[logicallee]

I believe it's listening based on an overwhelming amount of anecdotal evidence reported all over the Internet. (Where there's smoke, there's fire.) Including by intelligent, scientifically-minded people.

At the same time, it wouldn't take much at all for me to believe otherwise: a simple blinded test where some people leave their phones in a room, researchers come and talk about some subject (absent the owners), then leave without ever interacting with the owners of the phone.

There are two ways to do the next part: The owners of the phone 1 week later could be asked if they've seen any facebook ads on (topic). A control group would have to be compared (where researchers didn't speak on the topic.)

Alternatively, the subjects of the study could be given 5 random topics when commencing the study and asked to record any ads they've seen on any of the 5 topics. The researchers could then speak on 1 of the 5 topics (differing from person to person) and a statistical analysis could be performed.

A "proof" is a statistically valid correlation between what the researchers talked about out and what the blinded subjects reported seeing ads about.

A refutation is a lack of such a correlation. Easy, and it would convince me personally.

By the way based on the anecdotal evidence I strongly expect this study to conclude "facebook is listening."

-

Note: there are a few loose ends to take care of. The researchers who are tasked with speaking on a subject are more likely to google it on their own phones. (Having been exposed to the topic), and those are in geographic vicinity. There are other similar possible mechanisms.

Perhaps the best approach is if subjects' phones are in a sound-proof vault and the researchers' speech is either fed into it via speaker/microphone, or not done so, but the researchers do not know for any specific phone whether it is able to hear them. (Making the study "double-blind", as neither the subjects nor the researchers know whether the subjects' phones have heard anything on a subject.).

[momeunier]

I don't agree with your proposed way of studying this. You're trying to study a symptom and then conclude on the cause of the symptom. That makes no sense. If you want to prove that Facebook is listening, then prove that Facebook is listening. And don't try to prove that Facebook is listening and then targeting you with ads based on what you said. There are way too many levels of indirections that can trigger false positive for a vast number of reasons you don't control at all. What you are trying to do is replicate in large-scale displays of anecdotes but with a slightly more controlled environment. That won't prove anything since it will just be anecdotes and will again not sustain rational explanation by experts of Facebook ads mechanisms.

Without getting into the ads delivery part and the anecdotes, how would you prove that Facebook is listening? How would you prove that there is a set of information taken from your speech or your audible environment transferred to Facebook.

[logicallee]

I think you didn't read my proposed methodology carefully.

    Box     Researcher    | Owner of phone (outside room)
  [     ]~       x        |      o

The researchers speaks about some subjects, but not others. The ~ represents that in some cases the researcher's conversations are being fed into the box, and in others they are not. The box is otherwise soundproof, and inside is the phone being tested.

We can pick subjects such as:

   - #1 Adult incontinence
   - #2 Cat food
   - #3 Last-minute trip
   - #4 ..
   -    ..
   - #10

The test group is that the researchers' voices are being fed into the box. The control group is that researchers voices are NOT being fed into the box.

It is important in order to maintain double-blind environment that the researchers not hear whether they are being amplified into the box.

The results might potentially look like this:

https://imgur.com/a/y7852

Of course, I just made this up. (I imagine the subjective 1-5 scores being whether the given subject reports seeing such an advertisement, from 0 definitely not to 5 definitely yes.) I even made subject 3 unsure about topics 1 and 3 to mimic that humans are fallible. Likewise subject 2 does not really report any advertisements. (This is likely in the real world - for example subject 2 could be explicitly excluded by advertisers for some reason.)

The attached is the kind of graphs that I would expect based on dozens of scientifically-minded people trying them.

If these are the two graphs that we got, and if the test and control groups were truly randomized, what other explanation could you offer?

Of course, my proposed experiment is orders of magnitude more scientific than what people are doing with their n=1, unblinded personal experiments. But theirs has some validity also.

[momeunier]

Thanks for explaining further. That would actually be an interesting experiment. Who's going to run it?

[logicallee]

Not sure who would run it. Nobody really cares that much.

[superbrama]

I disagree with your feelings about the anecdotes given there is an official account that’s more likely. Why would fb come out officially saying no, knowing they’ll be uncovered eventually if the answer ever were yes?

Also, your techniques could reveal a false negative both in the existence of a bug in the particular build of fb code being tested, and that such a mechanism could have existed in previous versions but already removed.

[logicallee]

Likelihood has to be judged against the evidence. A lot of scientifically-minded people have tested it for themselves and found ads to appear when they spoke about subject explicitly to test whether ads would appear. I recall several specific cases. It's not quite scientific enough for me, but I'd expect a scientific study to have the same conclusion.

Just like if you do a scientific study of whether baked beans make you toot, yes, you will find that they do.

Although you state there is a high likelihood of a false negative, it would be a useful study for me and others in the more likely case that it matches the informal n=1 studies numerous technical people performed by themselves.

