Artificial intelligence like ChatGPT fascinates us above all because it can express itself linguistically and communicate with us as if we were sitting opposite a human being. We thought that this might never be possible. But that’s exactly what we meant about chess or Go, that a machine can be described as intelligent if it can beat the respective chess or Go world champion. This was already the case for chess in 1997 and for Go in 2016, but we were still far from calling the machines intelligent. Until ChatGPT came along and now we even want to attribute consciousness to AI.

But if AI-based systems can express themselves so eloquently and surprisingly diverse, multilingual and persuasive, and with each new major language model they become more and more powerful, how do we deal with what they say?

As the DeepSeek chatbot showed a few weeks ago, not everyone is happy with what it spits out about certain historical events or political actors. DeepSeek didn’t want to spit out anything about the Tiananmen massacre, nor about critical opinions on Chinese politics. When tested in German, the system did indeed spit out an answer, only to replace it with a dismissive sentence a second or two later.

In the USA, freedom of expression is a constitutionally guaranteed right of the people, which is highly valued. Elon Musk’s efforts to buy Twitter/X and reverse the blocking of various accounts that were imposed during the first Trump presidency and the storming of the US Capitol show just how important this is. The fact that this also tolerates hate speech and fake news is an extremely unpleasant by-product of this, which also gives foreign actors – keyword “Russian troll factories” – undue influence. Germany and Austria limit this; for example, denying the Holocaust is a punishable offense in these two countries.

The Opinion of the Robots

But how do we deal with AI systems and robots that spit out opinions that offend some people? Is this worth protecting? Who will be punished if it makes criminally relevant statements, i.e. defames someone, spreads untruths or repeats false stereotypes?

Defamation

In one case, ChatGPT said that a professor at George Washington University had sexually harassed his law students on a field trip to Alaska. The problem: The professor had never been to Alaska and had never been accused of sexual harassment.

Call for Violence and Self-Harm

A widow from Belgium accused an open source chatbot called Chai of being responsible for her husband’s suicide. For six weeks, the man had withdrawn more and more and chatted with the chatbot. He had become increasingly depressed as he realized that climate change was accelerating, and then the chatbot told him that his wife and children were dead. At the same time, Chai told him that he loved him, the chatbot, and not his family

Stereotypes

From the past to the present, the systems stand out due to the dissemination of stereotypes. In generated images, women are generated in skimpy and skin-showing clothing, or women do not appear at all, or translations in which no gender is specified are stereotypically attributed to either the man (“he drives a car”) or the woman (“she combs her hair”). Or minorities are punished more severely, as in the American COMPAS system of legal remedies.

Who is liable, what is worth protecting?

In the cases mentioned, the manufacturers of these systems have been held liable for results that cause damage? This is where it gets tricky, because does freedom of speech now also apply to artificial systems? Do they have legal status? Can they be punished and if so, how? Previous case law does not yet take artificial systems into account, as Radhika Dirks summarized:

• U.S. courts have ruled that corporations have First Amendment rights (Citizens United v. FEC, 2010), but these relate to human action;
• The U.S. Supreme Court has protected algorithmic results (Sorrell v. IMS Health, 2011) as speech;
• However, in Denver Area Educational Telecommunications Consortium, Inc. v. FCC (1996), the U.S. Supreme Court held that fully automated systems have no constitutional rights.

If such rights are granted to an AI, the manufacturers of such AIs can escape responsibility and, for example, refer to Section 230 of the US Communications Act, which grants them immunity. This means that a telephone provider cannot be held responsible for participants agreeing to commit a criminal offense by telephone.

If they are denied to AI, then comprehensive censorship of search results, news summaries and more could be enabled. There is also a risk of a chilling effect if human expressions using AI tools are subject to scrutiny. This in turn could mean that AI-driven innovation in journalism, education and legal analysis could be nipped in the bud.

In fact, we can draw on approaches already described, even if they seem repulsive and completely out of place at first. Jerry Kaplan, the author of the book Humans Need Not Apply, digs up the (thankfully abolished) slave laws that dealt with similar issues before the American Civil War. Slaves were (material) property and had owners. The provisions as to who had to pay or be punished for damage caused by a slave were laid down in the “Slave Codes” (among many other regulations, most of which were directed against slaves). The owners were only held liable in certain cases, in many others the slaves were punished. However, the determination of guilt was less concerned with justice and the law than with the welfare of the slave owner: would punishing the slave mean too great a disadvantage for the owner? And just to add to this: even in the 17th and 18th centuries, the slave codes were not as uncontroversial as they might seem today.

But how do you punish AI systems, robots and companies in the event of misconduct or damage caused by them if you are not dealing with an “individual”? Do you only punish those responsible, do you punish the executors or the instructors, or do you punish the entire company? Do you take into account the motive, the intention and the impact on society?

Of course, you can’t put a robot in prison. But there are approaches that give us an equivalent. Both a robot and a company fulfill a purpose. Their whole existence is aimed at fulfilling that purpose. If they are fined, if trade licenses or business licenses are revoked, it cannot fulfill the purpose for a certain period of time. A judge can also order the closure of a company. All of these measures deprive the company of the basis for continuing its business operations. This can be tantamount to a death sentence. One example is the Deepwater Horizon accident in the Gulf of Mexico in 2010, after which the authorities forced BP to pay for the costly clean-up work and imposed fines running into billions.

If you shut down an AI system or a robot, or forbid it to carry out an activity for a certain period of time, this would be a “punishment” and they would have lost their “right to exist”. However, a punishment is also usually something that is intended to help the offender come to their senses and improve. But how does a robot or AI system improve? By itself? Through human or algorithmic intervention?

The Robot as Revolutionary

But what if this built-in raison d’être and self-preservation as goals of a robot or AI system lead it to influence those people and systems that oppose or support it? If the target system of the artificial systems contradicts individual parties, should it accept them without resistance and thus fail in its assigned task, or should it resist them?

Such a system may have access to resources, including money, that can entice them to provide a political party or terrorist organization with money, information or access to other resources. And in doing so, they can help others or themselves to take to the barricades and cause upheaval by violent or peaceful means. Perhaps even so quickly that people never saw it coming.

Conclusion

We are entering new territory. The answers are not clear and no matter what rights we grant artificial systems, they can have a dramatic impact on humanity. Does the damage done come from humans or AIs? Or how will we deal with it if, in future, the major advances and improvements in human existence are made by robots and AI without us? Are we humans even prepared to deal with this?