[Aaron]

Very Cool

http://espgame.org/

[gg3]

Very

very

interesting.

Oh how I would love to know the hypotheses they're testing with this, and how they've operationalized the variables.

Luis von Ahn has three "games" running right now: "The ESP Game," "PeekaBoom," and "Phetch" (Fetch). His research interests include human computation, pattern recognition, cryptography, and steganography.

In each case, you log in and get randomly assigned to one other person, or, in the case of Phetch, a bunch of people.

In The ESP Game, you and your random partner each see an image on the screen and attempt to type a series of words that describe it. There is already a list of words that are "obvious" and are excluded, so you have to look for less-obvious ones. When you & your random partner type the same words, you get a new image, and the goal is to get through 15 images in 2-1/2 minutes (average of 10 seconds per image). Point values are assigned for degree of non-obviousness of word hits.

Strictly speaking, it doesn't take ESP (nonlocal perception/action) to score on this. That is, the brain has an enormous amount of latent information processing capability and can pull a remarkable amount of information out of a relatively small amount of target or input data. And it would appear that both parties have the same target (or do they?) so there are no hidden variables except for the perceptual and cognitive skills & priorities that the other player has when evaluating the target.

So why would von Ahn call this an "ESP" game...?

A few things are fairly well established about nonlocal perception & action: The effects are robust (i.e. similar statistical deviations occur throughout the research). They are pervasive (they are a constant part of the perceptual/cognitive background activity and can be found in research data sets that were not even looking for them, see also "retrocausality"). They are relatively small (the statistical offsets are small and become obvious only over a large number of trials). They are "psychologically lawful" (they conform to patterns established for other perceptual, cognitive, and suchlike skills, for example performance increase with feedback, and performance decline under hostile scrutiny). Last but not least, they appear to conform to other physical models & examples of nonlocality & entanglement (for example including the need for "local" communication to confirm results).

So we could hypothesize that people who are above average at nonlocal perception/action tasks might do slightly better at the ESP Game task than people who are average or below. The game's scoring system could select for such people by rewarding their performance advantage compared to others. But here we are only talking about a few percentage points of advantage. This might be a helpful efficiency increase if the goal is, as stated, merely to label web images in order to simplify image searches, but in a way it seems like a "too mundane" application of a capability that has potentially far more interesting uses.

Note that the Game assigns higher point values to images with more excluded words, and thus, with fewer obvious descriptors. Thus there is an advantage to seeking out the less-obvious descriptors. Since the available "local" information is less, the use of nonlocal information could make a bigger difference in outcomes. But toward what end...?

The general case here is "the ability to make concordant observations of obscure or non-obvious details from a shared data-set." Beyond that, the general case here is to train people in that ability by giving them positive feedback for successfully using it.

I have some intuitive speculations about this one, in terms of how it could use nonlocal perception to leverage results in cases where the outcomes are more important than simply cataloging every image on the web. Think about maps, faces, buildings, vehicles, etc. Think about identifying targets such as hostile troops, missing persons, vehicles used in criminal activities (e.g. kidnapper's car). Think about picking sites for geological exploration for minerals, picking investments for a portfolio, etc. As a generality, cases where something of high value is at stake: lives, large sums of money, etc. A persistent slight advantage in such cases could have a significant impact on outcomes.

I'll have more to say about von Ahn's other games in the next few days...

[gg3]

Something more about concordant descriptions of targets...

In my graduate research on communications & states of consciousness, one of the branches I was exploring was how to operationalize "empathy."

Given a shared altered state with shared content: mutually-induced hypnotic trace with dream-like experiences. We quickly found that the degree of implicit communication was absolutely enormous: a few words between the subjects in these experiments would convey pararaphs of information. More interestingly, this communication at times appeared to occur so rapidly that it was not even possible to parse out a time-sequence of cause & effect: it was as if certain elements occurred simultaneously or instantaneously. I wasn't seeking to parse "local" from "nonlocal," since that would have required more sophisticated experimental designs and psych lab facilities (this was planned for down the road a ways, but there wasn't the time/money to continue).

So about empathy. I operationalized empathy in terms of "concordance," defined as the ability of one person to successfully describe elements of another person's subjective state & experience. The test for this was to have each person (after completing a shared states session) write out two lists of descriptors: their "A" List was the list of descriptors of their own experience, and their "B" List was the list of descriptors of the other person's experience.

What I was looking for was matches where items on each person's B List, minus items on their own A List, matched items on the other person's A list.

In other words, how much of what I think you experienced, minus the amount of it that I also experienced and thus could be considered "obvious" common experience, did you actually experience?

I was able to run this particular set of tests a few times with experienced subjects. However, what I got was not a set of nice neat word lists, but something that looked more like free-verse poetry: the kind of thing you might expect from subjects in a psilocybin experiment (not surprising: mutual hypnosis can produce psychedelic experiences, and we saw our share of those throughout the project). In other words, the results I got were "not very useful" for the purpose I had in mind: too much subjective judgement on the part of the evaluator would be required to match up the A Lists and B Lists. So I figured that my operationalization of empathy was flawed, and left that branch of the project to go on to other branches.

NOTE: Do not mess around with mutual hypnosis at home, this is not a casual procedure, it has similar risks to those associated with psychedelic drugs and should only be done under proper supervision in a controlled setting. If you're interested in shared altered states, look up "shared lucid dreaming" and go for that; it has the intrinsic safety factor that dream states are self-limited by waking up after one's body/brain are sufficiently rested.

SO how does this relate to von Ahn...?

Seems to me he is going for something similar in form though with a different goal in mind, and with the procedures stripped down to the essentials: there is not a shared altered state, but there is a common set of experiential data, i.e. the target image for each run. Each person's task is to guess what the other person will guess, i.e. to, in a manner off speaking, empathize with the other person in a generalized abstract way. This has me wondering whether von Ahn's approach, if modified to fit in with my experimental protocols, could be used to measure "concordance" and by implication, measure degree of empathy...? (If I had a decent research grant..!)

[I_Like_Plants]

On the site www.urbansurvival.com along with the bad spelling and interesting to say the least financial prognostications, you will find references to, and links to, a sort of project involving a bunch'a computers that analyze internet, news, etc traffic and use it to predict the future.

I guess it's kind of like the drunk who never consciously knew he was going to throw the beer bottle through the window, but deep inside, yeah, he knew he was going to throw the beer bottle through the window. These programs attempt to predict from the mumblings of the drunk when the beer bottle will be thrown.

They call the computers "time eggs" it gets pretty strange, but it sounds like the same kind of thing that these games operate on. Get a bunch of poeple together and they might just know more than they know they know.