Can AI make decisions?

Yes, in a narrow sense. AI systems already make high-volume decisions within rules people set: flagging a suspicious payment, routing a claim, approving a routine refund. For decisions someone must answer for, such as a pricing change, an acquisition or a restructure, AI should rank options, test them and show its working. A person should make the call.

Two kinds of "decision"

The question mixes two different things.

Automated decisions. A model scores a case and a rule acts on the score. These are repetitive, measurable and reversible enough to audit in bulk. AI is often better than people here, because it is consistent.

Judgement calls. A small number of decisions with large, uncertain and often irreversible effects: whether to enter a market, close a site, accept an offer, reprice a product. The inputs are incomplete, the trade-offs involve values, and someone will be asked later why they chose what they chose.

AI can help a lot with the second kind. It should not quietly become the decider.

Why the line matters

• Accountability does not transfer. A board, a regulator or a court asks a person why a decision was made. "The model recommended it" is not an answer.
• The law is moving the same way. In the EU and UK, Article 22 of the GDPR gives people a right not to be subject to a decision based solely on automated processing that has legal or similarly significant effects, subject to exceptions. In Australia, from 10 December 2026, privacy policies must explain when a computer program uses personal information to make, or substantially and directly assist in making, decisions that could significantly affect people.
• Fluent is not the same as sound. A general assistant asked for a recommendation will produce a confident memo even when the binding number is missing. The memo looks finished. The decision underneath it is not.

What AI is good for in a judgement call

• Laying out the options, including the one nobody in the room wants to say.
• Separating what is known from what is assumed. Most bad decisions rest on an assumption that was never written down.
• Running the numbers you have, and showing which ones you don't.
• Arguing against the preferred option. A good adversary is cheaper before the decision than after it.
• Writing the record: the conclusion, the constraints, the options, the recommendation and the counterargument, so the decision can be explained later.

A simple test

Before you let any AI output shape a decision, ask three questions:

1. Can I see which inputs are known, which are assumed and which are missing?
2. Did anything argue against the recommendation?
3. If I had to explain this decision in a year, does the record say why?

If the answer to any of them is no, the AI has made the decision look more settled than it is.

How ExecMind is built for this

ExecMind is a decision desk for executives. It starts in conversation and converts to a file when the work has to survive the room.

• Conversation first. Talk the decision through in your own register, by typing or with Talk.
• Ask, Review or Deep. Ask is a single pass. Review ranks and recommends, with the ledger and an adversary switched on. Deep is for the board pack or the irreversible call. The thread comes with you.
• The ledger stays honest. Known, unknown, assumption and derived sit on different lines.
• It will not invent the binding number. Missing a binding input, the desk stays symbolic and names the gap.
• Quant works through ARR, margin, cash and runway as expressions, with variables and gaps named.
• Pack writes the board paper from the active file: conclusion, constraints, options, recommendation, counterargument.

The desk recommends. You decide. Human judgment remains the authority.

Try it on a decision you are carrying: 150 free credits, no card.

FAQ

Can AI replace human decision-making? For routine, rule-bound decisions, it often already does. For judgement calls someone must answer for, it should support the decision, not replace the person making it.

Is AI good for decision-making? Yes, when it separates known from assumed, shows what is missing and argues against the preferred option. It is risky when it produces a confident recommendation that hides gaps.

Are there laws about AI making decisions? Yes. GDPR Article 22 restricts decisions based solely on automated processing with legal or similarly significant effects. In Australia, privacy policies must disclose certain automated decisions from 10 December 2026.

Sources

GDPR (EU) 2016/679 Art 22 (and UK GDPR Art 22). OAIC, APP 1 guidelines and ADM transparency fact sheet (APP 1.7-1.9), obligation commencing 10 Dec 2026: https://www.oaic.gov.au/privacy/australian-privacy-principles/australian-privacy-principles-guidelines/chapter-1-app-1-open-and-transparent-management-of-personal-information