TL;DR

Complex computer procedures are used to extract signals of the desired frequency from general radio noise, which likely makes such extraction unreliable. Could this factor be the cause of distorted radio signal data?

I recently read a paper by scientists about the results of a search for radio signals from extraterrestrial civilizations using the Chinese FAST telescope. After carefully screening out candidates, they were left with one particularly suspicious signal, emanating from the direction of the star K2-155. However, when they searched for similar signals at similar frequencies in data from the same observation day, they found similar signals (three, I think). The similarity was only in polarization direction; in other respects, they differed significantly. Those signals looked more like interference, leading the researchers to conclude that this signal could also be of terrestrial origin.
However, they were unable to link these signals to any external radio source, but they hypothesized that they could be instrumental interference.
This is where I'm unclear. The telescope itself shouldn't emit anything, as its purpose is to receive, not send, signals. The process of extracting narrowband signals from the general radio noise, as I've read, is a rather complex procedure performed by a supercomputer, which divides the general noise into millions of narrow frequency bands. The different polarization channels are obviously processed separately. Could it be that the absence (or lack of power) of the signals they found in another polarization channel was not due to their being highly polarized, but rather because the computer failed to identify them in the other channel or incorrectly estimated their power (underestimated)?
...
Link to the study: https://iopscience.iop.org/article/10.3847/1538-3881/ae8f34

There are several thousand satellites, any of which could have transmitted, or scattered a signal. There are too many sources of interference to consider. Polarisation can be affected by reflection from surfaces.

One problem with radio astronomy, is that the signals are variations of noise density over time, direction, and frequency. Noise can sometimes randomly reinforce, producing a statistically unlikely detection, but something that is nevertheless possible.

Transforms are used to detect patterns in the received signals. Transforms effectively convolute the received signals. A transform has a conversion gain, which is a lowering of the noise floor. That happens because each output data point, is the product of many input data points.

When there is little data, and no repeatable result, there can be no definitive conclusion.
Anything is possible.

I am not familiar with FAST telescope, but I can answer more generally.
In general, those supercomputers will not be tasked with anything similar to "AI" - at least not at the level of a simple signal detection. They will be tasked with applying well-defined "filters". In all cases, the purpose and limits of that "filtering" will be well-known.

As far as received-induced artifacts (instrument interference), there are many potential ways for the receiving circuit to mess up the recorded data. Even if the amplifiers did not actually create EM noise, the sampling clock might be slightly affected by power or ground plane issues - yielding a recording with the modern-day equivalent of the vinyl "wow" and "flutter".