systems, and why a good score can hide a useless model.

Systematic methods for verifying what an AI tells you: cross-referencing, checking sources, and knowing which claims should never be trusted without one.

• Where bias enters — data, labels, proxies, and the choice of what to optimize
• Hiring, parole, and lending: three systems that went wrong in public
• What building fairness in would actually require
• Time permitting: the alignment problem — making a system want what we want

Surveillance, inference, and consent when a system can deduce what you never disclosed. GDPR, CCPA, and whether consent still means anything.

• Copyright and the training-data fights now in court
• Liability: who answers when the model is wrong
• Regulation against a technology that moves faster than legislation
• The costs nobody bills — energy, water, and the people who label the data

Checkpoint. You show what you have, get feedback from the room, and fix what isn't working.

• The rise of big data, and why databases hold the world up
• Beyond bar charts — what better visualization makes visible
• AI as research collaborator in drug discovery, astronomy, and climate
• What it changes about reproducibility and peer review
• Time permitting: VariableZoo.org and BioXplore.org, built with Claude Code

• Generative AI in writing, design, music, and journalism
• Collaboration versus replacement — what the tool is actually doing
• Authorship and attribution: who made it, who owns it
• What "original" means when the machine has read everything
• Time permitting: using AI to actually learn something, rather than to avoid learning it

Competing visions of humanity alongside AI, and your own view of where this goes.

You present what you built.

With fifty students, one showcase day is not enough. This slot has no date yet.