Moral dilemmas have been safely pinned to philosophy for more than 6,000 years. What do I mean by the expression »safely pinned«? As an issue in philosophy, specifically moral philosophy, it remains valid and crucial because it was never resolved. If it were resolved, it would cease to persist as a topic of philosophy. Philosophy’s primary value is to keep us engaged with questions we must decide upon in life in moments of truth. The narrative we develop in dialogue with philosophy helps us choose better in moments of truth.

With “thinking machines,” at least one moral dilemma has escaped the safe domain of philosophy. At the moment, all producers that develop self-driving cars face the moral question that underlies almost every moral dilemma in general:

Would you decide to kill one person (a child, for instance) for sure if the only other option were to most probably (but not necessarily) kill dozens?

The shift of the question from personal experience and philosophy to technology places a solution algorithm in a machine as an agent, so there is no place for questioning anymore. A human driving a self-driven car ceases to be an agent in the crucial moral dilemma. What does that bring?

Morality and moments of truth

First of all, how does human agency happen? With the help of introspection, but more significantly, with the aid of neuroscience, cognitive philosophy, evolutionary theory, and other disciplines investigating human behavior concerning neural activities, we are confident that moral dilemmas are solved a priori, pre-rationally. In moments of truth, we as agents act solely driven by our limbic system. We decide based on the vast residuum of 4 billion years of evolution on Earth, our own experiences, and last but not least, on those cognitive solutions that were repeatedly imprinted in our limbic system. That is why we can procrastinate until the last moment when we somehow trust that our intuition will “decide”. We »know« that our ratio is too slow for making decisions in moments of truth.

Please note that I employ the concept of moments of truth in my branding theory for a specific reason. Subconsciousness, limbic system operation, takes over agency in all moments of truth. It is no different when we face a critical situation while driving a car than when we decide which brand to buy in a supermarket.

With self-driving cars, a strange shift happens. A rational being constructs the car’s agency. Cars have to »decide« upon the pre-designed algorithm. There is no place for the limbic system anymore, for something so crucial for the evolution of humans and the best part of the animal kingdom.

Knowing nothing about moral algorithms, it should be clear that they are impossible. Or better: until machines develop subconsciousness and all procedures that form the limbic system, until they engage in moral philosophy, they will be unable to act as moral agents, or at least to pretend to act as moral agents.

But then let us check how some self-driving cars address the moral dilemma.

Thinking driving machines

In 2020, Mercedes-Benz made a strategic decision to scale back its ambitions for fully autonomous passenger vehicles. The company shifted its focus toward developing advanced driver-assistance systems (ADAS) and prioritized autonomous technology for commercial trucks, where the business case appeared more immediate and viable. I started with this post, then used this reference to my scepticism about self-driving cars. Yet, in 2025, Mercedes-Benz announced level four automated driving testing.  

Does that mean that my scepticism is unwarranted?

No, my scepticism did not vanish. Yet, I have to be more precise about the direction of my scepticism.

I firmly believe in the fast development of all kinds of digital technical improvements based on AI, including self-driving cars. I claim that there is a theoretically and practically unbreakable barrier for any AI. This barrier is easiest to explain with the example of self-driving cars.

Stakeholder analysis for AI

Who are AI stakeholders?

This question seems absurd. Yet, this is the reason why it should be taken seriously. 

While the main stakeholders of a self-driving car are the driver and the insurance company, other stakeholders also exist. Who is liable if something happens? Self-driving cars increasingly shift their responsibility to AI and less to the driver, who should be responsible. While it is true that a driver is finally responsible, it was the case already before the now known AI instances that if it is proved that the accident was caused by a failure of a tested part of a car, the driver was not penalized for the one who produced the malfunction. Yet, even if the brakes failed, it was hard to delineate who made a mistake, a driver or a car producer. If AI, which controls all parts of the cat and its surroundings, fails, then why blame a driver who used to be in control and is no longer? He paid for the system, so the AI system should be responsible as much as the brakes if they fail, not the driver. 

The answer to the above dilemma is not legal, even if it seems like a legal question. Pushing it into the legal domain means avoiding the real question behind it. 

Such a question becomes even clearer if we consider the safety of passengers and the rest. What happens when the situation presents no threat to passengers but to two pedestrians, one pedestrian and one cyclist, one driver in another car, and one pedestrian, and so on?

What should be self-evident is that the AI algorithm should consider all possible situations. “The machine” should be able to decide upon a weighted evaluation of an enormously large number of situations. Should it kill a 50-year-old adult before 50 a 50-year-old woman? Is an 80-year-old woman worth less than 20 a 20-year-old student? What if that student is seriously ill, so she will die in the next six months? Is 11 an 11-year-old child worth more or less than 12 a 12-year-old child? What about a 30 years old couple in comparison to 18 years old junkie?

Let us consider a dilemma that was exposed on German TV. Is Lars Koch guilty (or not) of killing 164 passengers in an aircraft while avoiding the only other solution to kill 70.000 people at the Allianz Arena watching a football game between Germany and England? Germany decided that he was not guilty. But this dilemma was relatively easy since passengers would die anyway.

What is quite apparent is that a moral dilemma is unsolvable through technology as long as the technology works on consciously devised algorithms. Even if such machines can learn, this does not solve the issue. If it were only to upgrade the base of possible situations, it would still rest on weights that would come from one moral dilemma solution devised by humans. Even if weights were adjusted, some other code based on which weights would be self-adjusted should have been consciously developed beforehand and implanted in such machines.

Moral machines?

To overcome such a gap, the “thinking machine” and the “rational machine” should evolve into a moral machine. That means that such a machine should be able to understand the moral dilemma and feel it. And we all know what makes us (humans) feel a moral dilemma: the fear of death. So, machines that could run self-driving cars should not only be able to die but also be conscious of and afraid of that option. Such a machine would be a negative zombie. It would not look like a human, but would feel like a human.

It is thus quite clear that such machines could only evolve if matter (machine) can really become alive and conscious through emergence, which is an underlying principle of evolution as we understand it at the present moment.

Should that not be possible, then we have an unsolvable problem in the development of self-driving cars and the present concept of evolution and current theories of how consciousness emerges from brain activity, for instance. If the emergence of consciousness and life rests on vast enough complexity only (materialistic solution of Darwin, Dawkins, Dennett, etc.), then such thinking and feeling machines are possible. If not, then the car industry, materialism, and evolutionary theory have an unsolvable problem. Then only God can implant the solution into a machine as He did according to creationist cosmogony.