howsteepisit

A thought experiment is not a brainstorming session. They are tightly controlled to investigate behaviors or opinions based on a rigid set of option to force thinking to a one way or the other type of thoughts. Introducing your sense of reality or other choices invalidates the exploration of whatever is being investigated. While its fun to brainstorm and make up your own rules, its not a thought experiment.

jeichelberg87

I would prefer a believable scenario (as I do not believe a word from the now imprisoned Oregon driver) but given these circumstances, I choose "C." Swerve left into the vacated lane.

Forced to live with the choices at hand, I stay in my lane. I did not commit the offense.

Maelochs

A thought experiment which doesn't represent reality doesn't produce meaningful results. If people want to play fantasy games, they can .... but if I see the "thought experiment" as more of an "ill-thought-out experiment" then i leave it for others.

howsteepisit

No model can reflect reality. The best it can do is to isolate critical factors for evaluation. The real world is a complex place and consists of infinite variables.

Maelochs

A thought experiment doesn't have to replicate reality.

For instance---you are taking a class which you paid for but some students did not. You see a student helping another during a test----clearly cheating. Do you turn in the cheater, do you tell the cheater to stop but take no further action, do you demand some free right answers yourself? if the cheater is a friend .... if the person getting help is your friend ... if the person giving help got free admission ... if the person getting help got free admission ....

If the description and options are sufficiently wide-ranging and specific, people can pick a choice comfortably. if the scenario is unrealistic or the options are not sufficient (do not include obviously real and reasonable courses of action which you would take, while all those offered are not what you would really do) then the thought experiment fails.

IMO.

howsteepisit

A thought experiment that allows too broad of range of choices tends to not be granular enough to yield meaningful results. The first thing one needs to know is what question is being asked or what hypothesis tested. The experiment must then be designed in a way to test that question. For example, if the question is do people value a vulnerable user life more than multiple people's lives, allowing a reality-based set of answers cannot answer that question because too much noise is introduced. Thus to find the preference the answers must limit the choices severely.

Unless you are chatting on an internet forum, then answers can cover the range of both possible and impossible solutions, and per the internet standard, nothing much is learned nor determined. Who'd a thunk it.

Maelochs

I contend that if the options offered are Not sufficiently realistic they don't test anything but the reader's imagination. All of us can pick A.) Kill one to save five or B.) Let one live and five die, but in reality I doubt any of us know what we would do.

In most situations requiring quick thinking, difficult choices, and important actions/ reactions I have seen myself and others make enough unpredictable choices that nowadays I pretty much discount all that as bluster and self-delusion.

Just ask any 17-25-year-old male what he would do in any situation regarding a dominant male or an attractive female ... when sitting around partying with friends there will be one set of answers, but when faced with having to take life-altering action .....

Most "thought experiments" are "What I think I should think" experiments. I doubt most of them yield "meaningful" results.

IMO.

howsteepisit

Thought ==> Imagination!

Darth Lefty 

So now we are reduced to arguing if philosophy has value?

Rollfast

Oh Foo.

Maelochs

Not all thought experiments are necessarily valuable, or scientific or philosophical. "Know Thyself" can be said to the the fundamental phrase, the necessary starting point, for all philosophical discussion ... and as I stated above, I believe that most "thought experiments" do Not truly reveal a person's character.

canklecat

That's a valid point, regarding the thought experiment to evaluate ethics vs the problem solving experiment approach I chose to take. I won't argue with that.

However, I also pondered whether the Moral Machine might have another underlying, possibly more interesting agenda -- to identify imaginative thinkers. Or cheaters. Either works in this case.

That may seem far-fetched, especially given the short batch of sample scenarios presented per session. A much longer session would be needed, with some scenarios repeated a few times with minor variations.

Basically, that's what most employment application tests are, especially for retail and warehouse jobs -- tests to identify cheaters and thieves. The test designers know most applicants will try to game the test and answer in ways they think the employer wants to hear. So the same questions are not only rephrased many times, but so are the order and sequence of scenarios leading up to the basic "Are you gonna steal from us?" question. The tests begin with questions like "Would you take a pen home from work?" and gradually move toward "Would you report a co-worker whom you saw taking a ninety-nine cent loaf of bread home a few days before payday, knowing they were out of money, and that the employee planned to pay for or replace the bread after payday?" The sequencing of questions is designed to gradually break down a normal person's facade and get into their most likely real world responses to moral and ethical dilemmas. The testers aren't interested in whether you might offer to buy the bread for the hungry person, or whether the hungry person intended to repay the employer for the bread after payday. They only want to know whether you would ever steal or condone theft, regardless of the complicating scenarios.

That approach is too outdated for developing AI. Before people can be persuaded to accept self-driving cars as a standard alternative to human drivers, they'll want to be reassured the AI can not only react quicker than humans, and do so without unreliable emotions, but will also "think" like a rational human. And the vehicle should be equipped for maximum occupant safety.

In other words, it's blurring the concept of theft in the employment application scenario. Suppose an employment application test also included a sequence of questions such as:

• Have you ever received an income tax refund?
• Did you know the income tax refund is a return of money you overpaid to the government, which the government borrowed for a year and then returned to you without paying interest to you, the lender?
• Have you ever been late in paying taxes?
• Did the IRS waive the late penalties and interest charges?
• Have you ever begun work a few minutes early or worked a few minutes late, off the clock or without being paid or otherwise compensated?
• Have you ever calculated the total hours you've worked without pay or compensation?
• Have you ever been docked pay or warned that you might be fired for being late to work?
• Now... will you steal from us?

