Each deck guides studio time for its lesson: questions answered on students' devices, work in pairs and small groups, peer activities, and discussion as a class. A deck can also be presented without a room, worked through on your own (the recorded walkthroughs are being prepared, and until a deck's walkthrough is recorded its self-study format runs without the voice), or printed as a facilitator guide. All 10 decks are ready.
Lessons
10 of 10 decks readyL · 01
AI Literacy as a Public Health Skill: Machine Learning and Neural NetworksOne New Brunswick public health office trying an approved assistant and weighing a vendor's model that predicts which children will be overdue for their 18-month vaccines: the premise of the course, the five parts of AI literacy, the evidence on use and on gains and the check for each kind of output; nested fields, rules and learned models, the three ways machines learn, held-out and external data, sensitivity worked by hand and prediction compared with counterfactual prediction; one neuron and one step of gradient descent worked by hand, and underfitting and overfitting in error curves; and the history from Turing and the AI winters to ImageNet, ChatGPT, Canada's part and the 2024 Nobel Prizes.
L · 02
How Large Language Models and AI Agents WorkOne Nova Scotia town's public health office preparing for its heat season with an approved assistant and a document assistant, from the generation loop, tokens, embeddings, attention and parameters, temperature and the ratio of two probabilities, the context window and scaling, through pretraining, instruction tuning, reinforcement learning from human feedback, sycophancy, constitutional AI, the system prompt, variation, the training cut-off and confident errors, chain-of-thought prompting, reasoning models, retrieval-augmented generation, tools and how Lewis answers, and one question traced claim by claim with document search off and on, to the agent loop, workflows and agents, the forms of agent in 2026, prompt injection, checkpoints for human review and the check each kind of task needs.
L · 03
Working with AI: Prompting, Context and VerificationOne British Columbia coastal town's heat season with an institutional AI assistant: the jagged frontier, the consulting experiment, centaurs and cyborgs and known-answer tests; the six parts of an effective request, five techniques, four ways to supply context and one request refined four times; five kinds of error, sycophancy, four core checks, a recomputation and checks scaled to the stakes; and the four classes of information, consumer and institutional accounts, SFU's guidance, APA citation of a chat and the AI use log.
L · 04
AI for Learning and StudyOne study group of four fourth-year students in a Nova Scotia coastal town preparing for a closed-book examination: tutoring as the benchmark, the Harvard physics trial worked by hand, the high school experiment in Turkey, the three features that separated help from harm, the principle of who does the thinking and what a cross-sectional survey of offloading can show; retrieval practice, the utility ratings of study techniques, a tutoring prompt that withholds answers, explanations at three levels, a generated quiz key checked with its source and the built-in learning modes; a reading guide, terms in context, a risk ratio recomputed and an AI summary checked claim by claim, and notebook tools; and Lewis, a weekly routine with an unaided step in each stage, the occasions for studying without AI, the illusion of explanatory depth and calibration.
L · 05
AI-Assisted Literature Searching and ReadingOne Newfoundland and Labrador town's evidence scan on community paramedic home visits for older adults: the four categories of AI research tool, how a ranked AI search differs from a bibliographic database in recall, reproducibility and transparency, and what a task needs before a tool is chosen; the six-part topic map, the concept table, checking headings, the known-item test, citation chasing, the librarian's review and the search record with the AI use log; extraction tables and the errors that flatter the evidence, the evidence on extraction accuracy, the checking routine and summaries without overgeneralization; and why references are fabricated or misattributed, what retrieval changes, the four-step citation check with its four outcomes, the depth of checking and the rules for AI in systematic reviews.
L · 06
AI-Assisted Academic and Professional WritingOne Vancouver Island town's heat warning check-in program and the four writing tasks of its co-op student: the six stages of writing, interview, outline and reverse-outline prompts, four ways to divide the drafting and the experiments on speed, quality and convergence; feedback against criteria and from a named reader, the five checks for every suggested change, second-language editing and stock phrasing; the ICMJE, COPE, publisher, tri-agency and SFU rules, why AI-text detectors misclassify writing and disclosure statements; and briefing notes, plain language and the Flesch-Kincaid grade level, a claim-by-claim check, adapted messages and the review of machine translations.
L · 07
AI as a Thinking Partner: Generating and Testing IdeasOne Saskatchewan town's pop-up blood pressure program and the co-op student who uses the team's assistant as a thinking partner: the four stages of idea work, the experiments on the number, quality and variety of ideas from language models and five techniques that widen the range; the testing loop, the strongest objections, steelmanning, other explanations for a fall in blood pressure and a check of a causal diagram; the four forms of sycophancy, how training on human preferences rewards agreement, countermeasures and the survey on confidence in AI and critical thinking; and FINER scores corrected with local facts, a pre-mortem on a study, a logic model review and what simulated perspectives can and cannot do before engagement.
L · 08
AI for Data Analysis and Coding in ROne British Columbia interior town's fall flu clinic outreach and a co-op student's first analysis in R of a public community health survey: the four forms of AI coding assistance, what an assistant knows about the data, the two randomized studies of programming speed and where an assistant helps a beginner; the six-step workflow, the data dictionary, eight pieces of R notation, warnings and errors, explanations and the AI use log; silent errors, the four checks, a blank-answer code counted as a flu shot, age groups with the wrong boundaries, the garden of forking paths and the pre-specified plan; and four classes of data, the reasons that keep confidential data out of consumer services, Population Data BC, help with code without sharing the records and the OCAP® principles.
L · 09
AI in Public Health PracticeOne population health team in the southern Interior of British Columbia weighing a news-scan service, a vendor's falls risk score, a public chatbot and routine use of an enterprise assistant: indicator-based and event-based surveillance, GPHIN, BlueDot and EIOS, the SARS and Wuhan records, chief complaint classifiers, the precision and recall of a filter worked from a sample of discards and the drift of Google Flu Trends; score, threshold and action, sensitivity and positive predictive value worked by hand, a proxy label, a proprietary sepsis model, CHARTwatch, TRIPOD+AI and six questions for any model; four levels of evidence on language models, the Goh trial, three chatbot designs, S.A.R.A.H., misinformation, TRIPOD-LLM and CHART; and routine tasks and their checks, Cohen's kappa worked by hand, the FASTER principles, the B.C. framework and the analyst's task map.
L · 10
Responsible Adoption and Continued LearningOne northern British Columbia town's public health office, eight months into using an enterprise assistant, and its new analyst: the five routes by which AI can widen inequities, the uses that can narrow them, OCAP®, the CARE Principles and the B.C. framework's commitments; the six layers of governance, the WHO guidance, the NIST framework, the EU AI Act and its 2026 amendment, the federal picture since AIDA lapsed, the B.C. framework's requirements and the energy used by a team's prompts; a six-step workflow with a prompt library, verification tiers, an AI use log, tools matched to tasks and the audit of an AI-drafted workflow; and nine questions for any new tool, structured trials, ways to stay current, deskilling and a schedule for continued learning.
Students: join studio time. Open the join page and enter the four-letter code on the main screen, or scan the QR code.
Join studio time