As AI becomes deeply integrated into law enforcement and autonomous technologies, a pressing concern arises: Are current criminal laws equipped to assign responsibility when AI systems make decisions that lead to harm or injustice? This discussion invites legal scholars and tech experts to explore whether our legal frameworks need reform, or if existing laws can adapt to the challenges posed by AI in areas like self-driving cars and predictive policing.
I believe existing criminal law struggles to fairly address AI-driven actions in areas like autonomous vehicles or predictive policing systems. Traditional laws were designed with human intent and behavior in mind, so they often don’t directly apply to AI's decision-making processes. For example, in the case of autonomous vehicles, if an accident occurs, it’s unclear who is responsible – the manufacturer, the software developer, or the vehicle itself. Similarly, predictive policing systems might target certain communities unfairly based on biased data, raising concerns about accountability and discrimination. I reckon that the law will need to evolve to create new frameworks that specifically address AI's role in these areas, ensuring fairness and accountability while also considering the unique nature of machine decision-making.
No, at present a machine cannot have criminal mens rea. This needs urgently addressing as you will see from the attached article.
It seems it is possible to teach AI to be dishonest and tell lies. This is a very useful tool for criminals because the AI cannot develop a mens rea because it is not a creature in rerum naturae.
That proved to be a problem for many years in establishing an effective method of dealing with non-human personages in manslaughter for instance. Now we can have corporate fraud committed by a machine we will have to extend the legal concept of the controlling mind to whoever uses the advice given by an AI system that proves to be fraudulent.
This is going to be a real handful for prosecutors.
The interesting thing is that in crimes of specific intent such as theft or fraud a defendant cannot be convicted unless evidence of actus reus (guilty act) and mens rea (guilty mind) can be proved to the criminal standard of evidence.
Theft is dishonestly obtaining property belonging to another with the intention to permanently deprive (s.1 Theft Act 1968) The elements of mens rea are dishonesty and intention. The actus reus is obtaining and depriving.
In the case described above the machine has neither obtained, nor deprived and since it is not conscious (in rerum naturae) it cannot have been dishonest or had any intention.
Clearly the machine is not benefitting from the proceeds of the crime and the individuals using it are engaging in criminal behaviour but they cannot be in criminal conspiracy because it too requires intention on the part of the conspirators.
Did the humans in this example incite the AI to simulated dishonesty? Did the AI incite them? Inchoate offences are often very difficult to prove.
This example indicates the need for some form of 'moral compass' being trained into AI. This is going to be incredibly difficult because that moral compass has no clear boundaries. Many transactions which may be immoral by the standards of most of us are not illegal and we cannot prosecute people for being immoral.
We train humans morality (right & wrong) by reference to societal and cultural norms by coercion, a system of sticks and carrots. Humans obey rules because they wish to avoid adverse consequences of them or because they do not wish group opprobrium for breaking those norms. We know very well that this is at best only partly effective.
Criminality can also be learned so we need to be very careful about what AI decides to do with its knowledge.
Existing criminal law faces considerable difficulties when applied to AI-driven actions in autonomous vehicles and predictive policing systems. The core challenge lies in the fact that traditional criminal law is based on human intention and voluntary acts, concepts that do not straightforwardly translate to decisions made by AI systems.
For autonomous vehicles, determining liability in the event of an accident is complex, as the AI operates independently of direct human control. It is unclear whether responsibility should lie with the manufacturer, the software developer, the owner, or another party. This gap indicates that current legal frameworks may not be fully equipped to address these situations fairly.
In the case of predictive policing, AI systems analyze data to anticipate criminal behavior, which raises concerns about bias and discrimination. Since these systems rely on data patterns rather than human judgment, holding someone criminally accountable for errors or harm caused by these predictions is challenging under existing laws.
Therefore, while existing criminal law provides a foundation, it may not be sufficient on its own to fairly address the complexities introduced by AI in these contexts. Adaptations or new legal mechanisms might be necessary to ensure accountability and protect individual rights effectively
Of course it can. You asked an abstract question, and the colleagues who gave answers above are considering abstract models. In Russia, we have an expression for such a case: ‘a spherical horse in a vacuum’ (as you understand, horses are not spherical and do not exist in a vacuum).
If we discard abstractions and turn to a concrete situation, to a concrete real-life case, then the question ‘who is to blame?’ will find a clear and logical answer immediately. And if you also define what exactly you mean by AI, everything becomes even easier.
I have articles on this topic in Russian; I will gradually post English translations here, so that everyone who does not speak Russian can read them and clarify such points for themselves.
Best regards,
Marina
Thanks Marina
I have been collecting Russian proverbs for ages. I can't wait to use that one on my Russia colleagues.
Please do post the Russian articles or send them to me, I look forward to reading them.
Point noted in the Russian context. The question whether criminal liability could be attributed to AI. AI is not a lega entity for now. If at all reasonable, those who created AI should be responsible. This is what is going through the HL on transparency of data. For AI cars, there had been fatalitie.
It is difficult to imagine how AI can be a legal personage. We may be asking too much here about the somewhat exaggerated hype of what it actually is at all.