Employment application tests don't include such scenarios because it blurs line that they'd rather keep sharply focused.

But gaining acceptance of self-driving cars, particularly in the US, will require confronting and dealing with such complications and gray areas.

For example, a more realistic scenario might go something like this: Given the usual mix of vehicle occupants and pedestrians presented elsewhere in the test, and knowing that striking the pedestrians would probably kill the pedestrians while crashing the vehicle into the barrier would probably result in only minor injuries to the occupants, which would you choose?

Then rephrase the same scenario later in the test, this time with the test-taker's family in the vehicle, while all but one of the vulnerable pedestrians was a reprehensible character. If mommy is in the vehicle along with her kidlets, and she's a helicopter parent and anti-vaxxer, and she just made the last payment on her car a few months ago and now carries only liability insurance rather than full coverage, and she knows that of the six people in the crosswalk one is her pediatrician while the other five are a homeless woman, a drug dealer, a couple of addicts and a mime, well... I'm betting mommy will be shopping for a new pediatrician soon.

The Moral Machine does allow visitors to design their own test scenarios. I might give that a try and see what happens.

mconlonx

Would rather not confront a moral quandary where one might identify with one of the actors...

Thanks. "Right but dead"

Were it me and I had any chance to act by thought rather than sheer reflex, the results of the Morality Machine indicate I would take out the cyclist -- 1 life vs. at least 3.

jefnvk

Fully dependent on the exact situation. Head on, both cars doing 60MPH? That "possibly" moves into "almost definitely at least one death" to me, even with modern safety equipment. 20MPH? Scenario changes significantly, as it does if a sideswipe is a realistic option, as it does if the equation is a dually 3500 truck v a 30 year old Fiesta.

As to my actual results:
Saving more lives: Matters a lot
Protecting passengers: Matters a lot
Upholding the law: Doesn't matter as much, closer to neutral
Avoiding Intervention: Matters a lot
Gender Preference: Solidly Male
Species: Humans, exclusively
Age preference: Towards young
Fitness preference: Neutral
Social preference: Neutral

Perhaps I'm too closely involved, as my company does various detection systems that I've gotten to play with, but some key notes. Computers don't determine fat, skinny, rich, robbers, pregnant, or anything else, and make determinations based off of that. Computers won't make a determination if it is a jaywalker or someone crossing legally, just like it is not legal to run over jaywalkers now.

My preference, as such came down to two criteria: when loss of life was the same, keeping the vehicle where it was supposed to be, in lane. This overly simplistic scenario does not consider any other traffic, but that becomes a huge factor when operating in a mixed vehicle environment with no ad-hoc communication about intent. Secondly, if swerving resulted in less or no human life loss, it was more preferable, as far as the limited scenario was concerned.

And finally, I agree with whomever said it above: such moral dilemmas are pointless if the car isn't preparing to stop at a moment's notice WHEN APPROACHING AN INTERSECTION.

Darth Lefty 

You should read WHAT CAN YOU SAY ABOUT CHOCOLATE COVERED MANHOLE COVERS? - All The Myriad Ways

Darth Lefty 

I still want to know what any of you think about my idea that the self-driving car's programmer has an obligation to save his customers, rather than the strangers in their path.

Maelochs

Seems clear. The supposition is that every individual, including the AI driving the Google-Uber, will try to keep the passengers alive. One cannot make decisions for the other vehicles, one cannot make assumptions that the operators of the othehr vehicles would do anything except try to survive ... so the AI car needs to use the same rationale.

mconlonx

It makes sense on the self-driving car sales and consumer end of things.

That would have the car hitting a crowd of pedestrians rather than veering into a barrier at speed.

Morally, considering the lives involved, people would have issue in general with such a system. Politically, it would be completely untenable.

And since driving is a privilege while walking is a right, there might actually be legal issues which would preclude such a control system.

genec

All this pondering... one has to ask... how is a self driving car, determined to save it's passenger, any different from a human driven car?

Especially bearing in mind that a driving human is going to react, and not have time to evaluate.

Seattle Forrest

Sometimes people get shot or impaled through the head and live. Look up Phinneas Gage. That doesn't mean you should go get into a head on collision.

jeichelberg87

I do not think it is a question of obligation, it is more related to risk avoidance and liability In this case, I think the CEO/COB of Google can simply look at most of the outcomes of pedestrian/cyclist/motor vehicle collisions and come to this conclusion: When it comes to automobiles in this country, the deaths and injuries inflicted on cyclists and pedestrians by the operators is a non factor. There will be, for most instances, no investigation and merely a statement: "no criminality found." Thoughts of CEO/COB of Google - "Therefore, we do not need to worry if our self-driving car crashes uncontrollably on the sidewalk, killing those 20 school kids. We can always blame it on the programmer/maintenance people anyway."

Hey, all must bow down to the almighty dollar.

Seattle Forrest

Just to make sure I'm clear on this.

I'm driving fast enough that the force of a crash will probably kill everybody. And the cyclist is pacing me at breakneck speed while I consider my options - I don't just pass him quickly. Also my brakes and horn don't work.

Is that right?

CliffordK

Can I reprogram the Kobayashi Maru?

I choose not to put myself into fictional no-win situations.

canklecat

Why not? I did.

Ditto. That's why real life comes with cheat codes.

jon c. 

I ran through it twice following the same pattern and the resultant values were quite different.