In physical science, observing something changes the something. Skillful observation methods are designed to make the change infinitesimal, and thus irrelevant. Try measuring the voltage across a 10 meg-ohm resistor with a 10 or 20 meg-ohm meter, or the waveform across a 22 picofarad capacitor with an 11 picofarad oscilloscope probe to see what I mean.
Predicting the behavior of human beings is difficult; predicting the behavior of people who know they are being observed - whether by other humans or by some computer program - is virtually like the proverbial blind squirrel seeking nuts in the Atacama Desert (it won't find one unless some one puts a nut in its path).