Nyquist showed that to distinguish unambiguously between all signal frequency components we must sample at least twice the frequency of the highest frequency component. To avoid aliasing, we simply filter out all the high frequency components before sampling.
Note that antialias filters must be analogue - it is too late once you have done the sampling.
This simple brute force method avoids the problem of aliasing. But it does remove information - if the signal had high frequency components, we cannot now know anything about them.
Although Nyquist showed that provide we sample at least twice the highest signal frequency we have all the information needed to reconstruct the signal, the sampling theorem does not say the samples will look like the signal.
The diagram shows a high frequency sine wave that is nevertheless sampled fast enough according to Nyquist's sampling theorem - just more than twice per cycle. When straight lines are drawn between the samples, the signal's frequency is indeed evident - but it looks as though the signal is amplitude modulated. This effect arises because each sample is taken at a slightly earlier part of the cycle. Unlike aliasing, the effect does not change the apparent signal frequency. The answer lies in the fact that the sampling theorem says there is enough information to reconstruct the signal - and the correct reconstruction is not just to draw straight lines between samples.
The signal is properly reconstructed from the samples by low pass filtering: the low pass filter should be the same as the original antialias filter.
The reconstruction filter interpolates between the samples to make a smoothly varying analogue signal. In the example, the reconstruction filter interpolates between samples in a 'peaky' way that seems at first sight to be strange. The explanation lies in the shape of the reconstruction filter's impulse response.
The impulse response of the reconstruction filter has a classic 'sin(x)/x shape. The stimulus fed to this filter is the series of discrete impulses which are the samples. Every time an impulse hits the filter, we get 'ringing' - and it is the superposition of all these peaky rings that reconstructs the proper signal. If the signal contains frequency components that are close to the Nyquist, then the reconstruction filter has to be very sharp indeed. This means it will have a very long impulse response - and so the long 'memory' needed to fill in the signal even in region of the low amplitude samples.
