Is it possible to read neural activity from the outside of the head and if you had an machine learning algorithm that could learn what the signals "mean" (for example moving the right arm)? Could that be achieved with having receptors at special places on the head? And what type of receptors that would need. I don't have any experience or knowledge with/from the brain. Sorry if this is a stupid question.
It's not a stupid question in my opinion.
However, it is perhaps naive. Brain function is very complex and how the different parts of the brain exactly relate to memory and function is still being worked out.
We can monitor neural activity outside the brain by a number of methods. The most common one you might have come across is an electroencephalogram (EEG). This measures physiological electrical activity at the surface of the brain (remember, the brain is a 3D structure), so this can't measure activity within the folds of the brain. It is generally done with electrodes placed outside the skull, so non-invasive and produces no harm, even if used long-term. It's closely related to the electrocardiogram, which measures electrical function of the heart. You will have seen an ECG on TV or in a movie - it is often portrayed as a regularly beeping device that goes emits a long sustained beep when someone dies.
A similar system, though invasive is electrocorticography (ECoG), which uses electrodes implanted inside the skull. It requires surgery to be performed so it is much more risky to do. Like the EEG this measures physiological function and can not be used to read purpose into the signals. I believe this system can be used as an interface/controllers for prosthetics such as hands/fingers, but I don't know how advanced this is and what is involved in getting it to work. I don't think, however, that this is a true reading of function as such, rather that the user is trained to set off a signal for the ECoG to read.
There are much more complex and functional methods of actually reading brain function and applying it to precise signals within the brain. All of these methods require the use of large equipment such as Positron Emission Tomography (PET) or Magnetic Resonance Imaging (the particular application here is called functional MRI or fMRI). These all require the patient to lie very still inside a large noisy piece of equipment to measure some aspect of the brain interpretable by equipment/software in question. These methods can be quite specific and do show the interior of the brain.
I don't know much about these methods other than general principles. However, fMRI measures blood flow to a particular area of the brain and requires the person being scanned to concentrate on the topic at hand while the scan is being performed. PET and the related technique Single-Photon Emission Computed Tomography (SPECT) often require the use of an injectable material that the scanner can pick up - this is usually (always?) a mildly radioactive one, often conjugated to a specific biological molecule (see here for a list of radioactives and localizations) so that it is trafficked in the body to a specific location.
In conclusion, people are still working on brain function and assigning particular things to particular locations, so we are still a fairly long way from having methods that can be used to natively control devices via brain function.