[superbrama]

Source?

[logicallee]

>Source?

[1] https://www.google.com/search?q=%22baked+beans+make+me+fart%...

[2] https://www.reddit.com/r/videos/comments/79i4cj/youtube_user...

[3] https://www.youtube.com/watch?v=jjbz1N4qBQA the host themselves did the same.

In the original comment of mine that you replied to, I suggested how to make the methodology even more scientific with a double-blind study. As far as anecodtal evidence this is going to be the best that you can do. I personally find it convincing. These are reasonable people testing things as scientifically as they can.

[superbrama]

Dude, cmon. I know this is hn and I’m supposed to put effort into crafting useful responses. But, how have those links convinced you that it must be the mic?

From the evidence in those videos, I’d rather be considering whether fb is part of a future AI continuum monitoring and influencing the past (from the future in which the AI exists) by messing with people testing this feature. Yep, cat is alive on Facebook/IG, days later. Microphones and big data/AI are but two slits through which photons can pass, in the experiment known as fb.

[logicallee]

What other kind of explanation do you expect? People have tested it by playing a Spanish soap opera on their TV for a few hours, leaving their phone in front of it to listen. They then started getting Spanish advertisements. I realize that there is a confirmation bias in that they are then actively looking for Spanish-language ads (to test their theory) but don't you think they would have noticed them at other times? They explicitly ran an experiment and got the verification.

In fact, some people were incredulous and re-ran the experiment thinking there's no way they would start getting Spanish ads: they did too. That's all it takes to convince me. I mean what other mechanism would get back to Facebook? As far as I know TV's don't report back which specific channel they're tuning into (all channels are streamed concurrently.)

I'm all for a more scientific test but at the moment I find the anecdotal evidence quite convincing. I just outlined how a scientific test might work here:

https://news.ycombinator.com/item?id=15782323

[_l8fa]

> it wouldn't take much at all for me to believe otherwise

...goes on to describe multi-week, double-blind trials involving human subjects.

[logicallee]

:D I get your point. The specific reason I said it "wouldn't take much at all" to convince me otherwise is that under a negative result, we do not know:

-> Was Facebook listening under previous versions?

-> Did the researchers adequately trigger ads?

-> Are subjects able to notice all direct ways in which listening data is being used?

Instead, I would simply accept a negative as showing that no, Facebook isn't listening. (I'd happily accept false negatives.) Instead, what is really interesting is that numerous people decided to do an informal experiment such as put their phone next to their TV set to Spanish-language programming and then, to their shock and horror, Facebook began showing them Spanish-language ads.

[paulcole]

>an overwhelming amount of anecdotal evidence reported all over the Internet

This is exactly why I believe pizzagate, am a 9/11 truther, etc.

[logicallee]

Public interpretation of shared information is not at all the same as anecdotal evidence!

Don't conflate these two things. I know beans make people fart because of anecdotal evidence I've heard from different people. That doesn't make this "fact" wrong.

On the other hand, interpretation of public information is quite different.

-

Aside:

Also I have a small aside here for you personally. I'd like to remove something from your mind so that you never speak about it again. I don't want 9/11 truthers derided here on Hacker News because I consider the example to be one that should not be included here. Flat earthers? Sure.

As for 9/11: I'm not a "9/11" truther. I've never called for an investigation into it. I don't care about it. But for you, if you care about this subject for some reason, the standard for whether you should be a 9/11 truther, in the sense that you believe anyone had any prior knowledge those acts would be committed, is whether you find any leaks to that effect by such a government official.[1]

The standard regarding whether it is feasible for 9/11 to have included a controlled demolition is the opinion of civil engineers, whose profession this sort of thing is.[2]

So while I can't speak for you, I do have my own standards. This does not mean I'm a 9/11 truther. I'd like you to stop using it as an example, though. Use things that are false (like flat earth) if you want to make some kind of cheap point, please.

Note: I want to reiterate that I am in no way, shape, or form, a 9/11 truther.

[1] https://www.youtube.com/watch?v=i6eMq5Rit1w - "From about August 23rd, every night, until September 2nd, September 3rd." This insider says "We are never going to look for who they are, and it's probably not worth looking for them." Her Wikipedia page is quite credible as an insider. In addition it includes major smears on her character and a complete refutation over the official channels you would expect. She meets the standard outlined above.

[2] https://en.wikipedia.org/wiki/Architects_%26_Engineers_for_9... - "The organization has compiled a list of criteria for a controlled demolition that it says the collapse of the World Trade Center meets: the destruction followed the path of greatest resistance, the debris was symmetrically distributed, the rapid onset of the destruction, explosions and flashes reported by witnesses, steel elements were expelled from the building at high speed, the pulverization of the concrete, expanding pyroclastic clouds, lack of pancaked stories in the debris, isolated explosions 20 to 40 stories below the wave of destruction, molten steel and thermite traces found in the debris."