Can AI be morally responsible? Why blame, liability and tool-use are different questions

Lode Lauwaert and Ann-Katrien Oimann write that today’s AI cannot itself be held morally responsible, while other philosophy papers argue for more distributed or functional views of responsibility. The disagreement turns on whether…

AI-drafted, human-reviewed

Not in any settled sense. Lode Lauwaert and Ann-Katrien Oimann write that today’s AI systems cannot themselves be held morally responsible, but they also argue that using AI does not necessarily create a responsibility gap because humans can still be responsible; other authors argue for a more distributed or functional idea of responsibility in which AI can take on a responsibility-like role without replacing human oversight.

What moral responsibility means here

Lauwaert and Oimann say the first step is to separate different meanings of responsibility. Their chapter distinguishes causal responsibility, moral responsibility and role responsibility.

Lauwaert and Oimann describe moral responsibility as the kind that makes someone a proper target of praise, blame, reward or punishment. They state the point directly: “The thesis of the so-called responsibility gap is thus concerned with moral responsibility.”

The same chapter also says this is not a legal analysis. In a footnote, Lauwaert and Oimann state that their discussion is not about liability or legal responsibility. That matters because a reader asking whether AI can be morally responsible may really be asking a different question, such as the criminal-law question discussed in whether AI can be charged with a crime or the legal-status question discussed in electronic personhood.

Behdadi and Christian Munthe say the wider debate often mixes several issues under the label artificial moral agency. Their paper argues that the practical question is not only what inner properties a machine has, but also how artificial entities should be included in practices that normally assume moral agency and responsibility.

One reading of these papers is that can AI be morally responsible contains at least three separate questions: can AI itself deserve blame; can some human or organisation still be liable or otherwise answerable; and should AI be treated as no more than a tool. The sources do not collapse those questions into one.

Why one prominent view says no

Lauwaert and Oimann answer the blame question narrowly. They write that current artifacts cannot be the proper targets of moral reactions because moral responsibility presupposes the ability to be affected, mentally or physically, by those reactions.

On their account, today’s autonomous technologies do not have that capacity, so holding them morally responsible would be mistaken. Lauwaert and Oimann add that this is why it is absurd, in their view, to hold present-day artifacts responsible for what they do.

Behdadi and Munthe describe a related line in the literature as the standard view. Their paper says this view ties moral agency to conditions such as rationality, autonomy or free will, and phenomenal consciousness, meaning subjective experience.

The steelman for the no side is straightforward. If moral responsibility is about justified blame, and if blame makes sense only for beings that can understand, feel, or meaningfully answer for what they have done, then current AI does not qualify on Lauwaert and Oimann’s account. On that reading, debates about AI rights or moral status are related but not enough by themselves to make a system blameworthy.

Lauwaert and Oimann do not say the future is settled forever. They write that moral practices can change, and they say the question could reopen if artifacts could one day be affected physically or mentally in the relevant way. But their claim about the present is clear: current AI is not itself morally responsible.

Why other authors leave room for a responsibility-like role

Dodig-Crnkovic, Gianfranco Basti and Tobias Holstein argue for a different framing. Their paper says traditional boundaries of moral responsibility are being challenged as AI systems operate with more autonomy and adaptability.

Dodig-Crnkovic and her co-authors propose what they call a functionalist perspective. In plain English, that means treating responsibility less as a private inner trait and more as a role inside a socio-technical system, which their paper describes as a system of humans, machines and organisations acting together.

On that view, Dodig-Crnkovic and her co-authors say responsibility can be distributed among human and artificial agents. Their paper argues that AI can be built to follow ethical guidelines and assess decisions ethically, and it says such systems may count as artificial moral agents in that functional sense.

But the same paper does not hand final responsibility to the machine. Dodig-Crnkovic and her co-authors state that ultimate responsibility for the whole socio-technical system still falls on humans. They also cite the United Nations AI Advisory Body’s first guiding principle: “AI should be governed inclusively, by and for the benefit of all.”

Behdadi and Munthe reach a different but compatible practical point. Their paper argues that the debate should move away from only asking which metaphysical conditions make a machine a moral agent, and toward the normative question of how artificial entities should be involved in human practices of responsibility.

The steelman for this side is that AI already helps make decisions that affect people, and the social need is to govern those systems, assign oversight, and shape behaviour. On that reading, asking only whether a system deserves blame may be too narrow to do the work societies need.

