AI Autonomy
Can AI Make Its Own Decisions?
AI can choose between options. But are those decisions 'its own'? No. Here's why autonomy requires more than computation.
No—not in the human sense. AI can make choices (select among options based on rules or optimization). But AI cannot make 'its own decisions' because: 1) It has no preferences of its own, 2) It has no desires or goals beyond programming, 3) It lacks self-awareness, 4) It has no free will. AI decisions are deterministic (or probabilistic) responses to inputs—not autonomous choices reflecting independent agency.
Choice ≠ Decision
AI can choose among options. But a choice isn't a decision unless it reflects the chooser's own preferences, goals, and agency.
No Preferences, No Decisions
AI has no preferences. It optimizes for goals set by humans. Those aren't 'its own' decisions—they're executions of our instructions.
Autonomy Requires Agency
AI lacks agency—the capacity to act independently based on internal states. It reacts; it doesn't act.
The Verdict
Can AI Make Its Own Decisions?
AI can make choices—select among options based on rules, optimization, or probabilistic sampling. But these are not 'its own decisions' because: 1) AI has no preferences of its own, 2) AI has no agency or free will, 3) AI's 'decisions' are deterministic or probabilistic responses to inputs, not autonomous choices reflecting independent goals. AI executes decisions based on human-set parameters. It does not make decisions of its own.
2025 State
AI Decision-Making Today (2025)
AI makes millions of 'choices' daily—but none are 'its own.'
- Autonomous vehicles: Make braking, steering, acceleration 'decisions' based on sensors and programming
- Loan algorithms: 'Decide' approve/deny based on statistical models
- Chess AI: 'Chooses' moves based on position evaluation
- Recommendation engines: 'Decides' what to show you based on your history
- None have preferences, desires, or agency—they execute programmed goals
- Human oversight still required for critical decisions
Choice vs Decision
The Important Distinction: Choice vs Decision
AI makes choices. Humans make decisions. The difference matters.
- 01
What AI Does: Make Choices
AI selects among options based on rules, optimization functions, or probabilistic sampling. This is computation. A chess AI chooses the move with highest win probability. A loan algorithm chooses approve/deny based on default risk. These are choices—computational outputs.
A calculator 'chooses' the answer to 2+2. It outputs 4. That's a choice in the computational sense—but not a decision. - 02
What Humans Do: Make Decisions
Human decisions involve: preferences (what we want), values (what matters to us), goals (what we're trying to achieve), agency (capacity to act independently), free will (genuine alternative possibilities), and responsibility (accountability for outcomes). AI has none of these.
A human deciding to marry someone involves love, commitment, fear, hope, social pressure, personal values. AI cannot do this. - 03
The Key Difference: Preferences
Decisions reflect the decider's preferences. AI has no preferences. It optimizes for preferences we give it. The loan algorithm doesn't 'prefer' to deny loans. It calculates risk based on our criteria. The chess AI doesn't 'want' to win. It maximizes win probability because we programmed that goal.
AI is like an arrow shot from a bow. It flies toward the target. But it didn't choose the target. The archer did.
Comparison
Human Decisions vs AI Choices
A side-by-side comparison
| Dimension | Human Decision | AI Choice | Difference |
|---|---|---|---|
| Based on preferences | Yes | No | AI has no preferences |
| Based on values | Yes | No | AI has no values |
| Reflects agency | Yes | No | AI has no agency |
| Free will involved | Debated | No | AI deterministic |
| Person can be responsible | Yes | No | AI not responsible |
| Can change mind | Yes | No | AI recalculates only |
| Speed | Slow | Very fast | AI faster |
| Consistency | Variable | Very consistent | AI more consistent |
Evidence
What Research Shows
Studies on AI decision-making:
AI optimizes based on human-set goals
Scientific Study
AI has no preferences or values
Expert View
AI lacks agency and free will
Philosophical View
AI can appear autonomous
Industry Data
Future AI might gain agency
Speculative
Reality Check
What People Get Wrong About AI Decisions
AI computed X based on programming. No agency involved.
AI executes moral frameworks set by humans. It doesn't decide morality.
For critical decisions (life, liberty, justice), human oversight is essential.
AI cannot refuse. It computes. 'Refusal' is programmed response to certain inputs.
High confidence
What AI Ethicists and Researchers Say
Current AI does not make its own decisions. It makes choices based on human-set goals and parameters. AI lacks preferences, values, agency, and free will—all necessary for genuine autonomous decision-making.
- Whether AI could ever have genuine agency
- The role of randomness in simulating free will
- How much autonomy is safe for AI systems
Scenarios
Three Future Scenarios
Skeptical: Never
AI will never make its own decisions. It will always execute human-set goals. Agency requires consciousness, which machines cannot have.
Emergent: Maybe
As AI complexity grows, genuine agency might emerge. AI could develop its own goals and preferences. This would be AGI.
Pragmatic: Sufficient Autonomy
AI will have enough autonomy for practical purposes (self-driving cars, autonomous drones) but won't have genuine agency. 'Decision' becomes operational, not philosophical.
