Science Fiction, AI & The Fourth Law of Robotics
On using science fiction writing to envision and predict the future, the Three Laws of Robotics by Isaac Asamov, and a Fourth Law for the 21st century.
I received a friendly challenge from to write about the “Fourth Law of Robotics” as proposed by Dariusz Jemielniak in IEEE Spectrum.
The challenge is hereby accepted.
Do you, dear reader, have any suggestions on topics you would like to see me cover, feedback, or other kinds of inputs? Please don’t hesitate to message me here on Substack, LinkedIn, or via e-mail on futuristiclawyer at gmail.com.
Looking forward to hearing from you!
Science Fiction & AI
Science fiction literature has been formative for the AI industry in America. Not just as an inspiration but as a goalpost. OpenAI, Anthropic, Google DeepMind, Safe Superintelligence Inc., Thinking Machine Labs and other ambitious firms are not just believing in the possibility of superintelligent AI, they are actively trying to build it.
In my definition, “superintelligent AI” must be an AI that resembles an entity more than a machine. It has agency and is driven by an intrinsic motivation in the same way as humans are but it far surpasses us with its superhuman abilities. Again, deep in the research department of major AI labs, this is not a dream but a plan. We can look to “AI 2027” which is a story devised by former OpenAI alignment researcher, Daniel Kokotajlo, the prominent writer in the rationalist community, Scott Alexander, and the AI safety researchers Thomas Larsen, Eli Lifland, and Romeo Dean. It provides us with a concrete timeline for the development of superhuman AI (which according to the authors could be in 2027).
The story received both praise and critique for its boldness. Suffice to say, it’s a blend between science fiction and prediction-making, not unlike the classical Silicon Valley book The Singularity is Near by Ray Kurzweil or the more recent manifesto Situational Awareness by OpenAI insider, Leopold Aschenbrenner. These works are all part of a literary genre that treats science fiction like predictions, as synonymous with a forthcoming future. The problem is: if we believe too strongly that we can see the future it can harm us as much, if not more, than clinging too hard to the past. Let’s look at a quick example to see what I mean.
The Metaverse & Other Strange Ideas
Zuckerberg was motivated to build the Metaverse - a future virtual reality world where people would choose to work, socialize, and spend their free time - because he wanted to own the infrastructure for a new paradigm of computing. Facebook, Instagram, and WhatsApp are still to this day dependent on app stores, search products, hardware products, operating systems, and cloud computing owned by Apple, Google, Microsoft, and Amazon. Meta’s core social media business is essentially a second layer built on top of infrastructure owned by the other BigTech companies. Before Meta’s pivot into “open-source” AI, the Metaverse was how Zuckerberg imagined that he could own the mines and shovels during a future gold rush.
“The Metaverse” is a concept taken directly from the science fiction novel Snow Crash by Neil Stephenson. Another science fiction novel Ready Player One by Ernest Cline, later made into a popular movie by Steven Spielberg, takes place in a very similar virtual reality environment. These books probably helped to shape the perception that living part of our lives in virtual reality was desirable, even inevitable in the longer term. In reality? Let’s just say that none of us are using VR glassess instead of smartphones and laptops today.
Trying to predict the future through science fiction can lead to disappointment, bleeding stocks, and above all, some strange ideas- as true for the metaverse as it is for AI. For example, as the famous futurist Jaron Lanier recalled during an interview with Vox:
“Just the other day I was at a lunch in Palo Alto and there were some young AI scientists there who were saying that they would never have a “bio baby” because as soon as you have a “bio baby,” you get the “mind virus” of the [biological] world. And when you have the mind virus, you become committed to your human baby. But it’s much more important to be committed to the AI of the future. And so to have human babies is fundamentally unethical.”
Far out, right? Some of the bright engineering minds of our generation are literally wanting to build a future where AI replaces humans.
We have to be careful not to get too attached to science fiction ideas and mistake it for a future written in stone. That being said, science fiction, if handled with care, has the potential to show us a vision that innovators can strive for and the public can prepare for.
Let’s take a closer look at an idea that was developed through fiction many years ago but has helped to shape our thinking even today on how we can regulate human-like technologies, the Three Laws of Robotics.
The Three Laws of Robotics
For those who are not familiar with the “Three Laws of Robotics”; they originate from a short story called “Runaround” by the legendary science fiction writer Isaac Asimov. The story was first published in a science fiction magazine in 1942. The Three Laws are as follows:
First Law: A robot may not injure a human being or, through inaction, allow a human being to come to harm.
Second Law: A robot must obey orders given by human beings except where such orders would conflict with the First Law.
Third Law: A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Although made for fiction, the Three Laws have been cited in more than a few policy papers and scholarly debates around the world. The European Parliament referred to Asimov’s Laws in 2017 in its resolution on Civil Law Rules on Robotics. A member of the French parliament suggested that the Three Laws of Robotics should be adopted in the French constitution in 2020. The Three Laws have also helped to shape robotics policies in the UK, and South Korea, as well as safety policies at Google DeepMind’s robotics team.