Unlike a corporate legal personage, where the directors are the controlling mind it is the intention of AI to be the controlling mind itself. Companies do not think autonomously but AI is supposed to. Placing legal responsibility on this who created it should dangerously like holding an innocent family responsible for the crimes of one of its offspring.
AI is a long way from autonomy, which requires full sentience and consciousness. There is every possibility that this will never be achieved and we should not be influenced by the hubris that is often a consequence of technological advances (remember the Titanic)
Without consciousness or sentience AI cannot form the mens rea for crime or be strictly liable. Neither can it be negligent, unless we can create a machine that has 'duty' and can 'care'. Once again don't be fooled by AI 'congratulating' you for your good questions or 'apologising' when it gets something wrong. AI does not have feelings and is nowhere near it.
For the moment we will have to rely on the law as it pertains to all defective equipment and the consequences of those defects. Trying to determine laws for something that has not yet happened is like wondering if visiting aliens from another world would be subject to our laws, or we theirs.
Point taken and agreed. There have been debates on AI persona even at the EU and some countries even accorded robots with certain frictional characteristics. All said, criminal law concern the guilty mind and the action that is in extension of the human mind. Some cases, strict liability offence requires only the doing of the prohibited act. A human or a legal entity must be the focus.
Dear Colleagues, I wrote in my first reply: 'Of course it can'. But ‘it’ is not AI. ‘It’ is criminal law. The first sentence in my first reply should read: of course criminal law can.
Chua Alan you wrote: 'The question whether criminal liability could be attributed to AI'. But Mr Raza asks a different question: 'Are current criminal laws equipped to assign responsibility?'. And I answer: yes, criminal law can it. The emergence of a new technology, commonly called AI, has not changed anything.
AI is not a legal entity, quite right. But legal entities are not criminally liable (at least in our country and some others). AI is not a person at all. And it will never become a person.
Barry Turner I agree with your comment in many ways, but not all. I will point out the main thing: it is incorrect to compare a child as an extension of the parents and a product as an extension of the manufacturer. An offspring is a person, like his parents and all other people. And that is why he is responsible for himself. AI is a product. Therefore, the manufacturer should be responsible for the quality of its work, just like any other product.
Best regards,
Marina
Barry Turner since you like proverbs, you might like this anecdote too:
‘- What wonderful children you have!
- Why do you eulogize our children?
- Your children are very good. But everything you do with your hands is very bad.’
This anecdote, in my opinion, shows very well the main problem of AI.
Best regards,
Marina
Marina
The problem with any analogy is that at the moment none of the tools that have even been designed since the palaeolithic have been capable of thinking for themselves. Even our most powerful computers cannot do this and are entirely reliant on human input. Faults occurring as a result of that input can be attributed to programmers/operators etc and direct or vicarious liability come into play.
A computer directed to commit a crime, for instance withdrawing money from a bank account without authority will be doing that at the command of a person (human). The problem occurs when the program does it on its own. A recent experiment was carried out on a stock trade with and AI program that was trained to recognise insider dealing.
The AI in a simulation (see below) went ahead with a transaction that it should have recognised as illegal. Te suggestion was it had taken a calculated risk with is quite alarming. Even more alarming is that when it was questioned over this it appears to have lied. A criminal 'mind' indeed.
Insider dealing is a crime of specific intent and cannot be prosecuted without proving mens rea to the criminal standard of proof.
The operators of the system did not direct the transaction and even could be said to have warned against it. It would be rather difficult to prove an inchoate offence of conspiracy on their part.
It might of course be possible ex post facto to prosecute them for benefitting from the proceeds of crime if they kept the profit. We will be seeing some attempts at this because we now know that AI can certainly commit the actus reus.
I should perhaps quote the Chinese proverb (curse) "May you live in interesting times". It is certainly going to be interesting.
On Russian proverbs we might want to consider this one in relation to AI.
С ложью далеко зайдёшь, но назад не вернешься
Will AI go too far? It is certainly the case that once it has gone too far, like the saying goes it will be very hard to get back.
Barry, it's not the AI that took the risk and lied, it's humans with crooked hands (кривые ручонки, all thumbs) and lack of intellect. These persons have made a bad product that does not work well. This product cannot be used. People make bad products often. This is not a product problem, much less a legal problem. It's a problem of the manufacturers of bad products only.
AI is incapable of thinking. What is presented in the media as thinking or a semblance of thinking is just mathematical calculations according to an algorithm.
In criminal law, besides intent, there is also negligence (as a form of guilt, Art. 24 of the Criminal Code of the Russian Federation) and criminal negligence (as a particular offence, Art. 293 of the Criminal Code of the Russian Federation).
Ok, I won't expand further on this topic. As I was going to do, I will post my articles here; it is discussed in detail there.
Best regards,
Marina
It is too late to prevent the extension of the use of AI. It is already in widespread use and just about every government in the world is in favour of it. Laws governing the use IT and computing have never been very effective and now by far the majority of fraud is carried out by using IT and the Internet.
The controlling hands are still of course, in the main human and the mens rea is human. Where the fraud is committed the perpetrators are often 'betrayed' by the
traceability of this action.
