Deception Detection Software Applied to UFO Transcripts

After some testing i think that sentence length might have something to do with it. In tests where the sentece leng was quite short we were getting true and longer sentences deceptive.

reply to post by PhoenixOD


Yea, that would certainly seem to apply to the stuff I wrote.

I guess I'm better at formulating proper sentences when I'm being creative.

reply to post by gortex

Has anyone tried typing something they know to be true in it to see what result they get ?
Lie detectors work on physical stimuli , I don't see how an algorithm can do the same thing with any degree of reliability .

Yes; I typed in a completely objective description of my situation at the time: "I am sitting in a comfortable chair. The monitor is one meter away..." etc. It decided I was being deceptive. Back to the drawing board.

Sadly, this bust of an invention will now be picked up by conspiracy theorists to "prove" Neil Armstrong was lying, chemtrails exist, etc, etc, etc. They will not accept the results, however, if the algorithm decides that Hoagland, et al., are being deceptive.

reply to post by gortex


I typed in several things that I know to be true. All of it came up as deceptive. Something about the value being too high. Maybe it's designed to analyze things that based on speeches, and the way that I write is a lot different than how I speak.

Reply to post by ZetaRediculian

Thanks OP I bookmarked the website and will use this for those "out there" threads that pop up on ATS. I know its not accurate but it is entertaining.


Posted Via ATS Mobile: m.abovetopsecret.com

I do not know what results everyone else is getting, but I have taken the time to type up little occurrences into the box on the website, some of which I made up, and some of which were true. It was able to accurately analyze each one, with a 100% success rate.

I am impressed. The first thing I began applying it to were sasqui sighting stories. I tried about 5 different write-ups by Janice Carter, who purportedly hangs out with sasqui in TN...Every single one read deceptive. I had already figured this was the case, so this doesn't really change anything with that particular case.

I will keep trying to see if I can fool it, but with the success rate I have had so far, I expect this algorithm to be accurate at least 80% of the time, although it has yet to fail me.

I even just inputted all of the text of this post, above this line, and it reads "normal." lol, this is cool.

"The U.S. government has no evidence that any life exists outside our planet, or that an extraterrestrial presence has contacted or engaged any member of the human race"

Deceptive

Iranian authorities deny reports several news agencies made about the explosion allegedly occurred near the city of Isfahan.

Deceptive

But I want to say one thing to the American people. I want you to listen to me. I'm going to say this again. I did not have sexual relations with that women, Miss Lewinsky. I never told anybody to lie, not a single time, never. These allegations are false. And I need to go back to work for the American people.

Deceptive.. haha

I'm not sure how this algorithm works, but it is most likely using a range of factors to analyse, such as absolutes and limiting values, that then cross references language patterns on smaller chunks of the text, with other chunks, and expands outwards comparing larger and larger portions until it has a probability value on the validity of the text overall. This is why it requires at least fifty words, so that it has a sufficient sample to make a valid estimation on the probability of accuracy.

For instance, if I say "the screen is one metre away from me", then I am stating an absolute that is technically, most likely, not true. There is no qualifier to substantiate my statement. However if I had stated "the screen is approximately one meter away", then this is a more accurate statement, and is less deceptive.

However, as the algorithm expands on larger portions of text, It might pick up a qualifier that makes a shorter deception to be seen as a truthful statement in relation to the overall text, for instance "I have set up my work area so that my computer screen is at an optimal distance based on operational health and safety requirements, where my eyelevel is in line with my screen, and the back of my chair is approximately one metre and fifty centimetres away from my screen, so the screen is one metre away from me, when seated." In this case, I have provided qualifiers to justify my deceptive statement of "the screen is one metre away from me".

It also seems that this software has some learning capabilities so that it will adjust its probabilities based on the information input into the database. As the database expands, it will be able to more accurately assess the probabilities of honesty on a given piece of text.

Note: Interestingly, the above text was input, and it came back as true. An interesting algorithm that seems to have a good probablility of analysing the truth/deception of a piece of text.

Originally posted by DJW001
reply to post by gortex

 

Has anyone tried typing something they know to be true in it to see what result they get ?
Lie detectors work on physical stimuli , I don't see how an algorithm can do the same thing with any degree of reliability .

Yes; I typed in a completely objective description of my situation at the time: "I am sitting in a comfortable chair. The monitor is one meter away..." etc. It decided I was being deceptive. Back to the drawing board.

Obviously the algorithm has no idea about actual deep content, because even humans who do have strong natural intelligence can't always distinguish.

What it appears to be doing is trying to tell if the text is in some way "manufactured" in an unusual way by various linguistic statistics. And telling yourself to make up some text and type it in a box is manufactured rather than spontaneous, and it may reflect subconscious effort or crafting. That's what it's getting at, I believe.

So perhaps it is not a false positive, though the results may better be called "normal" and "manufactured".

Originally posted by ZetaRediculian
"File Status: normal" means not deceptive.

I typed this in and got Normal.

My name is Alvarez. I once ate an entire cow. It was quite tasty. I bought it from a farmer in Peru. His name was also Alvarez. It turns out that we were actually the same person. I am a time traveler.

I don't think this works too well.

Edit: I have never eaten an entire cow, I imagine it would be tasty though. I have never been to Peru. I am not a time traveler and I am not and have never met an Alvarez.

I wrote on there and it came back deceptive and it was truth so thumbs down

They also claim to be able to determine how deceptive a website is:



Draw your own conclusions.

Well what a crock of crap.i have just put in two different stories and it told me that i was lying about them both when infact there true.So does it work NO NO NO!

I gave it a little test that I suspected this kind of thing might get wrong, but it didn't (unless I mistakenly got something wrong in the text):

Well, let's see now--the capital of Hawaii is, er, uh, Honolulu, and for Idaho it's, uh, well, Boise, and the capital of Maine is, er, uh, Augusta, and then, um, let's see, for Montana it's, uh, Helena, and the capital of Missouri is, er, uh, well, um, Jefferson City, then, oh, uh, for New Jersey, it must be Trenton, and uh, for Nebraska it's, uh, Lincoln, and for Alaska, you know, it's, uh, well, Juneau, and not to forget California, uh, not L.A. but it's definitely Sacramento. and the capital of Arkansas is, uh, er, uh, well, Little Rock, to name a few. Should I go on?

Verdict: truthful

I just tested the Deception Detector at stealthserver01.ece.stevens-tech.edu... and that program a major FAIL.

It got my gender right once, wrong twice ... It got a female's gender wrong ... It got a male's gender correct. Apparently it only counts the male or female pronouns in the writing ... that's it! So, a female talking about a male, or visa-versa will give the author a sex change.

Inputting known hoaxes from ATS gives them a clean bill of health ... normal. Even when the hoaxer later admits they hoaxed.

i guess it will only be a matter of time before someone with too much timeon their hands runs every publci figure conspiracy theorist through the algorithum. Might make for some amusing results.

Coast 2 Coast with Art Bell and Victor
(the Alien Interview footage)
Transcript - Click here
Results:
Normal (talking about the aliens and their nature)
Deceptive (describing his own motivation)