Lessons from the PEAR Lab: Exploring the Possibility of Anomalous Bias in Scientific Measurements
A Project for PSY/ORF 322 by Dan Recht
I. Introduction
In the quarter of a century since its founding, the Princeton Engineering Anomalies Research (PEAR) laboratory has, through dozens of studies, made a very strong case for the possibility that human beings can influence machines through some function of consciousness. The outputs of various “random” (chaotic or quantum mechanical) physical processes, referred to as “random event generators” (REGs) seem to respond in small but significant ways both to the stated intentions of human operators and to the general state of “subjective resonance” of a particular place and time (cf. Nelson et al. 1998). Needless to say, PEAR’s work is quite controversial in most scientific circles because of its surprising results and far-reaching implications. Thus it is usually disregarded by researchers in the natural sciences. Although it is probable that the anomalies observed by PEAR are explainable via the existing scientific paradigm, there is a very reasonable possibility that they are not.
If there is even a small chance that that the human mind can skew the probabilities of random events, then that risk must be addressed for science research to continue. Far too many high-precision measurements are made in laboratories for researchers to disregard the possibility of anomalous bias[1] in their results. After taking a good look at PEAR’s results, even the most dubious skeptic will admit that such anomalous influence is possible. Perhaps it is not likely, but it is certainly not unlikely enough to ignore. It follows then that the physics community must take action. But what should be done?
Obviously there are no straightforward answers to this question. That said, this paper is intended to serve as the first step on the long road to resolving this very serious challenge to the integrity of scientific measurement. By taking the anomalies observed by PEAR as given (if they are an illusion or experimental artifact there is nothing to worry about) and exploring different possible solutions to the problem of anomalous bias, this paper attempts to provide useful guidance for future studies and to encourage collaboration between the fields of physics and anomalies research.
II. Ignoring the Problem
One of the only well-documented, replicable correlations found in the parapsychological literature is that of a subject’s belief in the anomalous phenomenon being studied (usually extra-sensory perception) with their performance on a task requiring that anomalous ability (Jahn, 1982 p. 140). This is what is known as the sheep-goat effect. If this effect applies to the anomalies studied at PEAR, then experimental science has nothing to fear as long as the majority of its practitioners continue to disbelieve the existence of anomalous bias. This is because the sheep-goat effect has some very comforting consequences. Namely that virtually all replicated measurements made on or before the present day are free from anomalous bias as it is almost certain that at least one incarnation of any experiment was carried out by non-believers, and that, by the same logic, most measurements in the immediate future will remain free of such bias. Of course this solution is only temporary as the shield of ignorance will fail the moment that the scientific community accepts these anomalies as fact. Thus if the sheep-goat effect is indeed found to govern these human-machine interactions, anomalies research may be faced with a strange but serious dilemma. If there turns out to be no effective solution to anomalous bias except for disbelief, anomalies researchers throughout the world may need to cease publishing lest they accidentally convince their more skeptical colleagues!
To test whether the sheep-goat effect holds for anomalous bias, one need only conduct two relatively simple REG experiments. The main feature of the first’s protocol would be sorting a group of subjects unfamiliar with PEAR into believer and non-believer classes based on questionnaires given before their participation and then comparing the results of the two classes. The second experiment would be very similar to the first except that the classes would be formed via two differing treatments, one designed to foster belief the other intended to discourage it. It would probably be wise for the second experiment to include a control group as well. It is worth noting that the PEAR lab environment corresponds closely to a “belief” treatment. Thus the “non-belief” and control conditions would have to be held elsewhere.
If the sheep-goat effect is present, the quest to find a solution to the problem of anomalous bias gains urgency. The fact that science has as yet encountered no major problems due to a mysterious measurement inaccuracy implies one of two conclusions. If the sheep-goat effect does not apply, then anomalous bias is necessarily a small-scale problem if it exists at all. If the sheep-goat effect does apply, however, the possible size of anomalous bias could be much larger since it is not currently being observed. Furthermore, there is a finite time window in which to find a solution to anomalous bias before science begins to believe in it and thus discovers its exact magnitude first hand. Either way it is important to learn as much as possible about anomalous bias in order to determine whether it is a problem and, if so, how to solve it.
III. Size does matter
The first important question that needs to be answered about anomalous bias is that of scale. If the effect is of constant size, how big is it? If it’s magnitude is variable then on what does it depend? Clearly, anomalous bias of any constant size will eventually be large enough (relatively speaking) to matter. Luckily, there is reason to believe that the scale of anomalous bias may be inversely related to the precision of the measuring apparatus it affects. If, as some theories suggest, consciousness is “inserting information, (Jahn 1995, pp. 302)” into random processes, it is reasonable to assume that the amount of information “inserted” is constant to within a few orders of magnitude. From a physical perspective, one could view the interaction as a mind doing work on a random system to reduce its entropy. In this frame of reference, the above assumption is equivalent to the statement that the interaction usually happens within a relatively narrow spectrum of energies. This seems very plausible since data from several REG experiments reveal that the mean effect sizes of all individuals have fallen well within a two order of magnitude range. The same is true for the effect size of a single individual over repeated trials (RegDbase).
