Do you want to know the history of telepathy and how do we go about testing whether an impression of telepathy is a coincidence, a hallucination, a psychiatric problem, a fabrication, or genuine? We’ll use an example of two friends, Gail and Tom.

Gail tells you that she thinks she’s in telepathic contact with Tom. She seems to know what he’s thinking and vice versa, not literally with words but with intentions. She provides examples of spontaneous episodes that seem credible, and you know her well enough to accept that she’s not making up these stories. You decide to test her ability.

You begin by asking Gail to think of any number, at random, from 1 to 10, then to multiply that number by 9. If the resulting number is two digits, then Gail should add the digits together, and then subtract 5. (You should try this too, to see what happens.) Now you ask her to determine which letter in the alphabet corresponds to the number she ended up with (example: 1 = A, 2 = B, 3 = C, etc.).

Now ask her to think of a country that starts with that letter and to remember the last letter of the name of that country. Then ask her to think of the name of an animal that starts with that letter, and to remember the last letter in the name of that animal. Finally, ask her to think of the name of a fruit that starts with that letter.

Now ask Gail, are you thinking of a kangaroo in Denmark eating an orange?

If she is (and if you were), then you’ve learned the first lesson in how not to conduct a telepathy experiment. This trick works with the majority of native English speakers because we share common knowledge about the frequency of words and concepts, and this allows you to be led down the primrose path to ensure that you select the most common choices for those words. Magicians use similar tricks to force you to select a specific item that you think you’re choosing freely, but actually you’re just following well-known biases.

So let’s look at another design for a telepathy experiment. First you have Gail and Tom sit back to back. You ask Tom to think of anything that comes to mind, and then you ask Gail to report what he’s thinking. If you do this a few times, you may be amazed to and that occasionally Gail gets it exactly right. But maybe they’ve been passing clues through body positions or innocuous sounds. Or maybe they’ve decided to play a trick on you by secretly setting up a series of mental targets in advance.

So you separate them into different rooms and satisfy yourself that they can’t communicate in any conventional way. Not with sounds, vibrations, odors, or cell phones, or even via a confederate who might be persuaded to secretly pass messages between them. Now you try the experiment again, asking Tom to think of something at random, and seeing if Gail gets it. Sometimes she does, and with wildly good accuracy. Still, something doesn’t feel right. Surely telepathy isn’t that easy. If it was, it wouldn’t be considered controversial.

You decide that maybe because Gail and Tom know each other so well that what looks like amazing telepathic hits is actually due to shared knowledge. That is, if they both went sailing the day of the test, then it is likely that water sports would be on both of their minds, so Tom might select a water-sport related item as his telepathic target. But so would Gail, because people aren’t good at selecting items at random. Here’s where the combination of common knowledge and recent memory can easily bias the results to make it appear that Gail was telepathic.

To get around this, you devise a method where you select one target photograph out of a large pool of photos with a broad range of content. To make that selection without introducing your own nonrandom decision biases, you toss dice, or use a computer-based method to choose the target. But now you wonder how are you going to judge how good the match is between what Gail says and what Tom’s target was, without introducing yet more biases? If you select a photo with a red balloon, and Tom doesn’t say “balloon” but he does report the color pink, is that good enough to count as telepathy?

So you make it easier to judge a hit or miss by selecting four photos, each as different from one another as possible, and then you randomly select one of those photos as the target to give to Tom. Now you can judge Gail’s accuracy after Tom finishes mentally “sending” the target by showing Gail all four photos and asking her to select the one that best matches her impressions of what Tom was sending.

To avoid accidentally giving Gail a clue about which one of the four targets was the actual one, you decide to ask someone else to select the target photo and give it to Tom. Then, when you’re with Gail and evaluating the test session, you have no idea which of the four photos is the target, and when you hand Gail the four targets to evaluate, you use fresh copies so that Gail never even sees the photo that Tom handled. Now Gail selects one of the four targets, and she either gets it right or not.

Okay, so now you conduct the test session using this design, and at the end Gail selects the correct target, and so you have proof of telepathy, right? Well, not quite.

Let’s say that Gail’s impression was about a dog, the target photo was a picture of an elephant, and the other three decoy targets were grass, a tree, and a bird. Gail selected the elephant because it was the only animal with four legs among the four targets. While this would count as a direct hit, it isn’t a very impressive hit because she didn’t actually say the target was an elephant.

Maybe she was having a bad moment. You run the test again. Again she gets a hit, but again it wasn’t an exact match. So you run dozens of trials. Some are hits and some are misses. If Gail was not telepathic, then she should get the right answer purely by chance on average about one in four guesses, or 25 percent of the time. Barring telepathy, if the targets are really selected at random, there is absolutely nothing she can do to amplify this hit rate.

Now you decide to expand your test to include many different pairs of people. You gather data from dozens, then hundreds, then thousands of sessions because you want to see whether telepathy is true for people in general, rather than just for Gail and Tom. In the process of running these tests, you develop even more sophisticated controls to ensure that the participants can’t cheat or introduce biases, you add other methods to prevent yourself or your assistants from cheating or introducing biases, you devise fully automated techniques to record the data properly, and you use statistical methods that are simple but appropriate to evaluate the data.

After running several thousand trials, you observe that the telepathic “receivers” are on average correctly selecting the randomly chosen target about one in three times rather than the chance-expected one in four. It doesn’t seem all that impressive. But you calculate the odds against chance to see how unusual it is to get 33 percent when you’d expected to get 25 percent by chance, given the number of trials you’ve run. The outcome makes your head spin.

The above description is a brief recapitulation of the evolution of telepathy tests from the 1880s to today. A method designed in the 1970s, called the ganzfeld test, is the most recent iteration of our Gail-Tom test scenario. This design has become a standard technique used to test for telepathy. And it is directly relevant to our interest in the siddhis because the ganzfeld design was explictly developed to test whether telepathy would be easier to detect when sensory noise was reduced. This idea in turn was based on the yogic lore that withdrawal of the senses would assist the yogin to become more aware of subtle inner impressions, including telepathic experiences.