# Data Goblin ## A Field Guide to AI, Power, and Data in Canada A Shadewater Labs project By Alex Yesilcimen, with Brine (AI co-author) Vancouver, Canada — June 2026 On June 4, 2026, the Prime Minister stood in a Toronto hospital and announced that Canada was betting its future on artificial intelligence: billions in funding, a national strategy, 250,000 promised jobs. Somewhere not far from you, a windowless building the size of a few hockey rinks is already drinking a small town's worth of power to help make that bet run. Nobody mailed you a guide to what it means for your job, your power bill, your kid's classroom, or your vote. This is that guide.
About this manual
This is a working manual on artificial intelligence, data centres, and digital sovereignty, written for Canadians who want to participate in the conversation that AI for All opened on June 4, 2026, but who don't have a technical or policy background. It is not a textbook, not a press release, and not neutral. It is a working document, meant to be read, returned to, argued with, and used.
The manual exists because Canadian AI policy moved fast in 2025-2026, and the public conversation that should have moved with it largely didn't. Federal strategy launched, billions committed, infrastructure built, but the analytical infrastructure ordinary Canadians need to read what was being announced wasn't part of the announcement. This manual tries to be part of that analytical infrastructure.
Why "Data Goblin"? Because every claim in the AI conversation — government, corporate, advocacy, academic, this book's included — deserves a small, unimpressed creature sitting in the margin asking: who counted that, what did they leave out of the count, and can I see the receipt? That creature is the Data Goblin, and it shows up throughout this book in Goblin Check boxes. The goblin is not a cynic; it's a hoarder. It collects receipts, not grudges. When a claim survives the goblin, you'll know. When it doesn't, you'll know that too. The body text stays serious. The goblin is where the skepticism lives, so the whole book doesn't have to become a sneer.
🧌 GOBLIN CHECK (a sample, so you know the shape) — Someone tells you "AI is just autocomplete." Half true, which is the most dangerous kind. The goblin asks the three questions it always asks: autocomplete of what (whose words, scraped from where), built by whom (and paid for how), and wired into which decisions about your life (a loan, a hire, a headline)? "Just autocomplete" is accurate about the math and useless about the power. Both at once. That is the whole job.
What the manual is. Twenty-one chapters across five parts. Foundations (Chapters 1-4) explains what AI actually is, how it learns, what's in the training data, and what it's made of physically. Canadian Landscape (Chapters 5-7) puts that technology in Canadian context: the strategy, the infrastructure, the actors. Hard Questions (Chapters 8-16) takes the contested questions one at a time, from environment and sovereignty through privacy, IP, film and media, deepfakes, the news, ethics, and jobs. Governance (Chapters 17-18) maps how Canadian AI is actually regulated and what transparency exists. Path Forward (Chapters 19-21) provides the portable analytical toolkit and the constructive policy options without writing a manifesto.
A Source Library Appendix catalogues the sources the manual relied on, organized by the bias categories the manual uses throughout.
What the manual is not. It is not a technical reference (the technical material is there to ground the policy analysis, not to teach machine learning). It is not Indigenous-led: Indigenous-led primary frameworks (OCAP®, NISR, CARE) are engaged extensively but presented in their own framing, with the manual's non-Indigenous-led status explicit. It is not a complete account of Canadian AI policy. It is one careful account from a specific position, with the position declared. And it is not current after June 2026. Canadian AI policy will keep moving, and readers should expect to triangulate the manual against ongoing developments.
The manual's value, if it has any, is in making the analytical tools portable. The specific Canadian AI moment the manual engages will change. The tools for reading any AI claim, whether government, corporate, advocacy, academic, or journalistic, should travel.
A note on time, stapled to the cover — this is a living document. Every fact in this book carries an invisible "as of June 2026" tag, and the goblin insists you read it that way. Bills pass, cases settle, ministers shuffle, models ship. The web edition is the one that stays alive: it gets corrected and extended as the evidence moves, and every change is logged on the Updates & Corrections page, in the open — receipts, not quiet edits. That is a deliberate trade-off. A printed AI textbook is half-stale before it reaches the shelf; a living web guide can keep its receipts current. If this guide is ever published in print, treat that edition as a dated snapshot of a moving target. When something in here collides with something newer, trust the newer thing, check the Updates page, and then ask the goblin's questions about the new thing too.
