The Ethics Dilemma: How Does AI Decide in an Emergency?

When a crash can no longer be avoided, software has to decide in advance whom it protects and whom it doesn't. We unpack the famous ethics dilemma of self-driving cars, the philosophical conflict behind it, and what it means for drivers and lawmakers today.

Why the ethics question matters to every EV shopper

Anyone shopping for an electric vehicle today eventually runs into the topic of self-driving technology. Most new EVs already ship with driver-assistance systems that take over parts of the driving task, and the road toward higher SAE levels of autonomy is technically well mapped out. But the more decision-making power software gets behind the wheel, the more urgent a question becomes that, at first glance, has nothing to do with engineering and everything to do with philosophy: how should a car decide when a crash simply cannot be avoided?

This isn't an abstract thought experiment reserved for ethics seminars. It directly shapes the rules that will govern how AI learns to drive in real-world traffic, who is accountable for that decision, and whether there's even such a thing as a "correct" answer. One of the clearest illustrations of this problem comes from ethics researcher Patrick Lin, whose thinking was turned into a short explainer video by TED-Ed. It's a great starting point for understanding the dilemma before we look at how regulators, including those in Germany, are trying to handle it.

Source: TED-Ed: The ethical dilemma of self-driving cars - Patrick Lin

The trolley problem, now on four wheels

Before diving into worst-case scenarios, Lin makes an important point worth keeping in mind: on average, self-driving cars are expected to be significantly safer and more efficient than human drivers. The safety statistics on self-driving cars already show that most crashes today stem from human error that a consistently programmed AI simply wouldn't commit: distraction, fatigue, or ignoring traffic rules. Still, an uncomfortable truth remains: eliminating every crash is impossible. Sooner or later, even the safest system will face a situation where harm can no longer be prevented, only redistributed.

The thought experiment: swerve or go straight?

That's exactly where the video's central example comes in, a modern spin on philosophy's classic trolley problem. A vehicle suddenly finds itself boxed in with only three possible responses, each carrying a different risk:

• Continue straight and hit an object falling into the road
• Swerve left and clip an SUV
• Swerve right and strike a motorcyclist

A human driver would react instinctively in that split second, with no time for conscious weighing of options. For a self-driving AI, it's different: it doesn't make that call in the moment, it makes it long beforehand, in the form of a fixed rule baked into code. That difference is what makes the issue so charged. A spontaneous reflex is hard to hold anyone morally responsible for, but a pre-programmed rule is another matter entirely, since it was deliberately designed to sacrifice a particular person in a particular scenario.

Utilitarianism versus duty-based ethics

This is where the video lays out the philosophical conflict sitting behind every possible programming rule. On one side is the utilitarian approach, which tries to minimize overall harm, in other words, cause the fewest injuries or deaths possible, even if that means actively steering toward a particular person. On the other side is the deontological approach, which holds that certain actions are simply wrong, such as actively harming a bystander, regardless of how good the outcome looks on paper.

Each position leads to a very different piece of code. A utilitarian AI would probably avoid the motorcyclist in this scenario, since that's where injury risk is highest, even if it means actively changing course. A strictly duty-based logic would lean toward staying in its lane rather than actively steering into another vehicle, even if the outcome turns out objectively worse. There's no consensus on which approach is "right," and that's precisely what makes writing these decision formulas one of the hardest problems in the entire field of machine learning for vehicles.

Who's liable when the AI makes the call?

That ethical uncertainty immediately raises a legal one: who bears responsibility when a pre-programmed decision leads to harm? The video deliberately leaves this open, and it potentially touches several parties at once:

• The manufacturer that put the vehicle on the road
• The engineers who wrote the specific decision-making code
• The owner who uses the system in everyday driving
• Regulators, who have yet to set unified rules

Engineers are left with an unusual task: translating abstract ethical standards that philosophers have argued over for decades into concrete, executable code. That raises very practical value judgments, such as whether occupants deserve more protection than pedestrians, or whether the number of people affected matters more than how much fault they bear for the danger in the first place. For a look at how German law is already tackling these questions, see our guide on who is liable in a self-driving car accident.

Looking ahead: culture, regulation, and the future of vehicle ethics

The thought experiment in the video is deliberately extreme, but it points to a much bigger societal debate that has been building for years: can there even be one universally correct ethical answer for autonomous vehicles, or do standards inevitably vary by culture and region?

Cultural differences: the MIT Moral Machine

Researchers at MIT explored exactly this question with a global online study called the Moral Machine, which drew millions of participants from more than 200 countries and territories. The result: preferences vary noticeably across cultures, for instance how strongly younger people should be protected relative to older people, or how much weight should be given to obeying traffic laws in the first place. That's a genuine headache for manufacturers: an ethics formula considered fair in one country can be seen as unacceptable in another. For more on the technical side of how these systems get cleared for the road, see our background on how self-driving cars get approved, a process that, in Germany, runs through TÜV inspection bodies and the Federal Motor Transport Authority (KBA) and still barely accounts for these cultural differences.

Regulation in Germany and Europe

In Germany, this debate hasn't stayed purely academic. The country's Ethics Commission on Automated and Connected Driving has already published guidelines stating, among other things, that weighing human lives against each other based on personal traits like age or gender is impermissible. That rules out the most extreme utilitarian calculations, but it's far from answering every practical question raised by the thought experiment above. We go into more detail on how these principles feed into German traffic law in our guide to self-driving car laws in Germany. Work on unified standards is also underway at the European level, though it's moving considerably slower than the pace of the underlying technology.

What owners and buyers should know today

For most EV buyers, this ethics dilemma isn't yet a daily concern, since fully autonomous Level 4 or Level 5 systems remain rare on public roads almost everywhere. Still, it's worth understanding how the debate is likely to unfold:

1. Manufacturers will increasingly need to disclose the principles their systems follow in edge-case scenarios
2. Regulators are working toward binding minimum standards to prevent a pure market free-for-all
3. Insurers and owners will have to engage more seriously with liability questions once Level 4 vehicles reach mass production
4. The debate will increasingly intersect with the question of when true full autonomy actually arrives

If you're curious about the timeline for that shift, our guide on when Level 5 autonomy will realistically arrive goes deeper into the outlook. One angle that often gets lost in the ethics debate but matters a lot to our readers: the vast majority of autonomous vehicles being developed today are electric, partly because sensors and onboard computing integrate more easily with an electric drivetrain. If you're already getting familiar with charging an electric vehicle, you're indirectly preparing for the next generation of cars that will actually have to put these ethical decisions into practice.

In the end, the ethics dilemma of self-driving cars is a striking example of how tightly technology, law, and philosophy are intertwined in the future of mobility. A perfect, universally accepted solution probably doesn't exist, but a transparent, socially agreed-upon, and legally binding answer is exactly what's needed for autonomous vehicles to earn lasting public trust.

Additional Video

While the classic trolley-problem framing lays out the philosophical puzzle, this NBC News explainer digs into how real engineers, regulators, and lawmakers are actually grappling with self-driving car ethics today.

Source: NBC News – The Complicated Ethics Of Self-Driving Cars | Think | NBC News