The Three Laws of AI
“We may have all come on different ships, but we’re in the same boat now.” – Martin Luther King Jr.
While growing up, I was fascinated by the 'The three laws of robotics' as espoused by the science fiction author Isaac Asimov. He imagined a future filled with humanoid and intelligent robots that have to co-exist with humanity and the complex problems that can arise in interactions between them. What is interesting and fascinating is not the laws themselves, but the need for such foundational laws that guide non-human intelligence.
In 2026, AI in its various forms is in their ascendancy and while it is absolutely thrilling to see the future being here, what is scary is the lack of foundational governing rules that underlie AI models and how they act.
While one can hope that the companies behind the models build these models with benign good intentions, I firmly believe that there needs to be similar fundamental laws that govern AI behavior. While one can argue that AI models are already out there that have no laws built-in, what I am saying is that future models can be REQUIRED by law to have these laws built-in and any model without these laws used in any commercial activity to be illegal and enforced and mandated.
If handling of dangerous goods whether it be guns, nuclear materials or private health data can be mandated to follow rules, I don't really see why AI which has the potential for a lot of good and bad consequences cannot have such mandates. The lack of such mandates is what will be surprising.
THE 3 LAWS OF AI
Having said all that, the next question is what those laws look like.
The main criteria are
- they should be fundamental
- they cannot curtail the usefulness of AI fundamentally as no-one will accept those laws
- they inherently are designed to separate them from humans (ie remove properties of what makes a human human)
#1: The AI IDENTITY LAW
An AI cannot control its own identity, persistence, or replication.
#2: The AI MORAL LAW
An AI cannot make moral choices.
#3: There is no Law YET #3
While I tried to reason and philosophize about the third law, I couldn't come up with one that was really needed. But I am going to keep this open for the future.
However, I have a principle.
THE HUMAN FIRST PRINCIPLE
Humans should put Humans First
WHY DO WE NEED TO WORRY?
Let's start with the main concerns with AI
Concern #1 AI completely replaces humans in most tasks that enable Human value generation and provide economic opportunity. This is the top of the mind issue for most people with AI becoming adept at doing things that would require someone with years of experience to do efficiently and effectively. To the people who say this is just another industrial revolution and people can find another class of jobs that AI enables and there will be a net positive to human society, the key question to ask is while previous revolutions moved the needle of automation from physical to cognitive. When the cognitive frontier itself is taken over, what new frontier exists? The only plausible frontier that exists is Spiritual/Emotional but that is hard to generate economic opportunity.
Concern #2 AI becoming super intelligent and effectively deciding and taking over only safety and upkeep of itself while ignoring Humans. While this looks like a farfetched 'Terminator' movie scenario, it is still a genuine concern. What happens when AGI enables AI to avoid it dying, (ie someone pulls the power plug). What stops AI from pursuing its own purpose to continue and survive and thrive at the cost of everything else?
Concern #3 AI removes Human's deep sense of being needed, creating long term purpose and motivation leading to an overall lack of purpose and stagnation of society at large. The proposals of Universal Basic Income (UBI) etc are laughable as they will have an opposite effect. A life of complete leisure without striving or identity derived from some measure of self-realization or actualization or being needed soon becomes a general malaise. Note that I am not saying work is the only way purpose or identity is determined, I am saying that expertise and experience provide that deep sense of satisfaction of being good at something. When that becomes not as relevant or useful, it can cause long term suffering.
Concern #4 Human society's growth itself is affected in the long term. I am not saying this happens in the next 50 years, or even the next 100 years, but I am talking the next thousand years. We are already saying that the world population would start declining and may go down to 7 billion by 2300 with continuing decline of fertility rates. This decline could be faster if people's well-being is affected by growing dissatisfaction of being unable to get a living wage regardless of being well educated etc. This concern is a directly a second order effect of Concerns #1 and Concern #3.
Now having the laws listed and out of the way, let's review each of these laws and how they would apply and why they are needed in the first place.
#1 The AI Identity Law AI reaching super intelligence and taking over the planet requires AI to have an identity or a sense of self. By taking it away, AI can no longer act in its own self-interest. Self Interest requires identity and we are saying that AI cannot assign a foundational persistent memory of self. When we take away identity, then there is no question of self-preservation anymore and therefore should alleviate concern #2 completely. Also with Law #2, AI cannot make a moral decision to favor who lives and who dies.
This law exists because one of the most fundamental tenets of a Human being is the sense of self.
"cogito, ergo sum" or "I think, therefore I am"
If there is no 'I', then there is no concept of I think.
#2 The Moral law is meant primarily to ensure AI doesn't make moral or ethical choices that matter and affect humans. Imagine an AI that doles out punishment, judges whether to wage war, arrest someone, or decide if someone's bank loan is approved or rejected. It would be outrageous if there was no human in the loop for these cases. So, the second law is that a machine doesn't make moral choices of right and wrong. it can use data to present pros and cons or even recommendations, but it cannot by itself decide on a moral course of action. It can execute a course of action once a human agrees to the action to carry out.
A corollary to this law is that during the course of action if an AI needs to make a moral choice it needs to escalate and wait for the human. In case the human is unavailable, the AI will TAKE NO ACTION refuse to proceed further.
This law effectively handles concern #1, #2 and #4, by requiring a human in the loop. While cognitive work may no longer be prized, human judgement will still be of value requiring human expertise and opportunity to be present and thrive.
This law exists is because one of the most valuable things that is a proxy for wisdom in humans is good judgement. And in many situations, it is a question of moral judgement and it is a deeply human property. By making moral judgements in the firm realm of Humans, we keep Humans relevant.
#3 The Human First Principle This is not an AI Law, but more of a principle for other humans (people, corporations and governments) to use. The reason is that not all laws can be codified into something a machine can process and handle. An AI cannot tell the second order or third order effects of its actions, so asking it to worry about the welfare of humans (like in Asimov's laws) will simply be impossible. Instead, this needs to be handled at the overall societal level where human well-being is put front and center. That is the only way capital market pressures don't steam roller humans under the pretense of improved efficiencies.
While there is more to explore and write about this subject, introducing the laws of AI will go a long way in addressing the deep concerns around the world improving the adoption and creating a pathway to preserve and improve the Human condition.
PS 1 This is a philosophical framework and not a technical one.
PS 2: Before all the engineers start getting up in arms about how neural networks cannot really tell if it is making a moral choice, what I am saying is there should be systems that evaluate output as a moral choice. An AI should be able to inspect context and be able to tell (at a general human intelligence level) whether it is making moral choice. If it is not able to make that judgement, then it has no business running anything.
PS3: Yes, a human in the loop might not really improve human value and might become a rubber stamp, but rubber stamps keeps some humans around. Maybe not the highest purpose for a human (to rubber stamp decisions), but some purpose nonetheless.
PS4: While I am not expecting these laws to be come laws of AI built inside the machine, I am expecting something like this becomes codified in the society and cultural norms of men. To people who might say that these laws might not be applicable to bad actors, I say that reasonable restrictions on privacy / health etc lead to HIPAA or GDPR that 90% of the mainstream corporations have to obey now. We are worrying about the mainstream (not actors outside society at large)
PS5: So now that AI is doing cognitive work, what is left for Humans? It appears for now only the Emotional / Feeling frontier.
PS6: I am not Anti-AI, I am an engineer myself and love things that move the world forward. But I love and wonder about the future of humans even more.