The problem is with fully autonomous systems (AI is not yet fully autonomous) carrying out acts which if done by a human would be criminal.
If a machine breaks the law without human intervention the remedies are limited. If a machine breaks the law and the humans who are using it do not inform the authorities they then will have a degree of culpability as accessories after the fact and could be prosecuted for benefitting from the proceeds of crime.
There could be a strict liability offence of being in possession of a machine that commits offences such as insider trading or fraud. Health and Safety Law, which I used to teach is clear on this. Operating machines that injure people because they are faulty is already an offence under strict liability in the UK. Although Russia does not have a traditional "strict liability" category in its criminal law in the same way that common law systems do.
There is some parity with the common law system of no fault liability. To a large extent the common law systems while not entirely dependent on fault do have some degree of culpability. In circumstances where dangerous machinery is used there is a duty of care under the common law to ensure that it is safe as far as reasonably practicable. This is compatible with the continental European civil law tradition, requiring some form of fault (вина) for criminal liability.
Russian criminal law requires that "criminal liability for such consequences shall ensue only in cases where the person foresaw the possibility of their onset, but expected without valid reason that they would be prevented, or in cases where the person did not foresee, but could and should have foreseen" the criminal code in these consequences.
The Russian Criminal Code recognizes two main forms of fault (вина):
1. Intent (умысел) - which can be direct or indirect
2. Negligence (неосторожность) - which includes both conscious negligence (легкомыслие) and unconscious negligence (небрежность)
Russian criminal law operates on the principle that criminal liability generally requires proof of fault - either intentional or negligent conduct. Unlike some common law jurisdictions that have strict liability offenses (where no mental element needs to be proven), Russian criminal law maintains that some form of psychological attitude toward the act and its consequences must typically be established.
There are very limited exceptions where something approaching strict liability might apply, primarily in certain regulatory or administrative violations that carry criminal penalties, but these are rare and the general principle remains that criminal responsibility requires fault in the form of intent or negligence.
There is a comparison between the civil laws and strict liability model. Anyone operating a machine that could cause injury must take care that it does not. For instance by not allowing untrained persons to use it. This corresponds well with the Russian concept of легкомыслие (conscious negligence). While the machine is not ‘conscious’ its user is and should recognise the potential danger and the likelihood of it occurring.
In spite of the Russian code being civil codifed in origin it is remarkably similar to the UK Health and Safety at Work Act 1974. The single most obvious failure of Art 143 (Violation of Labour Protection Rules) is that it pertains only to victims who are covered by this protection. While an employee injured by an AI machine at work seems protected a pedestrian injured by an autonomous drive vehicle does not seem to fit into that category.
Perhaps unconscious negligence (небрежность) is a category that could be applied. The owner of an autonomously drive vehicle should know that it is potentially dangerous. That most certainly is a strict liability concept.
The clarification needed is which criminal law and what kind of AI-driven action is being discussed by Marina Komashko . I may have missed the point that is being raised. I don’t think th issue is about AI being liable but the consequences of the AI driven action could be addressed by current criminal law. As laws are different in each jurisdiction, it will be enlightening to know more.
Chua Alan 'criminal law fairly address AI-driven actions' in your post looks like a quote. But Mr Raza, who started this discussion, did not write such a phrase. I didn't write such a phrase either.
'Where do we start?' you asked. First we need to define the nature, the essence of AI (on the technical side). Perhaps because of the difference in languages, these words reflect what non-English speakers and English speakers understand by them, not entirely in the same way. But I described the nature of AI in some detail in a discussion with our colleague Barry.
Best regards,
Marina
It is possible to address the consequences of a critical incident caused by AI. At present AI is neither sentient nor conscious and in spite of vast amounts of media hype is not anywhere near it. I have been suing the most anthropic versions and while the designers have made very impressive imitations and simulations of human empathetic consciousness it is not aware of its own deliberations.
The most anthropic versions apologise when it cannot answer the question and even sound enthusiastic in some of the replies. I find it amusing that it says sorry since that is an empathetic emotional response. It most certainly does not experience remorse or disappointment.
I am often congratulated by AI for the 'importance' of my questions. This is a programmed response because it is not possible for AI to recognise what would be important in a human sense.
From a legal perspective this is interesting. A sentient human showing real remorse for committing a crime is usually given credit in the sentencing. Those who do not get harsher sentences.
For the foreseeable future AI is a non-sentient machine. It does not think and cannot possess mens rea. More to the point it is a machine that is the property of a legal personage, either human or corporate. I like the Russian/civil code approach to this. The owner, legal personage is responsible for the actions of a machine that they ought to know poses a danger. Art 143 perhaps needs to be revised, since at present it only recognises grave harm as a consequence.
The common law version works just as well. The owner is culpable if they fail to take all practicable steps to prevent harm caused by anything in their control. There is little doubt we will see legal actions based on either legal principle.
One the last three decades common law has extended the range of negligence into the criminal arena. In the UK corporate manslaughter is now well established as is gross negligence manslaughter. For th immediate future they must be used to prosecute those who precipitate serious harm via AI application.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology