So your issue is your secretary doing it's job poorly?

First, there is a way to tell it to not do that. With Google Now, you simply tap the menu and say "No more notification like this". With the assistant, you will probably be able to ask directly.

Second, let's be honest, humans fail pretty often too, so that's just a weak argument.

Lastly, I think it's unfair to dismiss a new technology just because it could maybe fail, without having even tried it.

How about if the system is working exceptionally well, you're a depressed person, and the next ad you see is auctioned off between a therapist, a pharma company, and a noose supply store in the 100ms it takes to render your MyFaceGram profile?

The awful success cases are far more interesting than the awful failure cases.

I have no problem with ads for therapists or pharma companies competing for advertising space in front of me because they have algorithmically determined that I am a qualified lead. That actually sounds great from a mental health perspective.

Your noose example is pretty contrived, however.

Obviously the first two aren't the problematic ones. The issue is that an algorithm wouldn't know what distinguishes those from the third, obviously.

How about sleeping pills? Opiates? Local extortionist cult?

I think that algorithms, and AI specifically, are perfectly able learn what distinguishes those. Maybe even better than someone who might not be in their best state of mind.

The handwaviness is telling. Why would an algorithm or its creators even care about the difference? The highest bidder is the highest bidder.

Because the whole of Google's ad business stands on people wanting to click on the ads shown, and buy the products offered through them. That's why they spend resources on detecting misleading or fraudulent ads, which by your reasoning they wouldn't care about as long as they paid. PR is very important for this business to be sustainable: If the goal was for every user to click through one ad, and then never again, that might not even pay one engineer's salary.

What's misleading or fraudulent about those ads? Maybe you mean "morally reprehensible," in which case I ask where you draw the line between the morally reprehensible (auctioning off the method of suicide to a depressed person) and the morally questionable (say, auctioning off the final bankrupting car purchase to a financially irresponsible person)?

Detecting misleading and fraudulent ads is just an example of things they wouldn't spend resources on, if following your reasoning of "short-term money is the only thing they care about."

There's not only the "morally reprehensible" metric ("Don't be evil"); there's also the "absolute PR catastrophe" metric that printing such an ad for a rope would mean.

> So your issue is your secretary doing it's job poorly?

I think the real issue is the casual deception which you just fell for: It isn't "your" electronic secretary, and the thing it just did might actually be a "good job" from the perspective of those who control it.