Every measurement has associated with it an experimental error, which any natural scientist worth his salt is an expert at estimating via well-understood mathematical techniques. If one measures some value to be with error , an additional measurement of that value falling within, say, with error does not contribute much additional information. Thus it does not greatly reduce the entropy of the experimental system. Equivalently, one could say that it would not cost much energy to bias further measurements (or even the original measurement) to different values within this interval. Since the human-machine anomalies in question evidently operate at a small, relatively-constant energy, it stands to reason that any anomalous bias would not be able to cause measurements to deviate far beyond the range dictated by the experimental error. While the textbook definition of a high precision measurement is a reading with a large number of significant figures, useful measurements of this sort are characterized by small error values. Thus, according to this theory, causing large absolute biases in such measurements should have a greater associated energy cost than causing the same absolute bias in a less precise figure. In other words, the susceptibility of a measurement to anomalous bias should be roughly proportional to its estimated experimental error.
If this is so, then the anomalous bias problem is solved. Steps that scientists normally take to reduce their experimental error such as purchasing precise equipment and controlling variables very tightly would also reduce the effect of anomalous bias on their results. Furthermore, since anomalous bias would not be able to cause observations to deviate from their true values by much more than , it would have no significant impact on measurement accuracy since the uncertainty due to it would already have been accounted for in the estimated error value..
To test whether the human-machine anomalies do indeed scale with experimental error it is probably best to use a variation on the pendulum damping experiment as it is the only PEAR study that includes a measurement with a calculable experimental error (Nelson 1994 pp. 472-475). In the pendulum experiments operators try to influence the damping rate, as measured by a photodiode. To manipulate the amount of experimental error, one could simply change the resolution of the photodiode. Although the original experiment used 50 nanosecond resolution to measure the pendulum velocity, which was then used to calculate the damping rate, lower time resolutions and thus higher error values would certainly be possible. One could have a group of operators all do trials at several resolution levels (preferably over a very broad range) and then compare separately each individual’s results at different resolutions. This would give a decent estimate of the extent to which this effect varies with error.
Sadly, there is some anecdotal evidence that offsets this potentially good news. Preliminary data from the aborted PlantREG experiment suggest that the anomalous effect can be much larger than is usually observed. In PlantREG a philodendron was placed in a dark room with a full-spectrum plant light whose brightness was proportional to the deviation of an REG from expectation. Before the experiment was cancelled due to technical problems, results of a size significantly greater than those found in all human experiments were observed (Dunne). While it may be that some property of philodendrons or plants in general is responsible for the larger effect size, it seems more likely that these results are due to a more basic feature of the anomaly being studied. Because the philodendron was in a dark room, it would certainly have died had it not influenced the REG enough to bring the light to a comfortable level. One possibility is that the high stakes of the situation facilitated the plant’s increased effect size. If this is so then anomalous bias in laboratory experiments, which often have very high stakes, could be much larger than the anomalous effects observed by PEAR. In other words, high-stakes situations might significantly increase the energy of the mind-machine interaction and thus invalidate the “small energy” assumption of the previous argument.
Testing the high-stakes hypothesis is a delicate matter since one cannot put human subjects in life-threatening situations. Consequently, experiments on humans would have to feature somewhat sizable monetary incentives in order to raise the perceived importance of the REG’s output. A simple comparison between a control group and a group informed that they would receive a reward if they caused a certain fairly large effect would probably suffice. If these results were unclear one could attempt to do high-stakes experiments on animals, perhaps giving food only after excellent REG performance. This, however, would be a very difficult and expensive project.
If it is found that anomalous bias is scales in such a way as to be an issue, it will be necessary for the scientific community to find a way to control it. The many nuances of human-machine anomalies make this problem far from trivial. The development of a solution requires much more information about these interactions than is currently known.
IV. The Nature of the Beast
If the studies described in sections II and III reveal that science will indeed have to deal with anomalous bias, it will help to know as much about the enemy as possible. A reasonably accurate method of detecting anomalous bias would be an important first step toward preventing it. As is explained below, such a method may be within reach.
At this point it will be useful to formally divide anomalous bias into two categories: volitional and situational. Volitional bias is the anomalous skewing of data in line with the conscious intention of one or more of the experimenters. Situational bias is anomalous bias due to some sort of “subjective resonance” in the laboratory environment. Although the actual mechanisms by which they occur may be quite similar, their statistical signatures should be noticeably different.
Volitional bias corresponds to the anomalies observed in the standard REG experiments. It stands to reason that a data set influenced in this way will look similar to data from those studies. It should be possible to look at the noise in the reported measurements and check with reasonable accuracy for the presence of anomalous bias just a one would for bias of a more mundane sort. Since volitional bias is based on an intention that remains constant throughout the experimental process, it should act in a consistent manner on the data. That said, it is not possible to identify the cases in which anomalously biased data will resemble a baseline intention versus when it will resemble a high or low intention because such a characterization would require knowledge of the true value of the quantity being measured. Luckily, in both cases there should be a significant change to the shape of the data distribution. Baseline intentions have a variance that is smaller than expectation while high and low intentions show a variance that is greater than theory (Margins p. 116). This suggests that for both types of volitional bias the measurement distribution will likely display a significant excess kurtosis. For the baseline type, this will be due to peaked-ness; for the high/low type, it will be due to tailed-ness. To approach the problem systematically, a D’Agostino-Pearson test for normality, which is based on skewness and kurtosis, should be used. If volitional bias is found, the experiment should be repeated by another scientist with a different or null intention.