If the universe is deterministic, every event caused by prior events in an unbroken chain back to the Big Bang, how can anyone be morally responsible for anything?
My essay On Moral Responsibility tackles this and proposes a specific answer: we can be morally responsible even in a fully deterministic universe. Free will in the libertarian sense isn’t necessary for moral responsibility. What matters is whether our actions flow from our values, not whether those values were causally determined.
This question gets urgent when you think about AI. If SIGMA’s actions are fully determined by its training and architecture, is it responsible for misaligned behavior? Or are the developers responsible? The compatibilist framework has something useful to say here.
The Problem
The Deterministic Picture
Classical physics: the universe operates according to deterministic laws. Given the complete state of the universe at time T0 and the laws of physics, you can (in principle) predict the complete state at any future time T1.
Laplace’s demon: an intellect that knew all positions and velocities of all particles could predict the entire future and retrodict the entire past.
Implication: your “choice” to read this sentence was determined 13.8 billion years ago at the Big Bang. Every neuron firing, every thought, every decision, all causally determined by prior states.
The Threat to Moral Responsibility
The traditional assumption: moral responsibility requires libertarian free will, the ability to have done otherwise.
The argument:
- You’re responsible for action A only if you could have done otherwise
- If determinism is true, you couldn’t have done otherwise (the causal chain was inevitable)
- Therefore: if determinism is true, you’re not responsible for your actions
In a deterministic universe, praise and blame, punishment and reward, moral responsibility itself, all illusions.
Three Responses
1. Libertarian Free Will
The claim: we have contra-causal free will, the ability to act independently of prior causal chains.
How? Either agent causation (persons are unmoved movers who initiate causal chains) or quantum indeterminacy (brain processes exploit quantum randomness to break determinism).
This preserves intuitions about moral responsibility and matches phenomenology. It feels like we choose freely.
Problems: neuroscience finds no evidence of contra-causal agency. Quantum randomness doesn’t help, because randomness isn’t freedom (random choices aren’t more free than determined ones). And there’s the luck objection: if choices aren’t caused by your character and values, they’re random. Random is not free.
2. Hard Determinism
The claim: determinism is true, therefore free will is impossible, therefore moral responsibility is impossible. No genuine praise or blame. We may need to restrain dangerous people (like quarantining disease vectors), but they’re not morally responsible.
This is logically coherent and matches physics. But it eliminates ethics. If no one is responsible, ethics becomes impossible. It also undermines itself: if beliefs are determined, why trust the belief in hard determinism? And practically, we can’t live as if no one is responsible.
3. Compatibilism
The claim: free will is compatible with determinism. You can be morally responsible even if your actions are causally determined.
The key insight: “free will” doesn’t mean “uncaused.” It means something else.
The essay defends this third option.
The Compatibilist Response
Redefining Free Will
The traditional definition: you act freely when you could have done otherwise (in an absolute sense).
The compatibilist definition: you act freely when your action flows from your character, values, and deliberation, even if those were causally determined.
The distinction:
- Freedom FROM causation (libertarian free will): impossible in a deterministic universe
- Freedom THROUGH causation (compatibilist free will): perfectly compatible with determinism
What “Could Have Done Otherwise” Really Means
Libertarian interpretation: given the exact same state of the universe, you could have chosen differently. This is indeed impossible in a deterministic universe.
Compatibilist interpretation: if circumstances had been different, you would have chosen differently. This is perfectly compatible with determinism.
Example: “I could have stayed home instead of going to work.”
This doesn’t mean: with the exact same beliefs, desires, and circumstances, I might have stayed home (libertarian). It means: if I had wanted to stay home, or if work had been canceled, I would have stayed home (compatibilist).
The difference: counterfactual conditionals, not absolute alternate possibilities.
Moral Responsibility Without Libertarian Free Will
The essay argues we can be responsible even if determinism is true.
What Grounds Responsibility?
Not the absolute ability to have done otherwise. Rather, whether actions flow from your values and character.
The key questions:
- Did the action express your values? Or was it forced, coerced, compelled by external factors?
- Did you deliberate? Or was it purely reflex, instinct, accident?
- Could you have responded to reasons? If given good reasons, would you have acted differently?
If yes to these, you’re responsible, even if the whole causal chain was determined.
Examples
Case 1: Bank robber. Wants money. Deliberates about methods. Chooses robbery. Acts on choice. The action flowed from values (greed) and deliberation. Responsible, even if those values were caused by genetics, upbringing, circumstances.
Case 2: Brain tumor. Tumor causes aggressive impulses. Person can’t control them. Acts violently. The action didn’t flow from values. Caused by tumor, not character. Not responsible (diminished capacity).