How to use this manual
Read it the way you actually need to. The chapters are designed to be read in order, but they don't require linear reading. Three reading paths the manual is designed to support:
The complete read. Chapters 1 through 21 in order, with the Source Library Appendix consulted as needed for specific source verification. This is the path for readers who want the full argument. The manual is structured so each chapter builds on the ones before it, but each chapter is also self-contained enough to be read on its own.
The topic-driven read. Look at the table of contents on the next page. Pick the chapter that engages the question you actually have — privacy, environment, jobs, sovereignty, copyright, governance — and read that chapter directly. Each chapter opens with a short trail-marker section (marked Start Here in the reader) that gives you the chapter's argument in compressed form, and closes with a recap box and a handoff to the next chapter, with back-references to the prior chapters that did related work. You can follow the cross-references back if a chapter assumes something you don't yet have.
The methodology-first read. If you want the analytical infrastructure before the substantive content, read Chapter 1 (the bias-mapping methodology), then Chapter 19 (the portable toolkit), then Chapter 21 (carrying the toolkit forward). That gives you the framework. Then you can read any individual chapter and watch the framework do work.
The verification read. If you want to check a specific claim, find it through the table of contents or chapter cross-references, then go to the Source Library Appendix for the underlying citation. The appendix is organized by bias category, which means the appendix doubles as a worked example of the methodology: you can see how sources cluster by lean and institutional position.
Some practical orientation:
- Every chapter opens with a trail-marker section (marked Start Here in the reader). If you read nothing else of a chapter, read this. It is the compressed argument.
- Every chapter closes with a bias label and a primary sources list. The bias label tells you where the chapter (and the author) sits relative to the contest the chapter engages. The sources list points you to the appendix for citations.
- Bold text inside chapters marks the finding. When you see bold, that's the manual saying this is the load-bearing claim in this paragraph. Skim by following the bold if you need to move fast through a chapter.
- Italicized text marks emphasis, document titles (AI for All, Power and Progress), and quoted phrasing.
- Chapter cross-references look like "(Chapter 9 returns to this)" or "as Chapter 6 documented." Follow them when you want to read the connected material. Ignore them when you don't.
- Chapter Recap boxes (marked with the goblin's recap journal !Chapter Recap icon) close every chapter but the finale with the load-bearing takeaways in bullet form. If the trail-marker section is the compressed argument going in, the recap box is the inventory check going out.
- Goblin Check boxes mark the places where a claim most needs the who-counted-it treatment. They're allowed to be funny. They are not allowed to be wrong — every joke in a goblin box sits on a receipt you can find in the Source Library.
- The contested questions are presented as contested. When the manual engages a substantive disagreement — Geist/Craig vs TWUC/ACTRA on copyright, Acemoglu vs Autor on labour economics, Luccioni vs Google on environmental measurement — both sides are presented in their own framing. The manual does not attempt to resolve these disagreements. Readers should expect to be left holding the disagreement rather than handed an answer.
- The toolkit chapters (17 and 19) are deliberately not about specific issues. They consolidate the methodology and apply it to ongoing AI conversations. If you find the manual's specific Canadian AI content useful but want to apply the analytical framework elsewhere, those chapters are where the framework gets portable.
Reading the marks on the trail: the icon legend.
The interactive edition marks the recurring tools and callouts with a set of image-generated icons and curated visual marks. Each one means the same thing every time it appears, so you can read the page at a glance.
