When a computer decides something about you

Every other page on this site assumes you get to choose. This one doesn’t. Some of the AI that affects you most runs inside other people’s organisations, decides something about your money, your housing, your health care or your work, and offers you no setting to switch off. There is still something you can do, and it is more effective than most people expect.

Where this actually happens

This is not a future problem, and it isn’t exotic. Software already makes or heavily shapes decisions in places most people meet at some point:

Job applications, sorted or scored before a person reads one.
Credit and loan decisions, and the limits and rates attached to them.
Tenant screening, when you apply to rent.
Health insurance — prior authorisation for a treatment, or the downgrading or denial of a claim.
Public benefits and eligibility reviews.
Fraud systems that freeze an account or block a payment.

You usually can’t tell from the outside. The letter says you were unsuccessful, or the claim wasn’t covered, or the application didn’t meet the criteria. Nothing in it mentions software.

The one idea to remember

You do not need to prove a computer made the decision. You only need to ask, in writing, whether one was involved, and to request that a person review it. Those two sentences do most of the work, and they cost you nothing but a stamp or an email.

Why asking works better than arguing

Most of these decisions are wrong for a dull reason rather than a sinister one: the information that went in was wrong. A mixed-up file, an old address, a debt that was settled years ago, a name close to someone else’s, a diagnosis code entered incorrectly. Systems built to process thousands of cases don’t notice any of that. A person looking at your file usually does, and often within minutes.

So the useful request is not “your algorithm is biased.” It is “please tell me what information this decision was based on, and please have someone check it.” The first version invites a defence. The second invites a look.

The four sentences worth keeping

Put these in writing — email, the organisation’s own contact form, or a letter. Writing matters because it creates a date and a record, and because it tends to be routed to someone with more authority than the person answering the phone.

“Please confirm whether an automated system or algorithm was used in making this decision.” You are entitled to ask. In a growing number of places they are required to tell you, and in the rest a refusal to answer is itself informative.
“Please tell me what specific information this decision was based on.” This is the sentence that surfaces the error. Ask for the report, the score, the file, or the reasons — whatever they call it.
“I am requesting review and reconsideration by a person.” Use the word person, or human. Say that you want someone with the authority to change the outcome, not a re-run of the same process.
“Please respond in writing, and tell me the deadline for appealing.” Deadlines are often short and are rarely volunteered. Asking starts a clock you can actually see.

Keep a copy of everything, with dates. If you later need to escalate, the record of a polite, specific, ignored request is worth more than any argument you could make about the technology.

Three situations with a route already built

Some of these decisions come with an appeal process that exists whether or not software was involved. You don’t have to mention AI at all to use them.

Anything decided using a report about you — credit, tenant screening, insurance, or an employment background check. In the United States, when a company turns you down based on such a report, they generally have to tell you so, tell you who supplied it, and let you get a copy. You can then dispute anything inaccurate with the company that produced the report, and they have to investigate. This is old, well-established consumer protection, and it works on exactly the bad input data that causes most automated mistakes.
Health insurance denials. There is normally an internal appeal, and after that an independent external review by someone who does not work for the insurer. A surprising number of denials are reversed, and a large share are never appealed at all. Ask for the specific clinical reason in writing and give it to your doctor — a short letter from them is often the whole appeal.
Public benefits. Eligibility decisions almost always carry a formal right to appeal, usually with a strict and short deadline printed somewhere in the letter. Find the deadline first, before you do anything else.

Where to go when the organisation won’t engage

Your state attorney general’s office and your state consumer protection office both take complaints, and both keep track of patterns. For financial products there are federal complaint routes; for insurance, every state has a department of insurance that handles them. A complaint rarely produces an instant fix, but it frequently produces a phone call from someone senior, which is usually what was needed.

What the law says, and why this page doesn’t list it

Rules about automated decisions are arriving quickly and unevenly. Some states require notice when an automated tool is used in hiring. Some create a right to request human review of consequential decisions. Some cover insurance, or health care, or lending specifically. Others have nothing yet. Outside the United States the picture is different again — European and British rules have contained a right not to be subject to purely automated decisions with significant effects for years.

Any list of specifics here would be out of date within a year, and one major state law has already been written, repealed and rewritten while this site has existed. So this page deliberately teaches the request rather than the statute, in the same spirit as the chapter on turning off unwanted features. If you want the current position where you live, your state attorney general’s website is the place to look, and it is worth ten minutes before you write.

What to expect, honestly

You will sometimes be told that no automated system was used, and that may well be true. You will sometimes get no answer at all. Human review does not mean a sympathetic review, and the second answer is often the same as the first.

But the request is cheap, it is frequently granted, and a startling proportion of these decisions come apart the moment somebody actually reads the file. The reason to do it is not that the system is malicious. It is that nobody has looked, and asking is how looking starts.

Why this belongs on a site about using AI

Everything else here rests on the idea that you decide what runs and what sees your information. This is the category where that isn’t true, and leaving it out would have made the rest of the site quietly dishonest. You can’t switch these systems off. You can make them show their work.