5 ms·

I listened to the recording in the article, and it sounds like a human. Further, it sounds like the same human. And I found myself reading the emotional message

by balex 4y ago

I listened to the recording in the article, and it sounds like a human. Further, it sounds like the same human. And I found myself reading the emotional message behind the sounds. So...yeah, it's an interesting theory, but I think it's more likely a prank or someone making a point.

FartyMcFarter 4y ago
There's a point in the MP3 that sounds possibly non-human at 0:18 to 0:19. The pitch goes quite high in a way that sounds artificial. But overall I agree that it sounds very human. For example, the grunt at 0:13 doesn't sound like random noise at all, it could easily be a sound effect in a movie or game.
atkailash 4y ago
dan-robertson 4y ago
Another way to think about it: there are a lot of planes and different ways the PA system could go wrong. You’re much less likely to read an article about a plane being ‘haunted’ by a strange clicking or buzzing or humming or white noise sound.
- ComplexSystems 4y ago
  That's because a strange clicking or buzzing sound is much less likely to be a human voice.
- balex 4y ago
  But why would I think that way? I listened to the recording, it sounds human to me. Why would I need to crunch further data?
  - dan-robertson 4y ago
    I think you’re trying to decide if it’s a prank or not. And (I assume) you care about trying to come to conclusions that are likely correct. Therefore I think you should care about things that would lead you to incorrect conclusions. Suppose the two possibilities are ‘prank made from human’s voice’ and ‘weird electromechanical phenomenon’. The logic I believe you’ve implicitly applied is that most electromechanical phenomena do not sound human and therefore this human-like sound is likely human. However I put it to you that this logic is flawed because you have not heard a randomly sampled electromechanical phenomenon but rather a sound that was sufficiently human-like to warrant an article about hauntings so thinking about the ordinary distribution of electromechanical phenomena is misleading. Furthermore I think that you’re assigning relativity equal weight to the two possibilities of a human prankster and any kind of PA system fault when I would argue that the latter is much more common. That is, I think you’re answering a question like ‘we flipped a coin and if it came up heads, we made silly noises into a dictaphone and if it came up tails we tortured the device in some way while it was recording. We’ll play the sound and you can guess which it was’ whereas I think the answer in this case hinges a lot more on the equivalent of the coin-flipping process (eg what if you needed 10 heads in a row for the first choice? What about 15? 20?) One can formalise this too. I’ve tried to come up with some priors that illustrate the point and which I think are reasonably generous towards your conclusion. Say H = some human pranking a flight with silly noises, M = some mechanical failure in the PA system, Mh = human-sounding mechanical failure Mm = mechanical-sounding mechanical failure S = human-like sounds And let’s invent some numbers for a random flight: P(H) = 1e-6 (feels too big to me) P(M) = 1e-3 (feels a bit small to me) P(Mm | M) = 0.995 (I.e. 99.5% of mechanical sounds sound mechanical) P(Mh | M) = 0.005 We read an article about S happening so we now wonder: P(H | S). Apply Bayes theorem: = P(H) P(S | H) / P(S) = P(H) * 1 / P(S), and in our model = 1e-6 / P(S). Now what is P(S)? We can calculate in our model (assuming we’re describing all possible causes and for simplicity ignoring a flight with a human-sounding mechanical failure and a prankster). P(S) = P(H) + P(Mh) = P(H) + P(Mh | M) * P(M) = 1e-6 + 0.005 * 1e-3 = 6e-6 And we can plug that back into the equation above: P( H | S ) = 1/6 You can fiddle around with the priors and get a feel for how the affect the results but the point I want to make is that it can be true that electromechanical sounds are very rarely human-sounding and still be likely that the sounds in this case are likely electromechanical. I believe that your argument was based on the incorrect logic I described at the start of this comment and not because you had strong priors that the very unlikely event of a human prankster was more likely than the combination of the unlikely events of a faulty PA system and a fault making humanoid sounds.

by balex 4y ago

FartyMcFarter 4y ago

There's a point in the MP3 that sounds possibly non-human at 0:18 to 0:19. The pitch goes quite high in a way that sounds artificial. But overall I agree that it sounds very human. For example, the grunt at 0:13 doesn't sound like random noise at all, it could easily be a sound effect in a movie or game.

atkailash 4y ago

dan-robertson 4y ago

Another way to think about it: there are a lot of planes and different ways the PA system could go wrong. You’re much less likely to read an article about a plane being ‘haunted’ by a strange clicking or buzzing or humming or white noise sound.

ComplexSystems 4y ago
That's because a strange clicking or buzzing sound is much less likely to be a human voice.
balex 4y ago
But why would I think that way? I listened to the recording, it sounds human to me. Why would I need to crunch further data?
- dan-robertson 4y ago
  I think you’re trying to decide if it’s a prank or not. And (I assume) you care about trying to come to conclusions that are likely correct. Therefore I think you should care about things that would lead you to incorrect conclusions. Suppose the two possibilities are ‘prank made from human’s voice’ and ‘weird electromechanical phenomenon’. The logic I believe you’ve implicitly applied is that most electromechanical phenomena do not sound human and therefore this human-like sound is likely human. However I put it to you that this logic is flawed because you have not heard a randomly sampled electromechanical phenomenon but rather a sound that was sufficiently human-like to warrant an article about hauntings so thinking about the ordinary distribution of electromechanical phenomena is misleading. Furthermore I think that you’re assigning relativity equal weight to the two possibilities of a human prankster and any kind of PA system fault when I would argue that the latter is much more common. That is, I think you’re answering a question like ‘we flipped a coin and if it came up heads, we made silly noises into a dictaphone and if it came up tails we tortured the device in some way while it was recording. We’ll play the sound and you can guess which it was’ whereas I think the answer in this case hinges a lot more on the equivalent of the coin-flipping process (eg what if you needed 10 heads in a row for the first choice? What about 15? 20?) One can formalise this too. I’ve tried to come up with some priors that illustrate the point and which I think are reasonably generous towards your conclusion. Say H = some human pranking a flight with silly noises, M = some mechanical failure in the PA system, Mh = human-sounding mechanical failure Mm = mechanical-sounding mechanical failure S = human-like sounds And let’s invent some numbers for a random flight: P(H) = 1e-6 (feels too big to me) P(M) = 1e-3 (feels a bit small to me) P(Mm | M) = 0.995 (I.e. 99.5% of mechanical sounds sound mechanical) P(Mh | M) = 0.005 We read an article about S happening so we now wonder: P(H | S). Apply Bayes theorem: = P(H) P(S | H) / P(S) = P(H) * 1 / P(S), and in our model = 1e-6 / P(S). Now what is P(S)? We can calculate in our model (assuming we’re describing all possible causes and for simplicity ignoring a flight with a human-sounding mechanical failure and a prankster). P(S) = P(H) + P(Mh) = P(H) + P(Mh | M) * P(M) = 1e-6 + 0.005 * 1e-3 = 6e-6 And we can plug that back into the equation above: P( H | S ) = 1/6 You can fiddle around with the priors and get a feel for how the affect the results but the point I want to make is that it can be true that electromechanical sounds are very rarely human-sounding and still be likely that the sounds in this case are likely electromechanical. I believe that your argument was based on the incorrect logic I described at the start of this comment and not because you had strong priors that the very unlikely event of a human prankster was more likely than the combination of the unlikely events of a faulty PA system and a fault making humanoid sounds.