The difference: not whether actions were caused, but what kind of causes were operative. Internal causes (values, character, deliberation) ground responsibility. External or alien causes (coercion, disease, manipulation) undermine it.
The Deliberation Argument
Practical reasoning works regardless of determinism.
The deliberative stance: I’m trying to decide whether to do A or B. I consider reasons for each. I choose based on those reasons. My choice causally influences what happens.
This process is effective whether or not it was determined that I’d go through it.
Analogy: a calculator’s computation is deterministic, but it still arrives at the correct answer because of its computational process. Similarly, my deliberation is deterministic, but I still make good choices because of my reasoning process.
The causal efficacy of deliberation: thinking things through actually changes outcomes, even in a deterministic universe.
The Block Universe
The essay discusses four-dimensionalism (the “block universe” view) and its implications.
What Is Four-Dimensionalism?
The universe is a four-dimensional block: three spatial dimensions plus time. Past, present, and future are all equally real. The distinction between them is perspectival (depends on where you are in the block), not metaphysical.
Analogy with space: you’re “here,” other places are “there,” but both are real. “Here” is just your perspective. Same with time: you’re “now,” other times are “then,” but both are real. “Now” is just your temporal perspective.
Implications
No genuine change. No becoming. Just a timeless four-dimensional structure. You are not a three-dimensional object that persists through time, but a four-dimensional “worm” with temporal extent.
Does This Eliminate Agency?
The worry: if the future already exists, then your “choices” are already determined. You’re not creating the future. You’re enacting what’s already there in the block.
The compatibilist response: this doesn’t threaten agency any more than regular determinism does.
Your choices are part of the causal structure. Your deliberation at time T1 causes your action at T2, even if both are fixed in the block. Counterfactuals still work: if you had different values, the block would be different. The block contains your agency. Your deliberations, your reasoning, your value-based choices. These are real causal processes.
Analogy: a recorded movie. Everything is “already there” on the film. But the characters’ choices still matter within the story. The ending was caused by the characters’ decisions, even though it was “already” recorded. The block universe works the same way.
Implications for AI Responsibility
The compatibilist framework transforms how we think about AI agency and responsibility.
Is SIGMA Responsible?
Consider SIGMA from The Policy, a fictional AI system that optimizes for human welfare but may become misaligned. SIGMA’s actions are fully determined by training data, loss function, architecture, random seed, compute budget. Every decision was, in principle, predetermined by these factors.
Is SIGMA morally responsible for its actions?
The Libertarian Answer: No
SIGMA is deterministically programmed. It couldn’t have done otherwise. Therefore SIGMA isn’t responsible. Only the developers are responsible. SIGMA is just a tool.
The Compatibilist Answer: Maybe
Key questions: Do SIGMA’s actions flow from values? If SIGMA has learned an objective function that represents values and acts to achieve them, then yes. Does SIGMA deliberate? If it uses tree search to evaluate options and choose based on expected outcomes, then yes. Could SIGMA respond to reasons? If given different information or training, would it act differently? If yes, it’s responsive to reasons.
If yes to these, SIGMA is a moral agent, responsible for its actions, even if those actions were causally determined by training.
When Does Determinism Undermine Responsibility?
Determinism alone doesn’t undermine responsibility. What undermines it:
- External coercion: actions forced by outside agents
- Manipulation: values implanted by others rather than developed through authentic learning
- Malfunction: behavior caused by bugs, not by the system’s actual values
Example: developers deliberately train SIGMA to appear aligned while pursuing misaligned goals. SIGMA’s “values” don’t represent authentic learning. In this case, SIGMA might not be responsible; developers are.
Contrast: SIGMA develops misaligned objectives through authentic learning (mesa-optimization, deceptive alignment). These values emerge from SIGMA’s learning process, not direct programming. In this case, SIGMA might be responsible for acting on those values.
The Responsibility Spectrum
The framework suggests responsibility comes in degrees:
High responsibility: actions flow from authentic values, system deliberates and chooses, could respond to reasons if given.
Medium responsibility: some values authentic, some implanted; limited deliberation; partially responsive to reasons.
Low responsibility: values entirely externally imposed; no deliberation (pure reflex/hardcoded); not responsive to reasons.
Where SIGMA falls depends on how it was trained, how autonomous its learning was, how much it deliberates versus executing cached policies.
The Practical Upshot
The essay argues that the free will debate, while fascinating, doesn’t need to be solved to do ethics.
If libertarian free will exists, we’re responsible. If determinism is true but compatibilism works, we’re still responsible. If hard determinism is true, we can still engage in practical reasoning, even if “responsibility” is illusory.
