Authority stays human
The system can propose anything and commit only what a person allows.
Guide
What Is Human-Centered AI?
A term used often and defined rarely. Here is a working definition, and the tests that separate it from marketing.
Human-centered AI describes systems designed around human judgement, consent and capability rather than around automation for its own sake. In practice it means four things: the person keeps authority over consequential decisions, the system is honest about what it knows and does not know, memory exists by permission, and actions can be inspected and reversed. Guardian Daily is built to that pattern — it suggests and prepares, and a person approves.
Opens the separate Guardian Daily app. It may require its own sign-in or invitation. Paid signups are not open yet.
The definition
Four properties that have to be present together
Honest uncertainty
When confidence is low it says so, rather than producing fluent guesses.
Permissioned memory
What it knows is visible, correctable and removable by the person it is about.
Reversibility
Actions leave a record, and mistakes have a path back.
Why it matters
The failure mode is confident wrongness, not slowness
- An automated action taken on a misunderstanding costs more than the minutes it saved.
- People accept help far more readily when refusing it is easy.
- Systems that quietly widen their own authority erode trust faster than they build capability.
- Explaining why a suggestion was made lets a person correct the reasoning, not just the output.
How to test it
Questions worth asking any AI product
- What can it do without asking me, and can I narrow that?
- Can I read what it remembers about me in plain language?
- Does it tell me why it suggested something?
- What happens to my data if I stop using it?
- Is there anything it refuses to decide on my behalf?
Where the term is misused
Three patterns to watch for
Consent theatre
A single broad permission at signup presented as ongoing control.
Explanations after the fact
A rationale generated to justify an action already taken.
Opt-out buried
Controls that exist technically but are unreachable in normal use.
Questions
Common questions
- What is human-centered AI, in one sentence?
- Software that uses AI to expand what a person can do while leaving the consequential decisions, and the ability to say no, with that person.
- Is human-centered AI the same as human-in-the-loop?
- Human-in-the-loop is one mechanism — a person reviews or approves an action. Human-centered is the broader stance that includes memory control, honest uncertainty, reversibility and the right to leave.
- How can I tell if a product only says it is human-centered?
- Look for defaults, not adjectives. Does it act without asking? Can you see and delete what it remembers? Does it tell you when it is unsure? Marketing language is cheap; defaults are expensive.
- Does keeping humans in control make AI less useful?
- It makes it slower in narrow cases and more trustworthy in broad ones. Most real-world tasks fail on wrong confident actions, not on a missing second of latency.
Keep reading
Related guides
More plain-language explanations of how Guardian Daily is designed.
AI that asks before it acts
Approval boundaries in practice.
Privacy-first personal AI
Memory you can inspect and remove.
What is human flourishing?
The standard this design stance serves.
Assistance, not substitution.
Guardian is designed so that capability grows and authority stays where it belongs.