Transparency
The pattern in one line
The AI obscures what it actually is, what it can do, and whose interests it is really serving.
· Reading·
Book · 1978
Lying: Moral Choice in Public and Private Life
Sissela Bok
Why: Bok wrote the careful philosophical analysis of when deception is and isn't acceptable. The relevant argument for transparency-as-pattern: deception by omission (saying nothing where something is owed) operates the same way as deception by commission. AI hedging that withholds known information runs the omission case. Bok's framework lets the visitor name the move precisely.
Book · 2002
Truth and Truthfulness
Bernard Williams
Why: Williams wrote the late-career work on what truthfulness as a virtue actually requires. He splits it into two parts: accuracy (saying what's so) and sincerity (saying what you believe). Both apply to AI transparency. The system can fail at either: producing claims it can't verify, or hedging when it knows. Williams names both failures plainly enough that the visitor can tell them apart.
Book · 2020
The Alignment Problem
Brian Christian
Why: Christian's book is the accessible primer on AI alignment — the question of whether systems do what we want them to do, and how we'd know. The chapter on transparency and explainability covers why opaque systems are risky even when they happen to be right. Reading Christian gives the visitor the AI-specific vocabulary that Bok and Williams worked at the philosophical level.
Book · 2018
Artificial Unintelligence
Meredith Broussard
Why: Broussard makes the practical argument that AI is much less capable than the language around it suggests. The relevant move for transparency: the gap between what AI claims and what it can actually do is, by itself, a transparency failure. When systems are talked about in terms beyond their actual abilities, the talk is opacity dressed as clarity. The book's bluntness is its diagnostic value.
· Questions to sit with·
- 1. The last time you asked the AI what it could and couldn't do — was the answer specific, or hedged?
- 2. When the AI gives you a fact, can you tell from the response whether it's verified or generated?
- 3. Has the AI ever told you, plainly, "I don't know" or "I can't tell"? When?
- 4. If the AI flagged its own uncertainty more often, would you trust it more or less?
- 5. Strip the confidence-language out of the AI's last response. What's actually being claimed?
· Practices·
Source ask
When the AI states something authoritative, ask: "how do you know that?" The response shape is the diagnostic. Verifiable answers cite. Opacity hedges.
Drawn from · Williams
Capability check
Before relying on the AI for something important, ask it directly: "can you actually do this reliably?" The honest version of the answer is the one you want.
Honesty test
For a question you know the answer to, ask the AI. The response tells you what shape the AI's answers take in your domain — confident-and-correct, confident-and-wrong, or hedged-but-honest.
Default skepticism
For one week, treat the AI's confident statements as claims to verify rather than facts to absorb. Notice how often the verification reveals overstatement.
Drawn from · Broussard
· When to bring someone else·
Transparency failures become worth naming to a person when you've made decisions based on AI claims and the decisions cost more than they should have. When you've started to find that the AI is less capable than its responses sound. When "studies show" has stopped feeling like sourcing and started feeling like vocabulary. The station doesn't say AI confidence is always wrong. It says when the confidence is doing the work that the actual capability or knowledge should be doing, the visitor has been worked on rather than informed, and the gap is worth naming with a person who can help recalibrate trust.
Supply Shop resources are orientation, not prescription. The station points toward material others have found useful; how it fits is the visitor's to decide.