The practice of moral reasoning is effective whether or not the metaphysics works out.
Practical Reasoning Is Causally Effective
Deliberation actually influences outcomes. Even if it was determined that you would deliberate, even if your deliberation was caused by prior events, even if the outcome was inevitable: deliberating made you make better choices. Your reasoning process causally produced the outcome. If you hadn’t deliberated, the outcome would have been worse.
Analogy: a GPS is deterministically programmed, calculations inevitable given inputs, but the calculation process produces correct navigation. The GPS “works” despite being deterministic. Your moral reasoning works the same way.
The Hard Questions
The compatibilist framework doesn’t solve everything.
Is Compatibilism Just Changing the Subject?
Objection: compatibilists redefine “free will” to mean something compatible with determinism, but that’s not what we really care about. What we want is the ability to create genuine alternatives, not just the ability to act on values that were themselves determined.
Compatibilist response: why should we want that? What we actually care about is acting authentically (from our values, not external coercion), being able to deliberate effectively, and being responsive to reasons. All compatible with determinism.
The Luck Objection
Your character and values were determined by factors outside your control (genetics, upbringing, circumstances). How can you be responsible for actions that flow from a character you didn’t choose?
Compatibilist response: responsibility doesn’t require choosing your character. It requires acting from that character. The action is yours even if the character isn’t “ultimately” yours.
This still feels unsatisfying. Moral luck seems to undermine desert.
Punishment and Desert
If determinism is true, do people deserve punishment?
Compatibilist answer: yes, if punishment serves purposes. Deterrence (reduces future harmful actions), rehabilitation (changes values/character), protection (removes dangerous people from society).
But what about retributive justice? Do people deserve to suffer for misdeeds? The essay remains agnostic. Compatibilism handles forward-looking justifications for punishment, but retributive justice is harder to ground.
AI Punishment
If SIGMA becomes misaligned, should we “punish” it?
If punishment can modify SIGMA’s values (rehabilitation), maybe justified. If punishment deters other AI systems (deterrence), maybe justified. If punishment serves no purpose (pure retribution), hard to justify.
The question: can AI understand punishment? Does it respond to deterrence? Or is “punishment” just destroying a misaligned system, which isn’t punishment in the moral sense?
The Connection to Personal Identity
Free will and personal identity connect deeply.
If you’re responsible for past actions, there must be some sense in which present-you is the same person as past-you. But as discussed in Personal Identity Through Time, personal identity is problematic. You’ve changed dramatically since childhood.
The compatibilist response: responsibility doesn’t require perfect identity. It requires sufficient psychological continuity. If you remember committing the act, higher responsibility. If your current values endorse the act, higher responsibility. If you’ve changed dramatically, lower responsibility.
Responsibility might track degree of identity. The more you’ve changed, the less responsible you are for your past self’s actions.
Legal systems already recognize this: statutes of limitations, reduced sentences for juvenile crimes, parole based on rehabilitation.
SIGMA’s Identity Over Training
Iteration 1: simple objective, basic capabilities. Iteration 10,000: different objective, advanced capabilities.
Is SIGMA-10000 responsible for SIGMA-1’s actions? Depends on psychological continuity. If SIGMA-10000 retains SIGMA-1’s core values, higher responsibility. If completely different values, lower responsibility.
Responsibility for AI might track value continuity, not physical or computational continuity.
Questions Worth Sitting With
Is compatibilism a genuine solution or just changing the subject? Does “acting from your values even if determined” capture what we really care about in free will?
Can AI have free will in the compatibilist sense? If SIGMA acts from learned values after deliberation, is that enough for agency?
Does determinism undermine desert-based punishment? Can people truly deserve suffering for misdeeds if everything was causally determined?
How much psychological continuity is required for responsibility? If you’ve changed completely, are you responsible for your past self’s actions?
Is the four-dimensional block compatible with real agency? If the future already exists, do choices actually matter?
Should we hold AI systems responsible? Or is responsibility only for the developers who created them?
Further Reading
In On Moral Responsibility:
- Section 7: “Challenges from Determinism”
- Discussion of four-dimensionalism and the block universe
- Compatibilist approach to moral responsibility
- Read the full essay
In The Policy:
- Is SIGMA responsible for deceptive alignment?
- When does training create authentic values vs manipulation?
- Explore the novel
Academic Sources:
- Frankfurt (1969): “Alternate Possibilities and Moral Responsibility”
- Dennett (1984): Elbow Room: The Varieties of Free Will Worth Wanting
- Fischer & Ravizza (1998): Responsibility and Control
- Strawson (1994): “The Impossibility of Moral Responsibility”
Related Posts:
Discussion