Autonomous cars face the trolley problem: Who should they protect in crashes?

Potential casualties in unavoidable accident scenarios where autonomous systems must choose between protecting occupants or minimizing broader harm to road users.
A small human sitting with me
Magno's description of how the autonomous system negotiated a narrow street by reading and responding to another driver's intentions.

O veículo que não precisa de condutor já circula nas estradas do mundo, mas carrega consigo uma questão que a velocidade da tecnologia ainda não conseguiu ultrapassar: quando o acidente é inevitável, quem decide quem sobrevive? O dilema do comboio, durante décadas confinado às salas de filosofia, instalou-se agora no asfalto, transferindo a responsabilidade moral do condutor humano para engenheiros, fabricantes e reguladores. A sociedade avança para uma era de mobilidade autónoma sem ter ainda respondido à pergunta mais difícil — não como fazer o carro conduzir, mas como fazê-lo escolher.

  • Em milissegundos, um sistema autónomo pode ter de decidir entre proteger o seu ocupante, desviar-se para um peão ou colidir com o trânsito em sentido contrário — e alguém programou essa decisão.
  • A responsabilidade moral, que numa travagem humana é lida como reflexo e perdoada como tal, torna-se num algoritmo um ato calculado que exige justificação pública e legal.
  • A China lidera a adoção prática da condução autónoma, enquanto o Ocidente, com exceção da Tesla, permanece despreparado e a regulação europeia corre atrás de uma tecnologia que já não espera.
  • Apenas 10% dos consumidores desejam veículos autónomos para uso pessoal, mas os robo-táxis e serviços comerciais em zonas controladas prometem ser o verdadeiro campo de entrada desta tecnologia.
  • Portugal aprovou em abril o primeiro quadro legal específico para testes de veículos autónomos em vias públicas, sinalizando que a infraestrutura regulatória começa, finalmente, a acompanhar a estrada.

Antes de cedermos o volante por completo, há uma pergunta que não tem resposta limpa. Imagine um veículo autónomo numa estrada de dois sentidos quando a carga de um camião se espalha subitamente por ambas as faixas. O sistema tem milissegundos: seguir em frente e embater nos destroços, desviar para o passeio arriscando atingir um peão, ou invadir o sentido contrário. O dilema do comboio — clássico exercício de filosofia moral — chegou ao asfalto.

Foi sobre esta tensão que o jornalista tecnológico Sérgio Magno e o editor Diogo Teixeira debateram numa sessão do Auto Talks, gravada no ECAR Show. Magno foi direto: proteger o ocupante não pode ser a prioridade máxima. O sistema deve minimizar o dano para todos os utilizadores da estrada. A distinção é fundamental — o desvio instintivo de um humano é lido pela sociedade como reflexo; a decisão calculada de um algoritmo exige justificação. Quando uma máquina escolhe, alguém a programou para isso, e essa responsabilidade não desaparece.

A Europa avança com o AI Act, mas a regulação move-se mais devagar do que a tecnologia. A China está mais à frente: os sistemas autónomos chineses são programados para minimizar o impacto energético dentro da própria faixa, recusando criar novos perigos ou invadir o sentido contrário. Magno viveu isso em primeira mão durante uma hora de condução autónoma em Pequim, a bordo de um XPeng. O momento mais revelador não foi nenhuma manobra dramática, mas sim uma rua estreita onde dois veículos não cabiam lado a lado: o carro recuou, o outro condutor cedeu, e o sistema avançou. "Exatamente como um humano faria", descreveu Magno. A sensação foi a de "um pequeno humano sentado comigo".

Mesmo assim, o entusiasmo não se traduz em adoção rápida. Apenas 10% dos consumidores querem veículos autónomos nos seus carros pessoais. Magno antecipa que a entrada real no mercado virá pelos serviços comerciais — robo-táxis, plataformas de transporte partilhado, zonas controladas. Em Portugal, o governo aprovovou em abril o primeiro quadro legal para testes em vias públicas, com requisitos de segurança e cibersegurança supervisionados pelo IMT. A infraestrutura está a ser construída. A tecnologia avança. Mas a pergunta mais difícil — quem decide o que o carro faz quando algo corre mal — continua sem resposta.

The car of tomorrow won't have a steering wheel. But before we hand over control entirely, there's a question that doesn't have a clean answer.