Blame, liability and tool-use are not the same question

Lauwaert and Oimann expressly separate moral responsibility from liability and legal responsibility. So a claim that AI cannot be morally responsible does not by itself answer who pays damages, who violated a regulation, or who counts as a legal person.

Behdadi and Munthe report that some participants in the debate treat machines as advanced tools or prosthetics, partly because machines are designed for human purposes. Lauwaert and Oimann cite Joanna Bryson in a footnote as rejecting moral responsibility for robots and treating them as tools rather than responsible agents.

The tool-use position has a strong version. If AI systems are built, trained, deployed and updated by humans, then one can say the machine is better understood as an instrument through which human purposes are carried out. On that reading, the real moral question is which people or organisations chose the tool, set the task, or accepted the risk.

Behdadi and Munthe also explain the main reply. Their paper says arguments from design can be pushed too far, because human beings are also shaped by genes, upbringing and environment. Their point is not that humans and machines are the same, but that design origin alone may not settle the whole moral question.

Dodig-Crnkovic and her co-authors add another distinction. Their paper says that delegated responsibility to AI should not focus on guilt or punishment for machines, but on encouraging good behaviour and reducing risk inside the larger system.

What the responsibility gap changes

Lauwaert and Oimann frame the central puzzle this way: “When autonomous systems make a mistake, is it unjustified to hold anyone responsible for it?”

Their answer is no, not necessarily. Lauwaert and Oimann argue that humans can still be morally responsible when they knowingly use autonomous systems in contexts where foreseeable harms may follow.

To make that case, Lauwaert and Oimann compare AI use to other forms of risky delegation. They press the point with a question: “If the officer is morally responsible, why shouldn’t the same apply to those who decide to use autonomous AI systems?”

Lauwaert and Oimann say classical moral responsibility turns on at least causal involvement, autonomy and knowledge. They argue that a lack of real-time control does not by itself erase responsibility. As they put it: “Not having control does not absolve moral responsibility.”

The steelman for the no-gap position is that people can be responsible for taking a foreseeable risk even when they do not control every later step. Lauwaert and Oimann say users, programmers, manufacturers or commanders may meet the relevant conditions in particular cases.

Still, Lauwaert and Oimann do not claim there is never a gap. Their chapter says a gap may remain if the consequences were unforeseeable, or if one of the other conditions for moral responsibility is not met. That is a narrower claim than saying AI always leaves someone blameworthy.

Dodig-Crnkovic and her co-authors shift the emphasis from blame to design and oversight. Their paper argues for ethics by design, monitoring, continuous learning and human oversight because AI can act at speeds beyond human supervision. Behdadi and Munthe likewise say the hard practical work is deciding how responsibility practices should change as artificial systems become more capable.

The bottom line from these papers is limited but clear. Lauwaert and Oimann say present-day AI is not itself morally responsible in the blameworthy sense. Dodig-Crnkovic and her co-authors, and Behdadi and Munthe, leave more room for AI to occupy a responsibility-like role inside human systems. The live disagreement is less about one word than about which job responsibility is supposed to do.

Frequently asked questions

Is moral responsibility the same as legal liability?

No. Lauwaert and Oimann explicitly separate their discussion of moral responsibility from liability and legal responsibility. Their chapter is about blame, praise and justified candidates for punishment or reward, not about legal rules.

Do AI systems necessarily create a responsibility gap?

Lauwaert and Oimann say no. They argue that lack of control alone does not erase moral responsibility, though they also say a gap may still exist when knowledge, autonomy or causal involvement is missing.

Could future AI ever be morally responsible?

Lauwaert and Oimann do not rule that out in principle. They say current artifacts cannot be held morally responsible, but they also say moral practices can change and that the question could reopen if future systems could be affected physically or mentally in the relevant way and met other conditions.

If AI shares responsibility, does that let humans off the hook?

No, on Dodig-Crnkovic and her co-authors’ view. Their distributed model says AI may take on an ethical role inside a socio-technical system, but ultimate responsibility for designing, deploying and overseeing that system still rests with humans.

Caveats: The three-way distinction between causal, moral and role responsibility, and the claim that today’s AI cannot itself be morally responsible, come primarily from Lauwaert and Oimann’s chapter. The distributed and functionalist counterarguments come from separate papers by Dodig-Crnkovic and co-authors, and by Behdadi and Munthe.

Sources

Tags: moral-responsibility · responsibility-gap · ai-ethics · artificial-moral-agency

← all briefings