Thank you for the question.

I'd like to know who would assume the authority to put people in jeopardy where an artificial intelligence is confronted with the necessary choice to take a human life? That's the salient ethical question in my opinion.

Peter Thomas Byrne Gregory If you replace a human with that Artificial Intelligence, the Reaction Time of humans, is estimated 1 second. in many cases, the event might happen in a fraction of that 1 second. So, what will you have ? A driver with a frozen, fixed scared face , doing nothing, not being able to decide, running into 20 people, without being able to swerve to the other side, where there was only one cyclist, which, if he was able to do the Risk Analysis, based on Bayesian, Maximum or precaution principle, had a lower probability of being killed. So we contend ourselves with the inability of the driver to decide, ignoring all of his personal shortcomings, bad habits etc. But all seems moral and ethical ?

who would assume the authority to put people in jeopardy by allowing such a driver to drive in the first place ?

Yes I agree, there are bad humans who drive cars, design algorithms and write code, along with many other things. I assume you are not suggesting that robot makers are a cut above the rest of humanity and can make these determinations and dispense these permissions for the purpose of protecting the rest of us from bad drivers because we all hate (no argument there) bad drivers. How about killer robots and the authors of such protocols; call to:{contingent factors when it is necessary to choose whose death}

Is this an omelette making egg breaking argument because if so I find they tend to end by generalising about the quality of human types and their utility to other human types. Not a good outcome from my reading of history but who knows what the future may bring. Maybe we could build a robot better than human and all mass suicide for being slower dimmer lesser creatures. Who’d argue with that?

Peter Thomas Byrne Gregory I wouldn't put driving and writing a code in the same argument level, even though both are susceptible to malicious utility. I think Driving per se, is rather an act one does, with a promised responsibility, but not necessarily an output like income, award or significant result, but simply displacing more simple than walking between two ends. so all of his bad habits,inability etc will disqualify him to do it in the predicted manner or avoid ALL of the possible accidents.

instead, what does form in people's minds when they face a problem? normally a "Startup idea", how to do it better? so they code, make machines, tools. normally there is no evil plan behind it. someone else might later aprofit, no discussion about it.

after the industrial revolution, people have suicided because they had lost their jobs, but there was no complete lack of jobs, they had lack of tools to look or displace for similar jobs elsewhere, they had their own constraints that stopped them, not the machiens. either, if there was not a machine that was simplifying his job, maybe his ex-company would have not even be created.

Same might be true for cars. It's just intervention of machines in a more sophisticated circle of actions of human beings, rather than reciprocating, simple and without decision making

I would say that the trolley problem - or something recognisably similar to it - does arise for those who are designing algorithms for how autonomous vehicles should behave in situations where an accident is unavoidable. Should it swerve to minimise casualties - even if this sacrifices people (e.g. on the pavement) who otherwise would not be at risk? Or should it brake but not swerve, even if this means a greater number of casualties? There are also lots of other complications, not present in the original trolley problem - e.g., does the occupant of the vehicle - probably its owner - have a special status, or could they potentially be sacrificed too to save more lives? I have taught this in ethics seminars for engineering students.

The smartphone like the three laws of robotics was an idea born from science fiction but, unlike the smartphone, the three laws have been relegated back into the world of fantasy in order to make way for the new expression of authority which appears to reify acceptability out of another form of fiction. First Law: A robot may not injure a human...

So this doesn’t mean a robot can’t “kill” a human, only that they can’t “injure” a human right? I think the lawyers couldn’t stretch this one such that it didn’t appear nonsensical and so the three laws had to go. Robots marketed into the life world are programmed to kill people. Now there are no such prohibitions as those contained in the laws of robotics, just a new fiction.

Peter Thomas Byrne Gregory The Utilitarian Ethics does not exist in the 3 laws of Robotics, and a robot does not kill in case it is given the ability, by more power or weight or any characteristic that humans might have less of or lack, for a greater good. But I don't see this different than any accident a worker might have when working by an automated machine.

also accidential killing is very different than intentional killing. Cause you are held responsible for that even, with 2 conditions: being in control, and knowing what they are doing.

Also : “Should” implies ambiguity. When you say someone “should” do something, you’re implying that there are exceptions.

“Can” is a statement of ability. A robot certainly can kill a human, that’s why the laws are necessary. And even a robot that’s properly following the three laws can kill a human due to accident or malfunction.

so now if I have managed to filter the robot ( industrial and military) dilemma from the trolley problem, I would go back to Thomson's argument : is killing always worse than letting be killed?

and how would we arrive to the point that whether we should consider the Trolley problem in AVs or not at all ?

We wrote an entire article about this question where we develop and validate our approach to autonomous driving ethics on the trolley problem:

https://www.researchgate.net/publication/350811853_Autonomous_Driving_Ethics_from_Trolley_Problem_to_Ethics_of_Risk

Minimum two tonnes of metal speeding down public highways directed by artificial intelligence based on erring human design is to my mind a cause of the trolley problem. To say a robot should be extended the same right to mistake as a human is a peculiar kind of anthropomorphism. Isn’t the point of these creations that they are better than us at what they do and therefore judged by those standards rather than compared to ours? I cannot see any justification in using humans as test subjects without their consent. One cannot claim that this risk is in any way ethical. Kant’s categorical imperative ought to apply where those who would dispense risk, first put themselves directly in the path of that risk. The point of departure for sentient creatures and automated supports (driverless cars, etcetera) is where humans have to deal with the particular merits which apply to emergent events. I would be fascinated to learn how this level of consciousness and so right to determine matters of life and death is to be coded onto semiconductors. In my opinion creative emergent behaviour is an ontological phenomenon that cannot be developed into a suitable mathematical formula in order for AI to foresee what is yet to occur.

While the current AI tests are non hypothetical, in that they have already caused actual loss of life, how should we determine the trolley hypothetical? One suggestion; design an algorithm that allows the AI unit to quickly recognise an authentic Hermes Birkin but, I like to think there is an order of procedure where if we fix the most pressing practical ethical problem – the intentional endangerment of innocent lives - what follows is much easier to envision.