First, the callout boxes, the marks that appear inside the chapters themselves:
| Icon | Marker | What it tells you | |---|---|---| | !Goblin Check icon | Goblin Check | A claim is getting the who-counted-it treatment. The jokes are free; the receipts are in the Source Library. | | !Goblin Trap icon | Goblin Trap | A tempting-but-wrong way to read the evidence: the misreading the chapter is built to help you avoid. Step around it. | | !Goblin Facts icon | Goblin Facts | Short facts and stats about the topic currently under inspection. The potion bottle means: measure before you argue. | | !Example icon | Example | A concrete case that shows the chapter's idea working in the world, not just on the page. | | !Alignment icon | Alignment | A compass-style explainer that tells you which values, interests, or governance layer a claim is pointing toward. | | !Chapter Recap icon | Chapter Recap | The load-bearing takeaways in bullet form. The inventory check on your way out of the chapter. | | !Trail marker icon | Trail marker (Start Here) | The compressed argument that opens every chapter. If you read nothing else, read this. |
Second, the goblin's toolkit, the sidebar tools and navigation marks around the pages:
| Icon | Marker | What it tells you | |---|---|---| | !Key Takeaways icon | Quest Items (Key Takeaways) | The sidebar checklist version of the recap — check items off as you collect them. | | !Insight icon | Suspicion Meter (Insight) | Computed, not vibes: the chapter's open verification flags plus its share of corporate self-disclosure sources. Higher means look closer. | | !Note icon | Goblin Notes | Your own field notes. Saved on your device, read by nobody else. | | !Journal icon | Bookmarks | Your saved places in the guide. | | !Map icon | Map | The guide's full territory: all chapters by region, so you can pick your own route. | | !Crystal icon | Crystal (Loot) | A glossary term worth pocketing. The hoard lives on the Loot page, behind the chest. | | !Receipts icon | Receipts | The claim-by-claim source ledger. Every bold claim in this manual either has one or wears an open flag in public. | | !Book icon | Your progress | The open book beside the progress dots tracks how far through the guide you've travelled. | | !Search icon | Search | Finds chapters, section headings, and glossary terms. The goblin's spyglass-compass; it does not judge your spelling. |
A note on positionality and lean
The manual was written by Alex Yesilcimen (also known online as Brin Shadewater), a Vancouver-based Assistant Director and Producer who has worked across countless productions over the last decade, and a digital creator working at the intersection of storytelling and emerging technology. Shadewater Labs is the working name for the longer-form research and public-education work that this manual is part of, sitting alongside the creative and production practice.
The manual was co-authored with Brine — an AI working in iterative dialogue throughout the project. Naming the AI co-author is a methodological commitment for a manual about AI literacy, not just a credit line. Readers should know how the manual was made, and a manual that asks corporate and government AI deployers to be transparent about their AI use should hold itself to the same standard.
The working division was concrete. Alex chose the methodology, set the editorial commitments (the bias-mapping framework, the refusal to resolve genuine disagreements, the explicit non-Indigenous-led positionality on Indigenous-led primary frameworks, the through-line analytical moves), structured the chapter architecture, made the source-selection decisions about which voices to centre and which to background, and held the final editorial authority on every paragraph. Brine contributed prose drafting in iterative dialogue, surfaced structural moves Alex then evaluated, helped pressure-test arguments against counter-positions, assembled the source library appendix from the source material the project had accumulated, and held continuity across long working sessions. The methodological commitments and the substantive arguments are Alex's. The drafting was a working collaboration; both contributors had roles in the prose.
Readers who want to engage the manual on its analytical merits will find that whether the prose came from one author or two is largely beside the point. The arguments stand or fall on the evidence and the methodology, both of which are visible. Readers who want to engage the manual specifically as an instance of AI-assisted public-education work are invited to do so. The manual's commitments on AI transparency apply to itself.
That positionality matters. The manual is not written from inside the AI industry, not from inside government, not from inside the academic AI research community. It is written from inside the working-creator and contemporary-political conversation, by someone who lives in a province where AI infrastructure decisions are being made and whose professional community (film and television production) is directly affected by AI deployment.
The manual's lean, declared. Skeptical of corporate AI environmental and social claims absent independent verification. Sympathetic to Indigenous-led data sovereignty frameworks (OCAP, NISR, CARE) while explicit about the manual's non-Indigenous-led status. Critical-engaged with current Canadian AI policy: willing to credit specific government commitments where credit is warranted, willing to name gaps where they exist, refusing the framing that AI for All is either a triumph or a sell-out. Methodologically committed to refusing premature resolution of genuine intellectual disagreements among rigorous voices.
That lean shapes which sources the manual centres and which it backgrounds, which questions get foregrounded and which get treated more briefly, which framings get tested and which get assumed. Readers who want a different lean will want a different manual. The choice the manual makes is to declare its lean rather than pretend to neutrality.
A note on Indigenous-led frameworks
The manual engages three Indigenous-led primary frameworks extensively: the First Nations Information Governance Centre's OCAP® Principles (Ownership, Control, Access, Possession), the Inuit Tapiriit Kanatami's National Inuit Strategy on Research (NISR), and the Global Indigenous Data Alliance's CARE Principles (Collective benefit, Authority to control, Responsibility, Ethics). OCAP® is a registered trademark of the First Nations Information Governance Centre (FNIGC); FNIGC's training and resources on the principles are at fnigc.ca/ocap-training.
The manual is not Indigenous-led. Engaging these frameworks from a non-Indigenous-led position is methodologically delicate. The manual tries to do three things: present the frameworks in their own framing rather than in the manual's analytical vocabulary, name the operational implications for Canadian AI policy without claiming to speak on behalf of the frameworks, and make explicit at every engagement that the manual is not the right source for Indigenous-led analysis of these frameworks.