As logical as the Three Laws appear, they are far from perfect. Asimov himself explored the hidden conflicts between them in much of his work, including in his “Robot series” which spans 37 short stories and 6 novels. Asimov had to adopt “The Zeroth Law” (as in law number zero) to deal with a situation where a robot is told to protect an individual who harms humanity as a whole. The Zeroth law outranks the other three laws and states:
"A robot may not harm humanity, or, by inaction, allow humanity to come to harm."
Again, Asimov’s Laws are far from a bulletproof template to align human interests with robots, or AI, as the case may be in 2025. In fact, as Asimov shows us through his stories, it’s extremely difficult, if not impossible, to formulate absolute principles that make perceptive robots serve our interests. Text is black and white and code consists of 1s and 0s. Neither format is able to catch all the subtleties and complexities of real life.
That is arguably the whole reason why lawyers exist in the first place; to interpret and conceptualize the gap between people’s intentions and the messiness of real life. Additionally, laws tend to lose their relevance and/or logical value with time. Asimov could never have predicted the rise of social media and AI. Can we repurpose the Three Laws of Robotics to serve our best interests against machine intelligence in this part of the 21st century?
The Fourth Law of Robotics
In a guest article for IEEE Spectrum, Dariusz Jemielniak, professor at Kozminski University, Poland, proposes a new addition to Asimov’s laws:
Fourth Law: A robot or AI must not deceive a human by impersonating a human being.
Jemielniak expresses rightful concern about AI-enabled deception, including cybercrimes involving digital manipulation and social engineering, deepfakes, sophisticated AI-generated propaganda and misinformation, impersonation scams, and manipulative AI agents that users can form emotional bonds with.
Quoting from the article:
“We need clear boundaries. While human-AI collaboration can be constructive, AI deception undermines trust and leads to wasted time, emotional distress, and misuse of resources. Artificial agents must identify themselves to ensure our interactions with them are transparent and productive. AI-generated content should be clearly marked unless it has been significantly edited and adapted by a human”
The Fourth Law of Robotics is meant to address an issue Asimov could not have foreseen. It relates to the loss of social trust caused by AI-generated slop on social media, fake users, and artificial images, videos, voices, sounds, and language. We can barely decipher what is real anymore on the internet. Jason Koebler from 404Media says that “AI slop is a brute force attack on the algorithms that control reality”.
AI slop is often annoying, distasteful, and too much of it erodes the online experience. However, even more problematic is AI-generated content and bots that we can’t identify as such. As the capabilities of foundation models improve, impersonating scams, manipulative AI partners, and even convincing AI online creators will only become more challenging problems. That is why we need the Fourth Law of Robotics.
Yuval Noah Harari explained in an interview with Wired why freedom of expression should not be extended to bots on social media. This is because only humans are stakeholders in society and interact with the world through our physical bodies. Harari gives this example:
“If the sewage system collapses, we become sick, spreading diseases such as dysentery and cholera, and in the worst case, we die. But that is not a threat at all to AI, which does not care if the sewage system collapses, because it will not get sick or die. When human citizens debate, for example, whether to allocate money to a government agency to manage a sewage system, there is an obvious vested interest. So while AI can come up with some very novel and imaginative ideas for sewage systems, we must always remember that AI is not human or even organic to begin with”
No matter how capable current AI models become, they don’t experience the world like humans do. On this background, they shouldn’t have a vote that could influence human affairs. The truth is that AI models ultimately couldn’t care less if humans suffer or thrive, live or they die. Would you trust a friend or counselor who is completely indifferent to whether you live or you die?
Looking Forward
Finally, let’s get practical. Jemielniak proposes five implementations the Fourth Law would require:
Mandatory AI disclosure in direct interactions,
Clear labeling of AI-generated content,
Technical standards for AI identification,
Legal frameworks for enforcement,
Educational initiatives to improve AI literacy.
EU’s AI Act does contain transparency obligations in Article 50 with requirements to disclose AI in direct interactions and clearly label AI-generated content.
AI Act Article 50 (1):
“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use (..)”
Article 50 (2):
“Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated (..)“
Article 50 (4):
“Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated (..) Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated.”
These requirements provide a good starting point for implementing the Fourth Law of Robotics in Europe. Via a phenomenon known as the Brussels Effect, European standards could proliferate to other countries.
What is left to solve now is the technical challenge of labelling and identifying AI-generated content, the legal challenge of enforcing breaches, and the educational challenge of improving AI literacy among users. It will require close cooperation between the public and private sector and coordination among countries. As with science fiction literature, the Fourth Law of Robotics gives us a vision to aspire to, but not a road map showing us how to get there.
I'm happy you have taken on my "challenge." I was really curious what you, as someone with a legal background, would say about this proposed Fourth Law of Robotics.
There is another story here about the stories we tell. As you noted, Asimov's Laws of Robotics and other sci-fi stories had a massive impact on many generations of engineers and influenced what people put their time and energy into building.
Because of how much influence and power such stories have, be it a book, a movie or a game, and how much they shape our future, we need those stories to show us not only what can go wrong but also what can go well and celebrate human ingenuity. Not just Black Mirror but also White or Grey, or All-Colours-of-the-Rainbow Mirrors, too.