Consider this: you're riding in an autonomous vehicle on a two-lane road when a truck's cargo suddenly spills across both directions. The system has milliseconds to decide. Stay in your lane and hit the debris head-on. Swerve onto the shoulder and risk hitting a pedestrian. Cross into oncoming traffic and collide with a car coming the other way. This is the trolley problem—the classic moral dilemma that lived in philosophy seminars for decades—now moving directly onto asphalt.

Sérgio Magno, a technology journalist at Público, and Diogo Teixeira, publisher of Razão Automóvel, tackled this question during a recent Auto Talks session recorded at the ECAR Show. The stakes are not abstract. When an autonomous system makes a choice, someone programmed it. That responsibility—to the manufacturer, the regulator, society itself—cannot disappear. Magno was clear on one point: protecting the car's occupant cannot be the highest priority. The system must instead minimize harm across all road users. This distinction matters enormously. A human driver's instinctive swerve is read differently by society than an algorithm's calculated decision. One is forgiven as reflex. The other demands justification.

Europe's AI Act has begun addressing these questions, but regulation moves slower than the technology it's meant to govern. "In the West, aside from Tesla, we're not very well prepared," Magno noted. China is further ahead. Autonomous driving still follows UNECE guidelines, but those rules have gradually approved systems capable of lane changes and other maneuvers without active human intervention. When artificial intelligence faces the same dilemma, the answer is more straightforward: the algorithm minimizes impact energy within its own lane, refusing to create new dangers or cross into opposing traffic.

AI was supposed to solve autonomous driving's central problem—handling unpredictable situations, human behavior, the ambiguity of real traffic. Magno experienced this firsthand during an hour-long drive through Beijing in an XPeng with autonomous capabilities. The car navigated city streets without any input from pedals or steering wheel. The most revealing moment wasn't a dramatic maneuver. It was mundane: a narrow street where two vehicles couldn't pass side by side. The car encountered a van coming the opposite direction and began reversing. When the van's driver also reversed, yielding right of way, the system changed course and moved forward. "Exactly as a human would," Magno described it. The sensation was of "a small human sitting with me."

Yet enthusiasm doesn't translate to rapid adoption. Studies show only 10 percent of consumers want autonomous vehicles in their personal cars—a figure reflecting both the technology's cost and lingering distrust. Magno expects faster deployment through commercial services: robo-taxis, ride-sharing platforms, transport in controlled zones. "It's like calling an Uber and having it arrive and take you where you need to go," he said.

Portugal is laying groundwork for that future. In April, the government approved a decree establishing the first specific legal framework for testing autonomous vehicles on public roads, with safety and cybersecurity requirements overseen by the IMT. The infrastructure is being built. The technology is advancing. The harder question—who decides what the car should do when something goes wrong—remains unsettled.

The protection of the occupant cannot be the maximum priority. The priority must be creating the least harm for all road users.
— Sérgio Magno, technology journalist at Público
In the West, aside from Tesla, we're not very well prepared for this.
— Sérgio Magno
The Hearth Conversation Another angle on the story
Inventor

When the system has to choose between hitting a pedestrian or crashing head-on, who actually bears responsibility for that choice?

Model

That's the core tension. The driver used to bear it—their instinct, their reflexes. Now it's baked into code by engineers and approved by regulators. Everyone shares the weight, which means no one can escape it.

Inventor

But doesn't the algorithm need some principle to follow? It can't just freeze.

Model

Right. So the principle becomes: minimize harm to everyone on the road, not just the person inside the car. That's a shift from how we normally think about self-preservation.

Inventor

Is that actually what's being programmed, or is that what people say is being programmed?

Model

That's the gap Magno was pointing to. Europe talks about it. China is moving faster with actual systems. But there's no global standard yet. Each manufacturer could theoretically choose differently.

Inventor

So a Tesla might protect its occupant first, while a Chinese car protects the broader road?

Model

Potentially. Which is why regulation matters so much. Right now the technology is outpacing the rules that would make those choices consistent and transparent.

Inventor

Why is China ahead on this?

Model

They're willing to deploy and iterate. They're testing real autonomous systems in real cities. The West is still debating whether it's ethical to even ask the question.

Inventor

And the average person—do they want this?

Model

Only about one in ten consumers is interested in autonomous vehicles for personal use. The cost is part of it, but so is trust. People don't yet believe the system will make better choices than they would.

Contact Us FAQ