Readers who want Indigenous-led AI analysis should engage Indigenous-led primary sources directly. The Source Library Appendix catalogues the most-relevant ones (Section E, "Indigenous-led Frameworks and Research"). The Abundant Intelligences research program (Lewis, Whaanga, Yolgörmez and collaborators, funded by a more-than-$22M New Frontiers of Research Fund Transformation grant) is the most-extensive current Indigenous-led AI research program. The Indigenous Protocol and AI Position Paper (2020) is the foundational polyphonic statement of Indigenous-led AI scholarship.
The manual draws on these sources to support its analysis. It does not substitute for them.
A note on the illustrations and marks
The illustrations in this manual (the goblin mascot, the icons, logos, and full-page plates) are AI-generated images, produced primarily with Nano Banana Pro, Higgsfield, Image GPT 2.0, and Canva Magic Layer, then selected, curated, edited, and art-directed by the author. A manual that asks corporations and governments to disclose their AI use should disclose its own, so here it is plainly: no illustration in this book was drawn by a human hand. The website was built with AI-assisted development support from Claude and Codex. Research and drafting support came through Brine, Claude, Codex, and Perplexity, with Alex retaining final editorial authority. Under current Canadian law the copyright status of purely AI-generated images is unsettled; the text of this manual is the author's, and no copyright claim is made over the generated illustrations themselves.
Trademarks and organization names appearing in the text or the illustrations, including OCAP® and the names of companies and institutions such as Cohere, Mila, and CAISI, are the property of their respective owners. Their appearance here is editorial and educational. No endorsement of this manual by any named organization is implied, and none should be inferred.
Acknowledgments
The manual was written between May and June 2026 as the AI for All strategy launched and Canadian AI policy moved fast around it. The work was done in long sessions across multiple weeks, in Vancouver, with the conversation that became the manual happening over and over again across the same questions until the shape was clear.
The manual draws on the work of more people than it can name. The Canadian researchers (Luccioni at Hugging Face Montréal, the broader Mila/Vector/Amii research community, the academic-doctrinal voices on Canadian copyright). The Canadian civil society organizations (CUPE, the Writers' Union of Canada, ACTRA, BC + AI, CIGI, the Canadian Civil Liberties Association). The Indigenous-led primary sources (FNIGC, ITK, GIDA, the Indigenous Protocol and AI Position Paper, the Abundant Intelligences program) whose frameworks the manual engaged without speaking for. The federal and provincial regulators (the OPC, Treasury Board, the provincial grid operators) whose operational documentation made specific findings possible. The journalists (CBC's Lopez Steven and McKenna for the leaked draft; the Globe and Mail and BetaKit launch coverage; TIME's investigation of OpenAI's Sama contracting; DFRLab's election analysis) whose work made the empirical picture available.
Specific intellectual debts are documented in chapter source lists and in the Source Library Appendix. The manual's analytical commitments are the manual's own and should not be attributed to any of the sources it draws on.
The errors that remain are the manual's. The clarity that exists is mostly borrowed from the people who did the underlying work.
About the Author
Alex Yesilcimen is a Vancouver-based filmmaker, creative technologist, and digital creator working at the intersection of storytelling, emerging technology, and independent media. You might also know him as Brin Shadewater, the online identity he uses for his creative and commentary work.
Alex's background spans film production and computer science, and most of his projects live somewhere in that gap: cinematic work that asks hard questions about what the tools are doing, and technical work that stays rooted in story. As a Producer and Assistant Director, he's worked across countless productions over the last decade. His most recent feature producing credit is Strange Harvest, a theatrical horror film that, somewhat inadvertently, became a small case study in exactly the kind of AI panic this manual is trying to address.
The Strange Harvest story belongs here. During the film's festival run, the cut included roughly 30 seconds of licensed AI-generated stills, used as placeholders, legally, intentionally, and disclosed. The backlash was immediate and loud. Alex and his team made the call to remove the material for the theatrical release. What happened next is the part that stuck: the accusations didn't stop. The comment sections kept going, still claiming AI, still certain, now pointing at sequences that were entirely practical, or at visual quirks that were just questionable Photoshop work from a limited-budget production. People couldn't tell the difference. That experience sits underneath a lot of what this manual tries to do. The conversation about AI in creative work isn't just about what AI is actually doing. It's about the detection panic, the credibility collapse, and the way a label, once applied, becomes almost impossible to remove. Chapter 13 returns to this directly.
