BAI · Machines and authority to act
The more a machine can do, the more visible human responsibility must become.As a machine gains authority, human responsibility must become clearer.
A system may answer a question, call another tool, send a message, change a file or act in the physical world. These capabilities do not transfer intention, judgement or responsibility to the machine. BAI places each machine within a real need, defined access, limited authority, human approval and a clear way to stop it.
The question
Today’s machines do more than answer questions; they increasingly ask for authority to act.
A conversational model may only offer a draft. An agent may call other tools and act on a calendar, email account or file. A device or robot may carry that effect into the physical world. The point is neither to treat the machine as a person nor to fear every machine. The point is to keep clear who gave permission, who is watching and who remains accountable for the outcome.
Automation can hand over a task; it cannot hand over responsibility.
A basic distinction
A machine carries capability; a person carries intention, judgement and responsibility.
A system may look powerful without becoming a human subject or a source of truth.
The machine
Acts through the capabilities it is given
- Processes data and finds patterns
- Produces predictions, drafts or options
- Calls tools it is permitted to use
- Acts within a defined environment
The person
Carries the purpose and the account
- Defines the real need and intention
- Considers sources, context and people affected
- Decides authority, acceptance, refusal and release
- Carries responsibility for harm, correction and outcomes
Four levels of authority to act
Look beyond the name and ask what the system can change in the world.
One product may work at several levels. The purpose is not to memorise new technical labels, but to make human approval more visible as authority increases.
01 Model: produces an answer or prediction
It may create text, images, sound, code, classifications or predictions. An output may look right, but it does not become a judgement without sources and human review.
02 Assistant: interacts with people and recommends
It may explain, remind, draft or offer options. It does not replace a teacher, parent, doctor, lawyer, friend or the person responsible for a decision.
03 Agent and automation: takes a step
It may plan, call tools, send messages or change records. Define access, human approval at consequential steps and a way to reverse actions before use.
04 Device and robot: acts in the physical world
It may sense, measure, carry or act in a space. The safety of bodies, children, homes, workplaces and the environment cannot depend on software accuracy alone.
As authority increases
More power to act requires clearer permission, narrower access and stronger ways back.
A tool that only drafts cannot be supervised like a system that sends money, deletes a document, opens a door or speaks with a child. Risk comes not only from technical capability, but from the setting, whose name it acts in, which data it uses and whether its actions can be reversed.
Human oversight is not a symbolic approval button. It must be a real responsibility held by someone who understands, can stop the action and carries the account.
Before giving a machine authority to act
Answer the same six questions clearly.
The answers are not only for installation day. Reassess authority whenever the system, data, users, purpose or area of impact changes.
What can it access?Make access to files, accounts, cameras, microphones, location, messages, payments, institutional records or children’s data visible—and allow only what is needed. What can it change?Distinguish reading from writing, drafting from publishing, recommending from purchasing and software output from physical action. Which step requires human approval?Money, health, law, children, privacy, representation and hard-to-reverse actions are not left to automatic acceptance. What happens when it is wrong?Test possible harm, reaching the wrong person, fabricated output, discrimination and physical safety risks with real examples. How can it be stopped and reversed?Set up emergency stop, removal of authority, rollback, data deletion, human takeover and safe shutdown before use. Who owns the outcome?Name the person responsible for the decision and the error within the institution, team or use setting. “The machine did it” must not make accountability disappear.
In everyday settings
The same machine should be used differently according to the people and limits of its setting.
Children and learning
A tool may support curiosity and making, but it does not become a child’s teacher, friend, religious guide or invisible observer. Age, household permission, privacy and adult guidance remain protected.
Professional and high-risk work
Professional competence remains essential in health, law, finance, education and public services. A machine may draft or assist; judgement and professional accountability remain with qualified people.
Households and privacy
People should understand what a device hears, sees, stores and sends outside the home. Convenience does not make family privacy an invisible cost.
Fieldwork and the physical world
Systems used in farming, production, transport or care do not reduce people, bodies, soil and living beings to measurable data. Field practitioners and affected people remain part of the decision.
The limit between truth and representation
A machine’s output or action is not truth itself, nor is it human judgement.
A model produces a representation through a particular arrangement of data and design; an agent takes steps through that representation; a robot may carry it into physical action. A successful result does not mean the system encompasses the nature of a person or the created world. Sources, purpose, uncertainty, human intervention and the limits of representation should remain visible, together with error reporting, removal of authority and responsibility for ongoing care.
Let us recognise what a machine does well while keeping its limits, fallibility and means of stopping visible.
Building the future together as a MİLLET
Technical teams alone should not decide the place of machines in our lives.
Understanding a system’s real need, nature and consequences requires people across languages, generations and professions to share a voice and responsibility.
Professional competence
Doctors, teachers, lawyers, farmers, artists and other field practitioners help define the problem and the limits of use.
The voice of people affected
Children, households, workers and people receiving a service are not merely data sources or test users; they are participants in the decision.
Languages and places
An assumption that works in one language or country does not become a measure for everyone. Native-language explanation and local experience remain visible.
Builder and SEDD responsibility
The team creating the system defines access, authority, human approval, error and responsibility for care from the beginning. SEDD is not an approval stamp added at the end.
When trying a system
Begin with a small, limited and reversible step.
Not every model, agent, device or robot needs to become a shared programme or product. Sometimes limited use is right; sometimes more human review is needed; sometimes the right decision is not to use it.
Responsible use of machines
Would you like to examine the authority of a model, agent, device or automation you use?
Shared reviews of machines and authority to act within BAI are still in preparation. You can share the real need, what the system can do and a benefit or risk you have encountered. This is not a promise of an active product or programme.