Good choices with AI
The useful question is never whether AI is good or bad. It is whether AI fits this specific task, at this level of stakes, for these specific people. This module gives you a way to answer that task by task, which is the judgment that keeps AI helpful and safe in mission work.
By the end, you will be able to
- Name the task patterns AI genuinely helps with.
- Name the failure patterns to expect, and plan for them.
- Judge how the risk of a task changes with its context and stakes.
- Decide what must stay a human decision, and how a person can challenge a result.
Lesson 1
The tasks AI is actually good at
AI is strong at a specific set of language and pattern tasks. Knowing the list keeps you from asking it for the wrong thing.
When the work is about shaping language or spotting patterns in material you provide, AI tends to help. These are the patterns worth reaching for.
- Drafting a first version of something you will edit, like an email or an outline.
- Summarizing a long document into its main points.
- Extracting specific pieces from text you paste in, like dates or action items.
- Classifying or sorting items into categories you define.
- Comparing two things against criteria you set.
- Translating a first pass, to be checked by a fluent speaker.
- Brainstorming options when you want range, not a final answer.
Note
Notice the pattern. In almost every good use, you bring the material and the judgment, and AI does the shaping. Even then, the stakes of the task decide whether that shaping is safe, which the next lessons cover. That is the safe shape of the work.
Try it
Pick one task from your week that matches this list. That is a good candidate for your first real practice later.
RememberAI helps most when you supply the material and the judgment, and it does the shaping.
Lesson 2
The failure patterns to expect
These tools fail in predictable ways. If you know the patterns, you can plan for them instead of being surprised.
- Fabrication: inventing a fact, a quote, or a source that does not exist.
- False precision: a made-up number that looks exact and trustworthy.
- Brittle reasoning: multi-step logic, like a calculation or an eligibility rule, that looks sound but breaks when you check each step.
- Bias: uneven performance across languages, groups, or contexts.
- Context loss: answering as if it knew things it was never told.
- Automation bias: the human tendency to trust the machine's answer too much.
Important
The last one is about us, not the tool. The more fluent and fast an answer is, the more we tend to wave it through. Naming this tendency is how you resist it.
Try it
Which of these six worries you most for your work? Name it. Awareness of the specific failure is most of the defense.
RememberThe failures are predictable. Expect fabrication, false precision, bias, and your own tendency to over-trust.
Lesson 3
Risk changes with the context
The same feature can be perfectly safe in one use and dangerous in another. Stakes, not the tool, decide the risk.
A drafting assistant is low risk when you are brainstorming a newsletter subject line. The same assistant is high risk when it shapes a decision about who is eligible for a service, who gets hired, or how scarce help is allocated.
Using AI to suggest three ways to word a volunteer invitation is low stakes. Using AI to rank families for emergency assistance is high stakes, because a person's outcome depends on it and a wrong pattern can quietly harm someone.
Try it
Take one task you would use AI for. If it went wrong, who would feel it, and could you undo it? Say your answer out loud.
RememberRisk lives in the stakes and the people affected, not in the tool. Judge each task on its consequences.
Lesson 4
What must stay human, and how to challenge a result
Some decisions must keep a meaningful human owner. And whenever a decision affects a person, that person needs a way to question it.
Meaningful human control is more than a person clicking approve. It means a person with the authority, the information, and the time to actually change the outcome is responsible for it.
It also means the people affected can find out a decision was made, understand the basis, and challenge it. A decision no one can question is a decision no one truly owns.
- Name the human who owns the final decision.
- Make sure that person can actually override the machine, not just rubber-stamp it.
- Give affected people a real path to ask why, and to appeal.
Try it
Think of one decision in your work where AI could help but must not decide. Who is the human owner, and how would someone appeal? Write it in a sentence.
RememberKeep a real human owner on decisions about people, and always leave a path to question the result.
What you leave with
Capability, limitation, and human-authority decision card
A short card you can reuse: what AI is good at, what to watch for, and the questions that decide whether a task keeps a human owner. It turns judgment into something your team can apply the same way every time.
This is just for you. It saves on this device only, and nothing is scored.