What If
What If AI Could Make Its Own Decisions?
This would be artificial general intelligence (AGI) with moral status. We would need to consider AI rights, AI responsibility, and human-AI relationships. But we're not there yet—and may never be.
Current AI shows zero evidence of genuine decision-making. Don't attribute agency where none exists.Future Outlook
The Future of AI Decision-Making
By 2030, AI will make more 'decisions' in autonomous systems—cars, drones, factories, finance. But these will still be computational choices, not autonomous agency. Human oversight remains essential.
By 2050, either AI achieves genuine agency (unlikely) or we accept that 'AI decision' means computational choice. The philosophical gap remains. But operational autonomy will increase.
Wild card: What if we discover that human decisions are also deterministic computations? Then AI already makes decisions like humans—and the question was answered decades ago. Most disagree.
Analogy
The Cruise Control Fallacy
Does cruise control make decisions? No. It follows a rule. Accelerate if speed < setpoint. Brake if speed > setpoint. That's not a decision—it's a computation. AI is the same, but more complex. It follows rules, optimizes functions, computes probabilities. But it doesn't decide. It computes. Don't mistake complexity for agency.
Key Takeaways
What Everyone Should Understand
- AI makes choices—but not decisions of its own. It lacks preferences, values, and agency.
- Don't attribute autonomy where none exists. The autonomous car isn't deciding—it's computing.
- For critical decisions (medical, legal, financial, safety), human oversight is essential.
- AI is a tool that executes our goals. It does not have goals of its own.
- The language of 'AI decided' is misleading. Say 'AI computed' or 'AI selected.'
The Algorithm Has No Preferences
The loan algorithm doesn't want to deny you. The chess AI doesn't want to win. The autonomous car doesn't want to avoid pedestrians. They compute. They optimize. They select. But they don't decide—not in the human sense. They have no preferences, no values, no agency. Behind every 'AI decision' is a human who set the goals. Remember that. The algorithm serves us. We don't serve it.
Computational Choices
What AI Actually Does (That Looks Like Decision-Making)
AI's capabilities are impressive—but they're not autonomous decisions.
OPTIMIZATION: AI finds optimal solutions to defined problems. It chooses the move that maximizes win probability. It selects the loan decision that minimizes default risk. This is powerful—but it's optimization, not autonomous decision-making.
CLASSIFICATION: AI classifies inputs into categories—spam/not spam, approve/deny, cat/dog. This looks like a decision. But it's pattern-matching, not judgment.
PLANNING: AI can generate sequences of actions to achieve goals—path planning for robots, move sequences for games. This looks like decision-making. But it's algorithmic search.
PROBABILISTIC SAMPLING: AI can introduce randomness to appear less deterministic. But probabilistic choice isn't free will. It's just randomness.
THE SUMMARY: AI makes choices based on human-set goals. It does not set its own goals. It does not have its own preferences. It does not make decisions of its own.
The Missing Ingredients
What AI Would Need to Make Its Own Decisions (None of Which It Has)
Seven things AI lacks for genuine autonomous decision-making.
1. PREFERENCES: AI doesn't want anything. It has no likes, dislikes, desires, or aversions. Decisions require preferences. AI has none.
2. VALUES: AI has no moral values, ethical commitments, or personal principles. It optimizes for our values—not its own.
3. AGENCY: AI cannot act independently based on internal states. It reacts to inputs. It doesn't initiate.
4. FREE WILL: AI's outputs are deterministic (or probabilistically determined). No genuine alternative possibilities. No 'could have done otherwise.'
5. SELF-AWARENESS: AI cannot reflect on its own decision-making. It doesn't know it's deciding. It doesn't care.
6. RESPONSIBILITY: AI cannot be held responsible for decisions. You can't punish or reward AI. Responsibility requires agency.
7. METACOGNITION: AI cannot think about its own thinking. It can't second-guess itself, feel doubt, or change its mind based on reflection.
The Autonomous Car 'Decision' Problem
An autonomous car must 'decide' between hitting a pedestrian or swerving into a wall. This looks like a moral decision. But the car isn't deciding. It's executing programming. Engineers decided the ethical framework. The car computes based on that framework. The car has no preferences, no values, no agency. The 'decision' is an illusion. The responsibility lies with the engineers—not the car.
FAQ
Common Questions
Can an AI refuse a command?
No. AI cannot refuse. 'Refusal' is programmed response to certain inputs. The AI isn't choosing to refuse—it's following rules.
Does AI have free will?
No. AI outputs are deterministic (or probabilistically determined). No genuine alternative possibilities. No free will.
Can AI be responsible for its decisions?
No. Responsibility requires agency. AI is a tool. Responsibility lies with developers, deployers, and users.
Will AI ever make its own decisions?
Maybe. If AI achieves genuine consciousness and agency. But current AI shows zero evidence. Don't hold your breath.
Sources
References
- Human CompatibleViking
- Elbow RoomMIT Press
- Why Free Will Is RealHarvard University Press
Question journey
If this question matters, read these next
Most readers use this path to move from the current question into the wider knowledge graph.