You would be incorrect.
"In physics if you do an experiment and some of the results appear to be wrong, you would normally eliminate them, rather than adjust them to match your theory. You would then redo your experiment. In climate science, the technique seems to be to change the data to what you think it should probably be. Based on how you 'feel' about it."
You are quite incorrect. The larger part of the decision about how to handle outliers in raw data depends on the nature of the experiment. If you have all the time and money in the world and are conducting a controlled laboratory experiment then the choice may be to redo the experiments. Unfortunately, the Earth and it's environment isn't an ideal controlled experiment.
A second issue is that, in fact, the laboratory experiments definitively prove that the addition of GHG to the atmosphere raises temperature. The absorption spectrum of CO2, methane and other GHG are well established without any raw GMT data. CO2 and other GHG are identical in nature as they are in the lab.
A third issue is simply that the raw data does, in fact, prove that the global mean air and sea surface temperature has increased since the beginning of the industrial revolution and the use of fossil fuels. And this increase is caused by CO2 and other GHG.
The goal of homogenization of the raw data is primarily to deal with the lack of controls in the weather stations. One of the largest issues is that weather stations are not nicely spaced about the globe. If you were measuring the temperature in your house, with one room containing two thermometers while others has only one, each room being of different sizes, simply taking the average of the raw data will not yield a precise and accurate mean. If each thermometer is of a different manufacture and construction, they are not likely to be all yielding the same number given the same conditions. All in all, the goal of homogenization is to achieve the most precise measure of GMT given the lack of laboratory control.
In a perfect and ideal world, you would be right. Unfortunately, in the real world, science has to deal with real data and doesn't have the luxury of simple re-running the experiment (the year 1959 has come and gone. Science can't rewind the clock.) or throwing out measurements because they aren't good enough (measuring a bucket of water hauled up over the side of a ship is not the same as measuring the temperature at the intake of the ships cooling system.)
In a perfect world, you would be right. Real scientists have to do real science in a real world.