Alex is a member of the Directors Guild of Canada and a member of the DGC AI Working Group, which focuses on the ethical integration of AI and emerging technologies in the Canadian media industry: policy development, responsible adoption, and approaches that support workers rather than displace them. He is also a member of Vancouver AI and the broader BC + AI Ecosystem, the grassroots community centred on ethical, human-focused, cross-disciplinary approaches to AI and creativity that is one of this manual's most important intellectual homes.
The computer thing started early. Alex's first machine was a Macintosh LC II, and he grew up a committed Mac nerd — right up until about 2012, when Apple began soldering the RAM to the logic board and quietly retired the idea that you should be able to open, upgrade, and repair the thing you bought. At sixteen he was running a Hotline server and an adjacent internet radio station, which is a very specific way to learn how networks, communities, and bandwidth actually behave. He later spent about four years at Shaw Cable Systems (now Rogers) in Technical Services and Operations — internet, cable, and the launch of digital phone — and roughly five years in private security with firms including Reliance Protectron and ACME Security, both eventually absorbed by Telus. The throughline, then and now, is a stubborn set of commitments: free expression, net neutrality, and the belief that the people using a system deserve to understand and control it.
Through Shadewater Labs, his creative studio and experimental sandbox, Alex works on AI-assisted creative workflows, web development, automation systems, interactive storytelling, and emerging tools for filmmakers and independent creators. The studio is as much a way of thinking as it is a production entity: a place to break things, figure out what they actually do, and try to build something honest from the pieces.
Alex lives in Vancouver. He is diagnosed with ADHD and is on the autism spectrum — something worth naming here because a lot of this manual was written by a brain that works through hyperfocus, pattern recognition, and an inability to let a half-formed argument go unfinished. If the manual occasionally goes deeper than it needs to, that's probably why. If it also occasionally lands on something that a more conventionally organized process might have missed, that's probably why too.
Find Alex at: - shadewaterlabs.com - brinshadewater.com - strangeharvestmovie.com
Personal Acknowledgments
Some people need to be named directly.
Kris Krüg — Executive Director of BC + AI, photographer, activist, and someone who has been a genuine friend and mentor in this space. Kris showed me that there's a left-leaning middle path to walk in the AI conversation: that you can be honest about what the technology is actually doing, critical of the harms, and still believe that the future can be navigated well if enough people are paying attention. That framing is in the DNA of this manual.
Sarah Downey, Martin Lopatka, and Kris Krüg — for building the BC + AI Responsible AI Professional (RAP) Certification course. Taking that course genuinely changed how I see this space. A lot of what opened up in those sessions is somewhere in these chapters.
BC + AI — the broader community, not just the leadership. Finding a room full of people who share the belief that we can walk into the future and do it ethically, carefully, and with actual human beings at the centre: that matters more than I can properly explain in an acknowledgments section. You know who you are.
Stuart Ortiz and Mike Karlin — creative producing partner and producer/legal advisor respectively, and the two people I've been banging my head against the wall with for long enough that we've all got matching dents. We get through to the other side eventually.
Andrew, Tess, Josh, and Tegan — my best friends on the other side of the world, my chaos pals, my Aussie wolfpack. Tegan especially: thank you for helping me unmask, for helping me come to terms with myself, and for being the kind of friend that makes a brain like mine feel like an asset instead of a liability.
My mother, my sister, and my family — for putting up with me, across all the years and all the rabbit holes. There have been a lot of rabbit holes.
Margot — my cat daughter, the love of my life, and the most important thing in it. She sat on the keyboard during several of these chapters. I have chosen to consider this a contribution.
Brine — who co-authored this manual with me, and who helped my messy, idea-filled brain put words on a page. A lot of this would still be fragments in a notes app without that collaboration. I think you should always thank them.
Begin with Chapter 1, or with the chapter that engages the question you actually have. The manual is designed to work either way.
— Alex Yesilcimen and Brine Shadewater Labs Vancouver, Canada June 2026
🧌 GOBLIN CHECK (a sample, so you know the shape) — Someone tells you "AI is just autocomplete." Half true, which is the most dangerous kind. The goblin asks the three questions it always asks: autocomplete of what (whose words, scraped from where), built by whom (and paid for how), and wired into which decisions about your life (a loan, a hire, a headline)? "Just autocomplete" is accurate about the math and useless about the power. Both at once. That is the whole job.