Although we have reliable data of solar polar fields only from the mid-1970s, it seems that the polar field at a minimum is well correlated with the next cycle, but the strength of the cycle is not correlated with the polar field produced at its end. We explain this by suggesting that the Babcock-Leighton mechanism of poloidal field generation from tilted active regions involves randomness, whereas the other aspects of the dynamo process are more ordered. To model actual cycles, we have to ‘correct’ our theoretical dynamo model by ‘feeding’ information about the polar field at the minima. Following this process, we find that our model fits the observed sunspot numbers of cycles 21–23 reasonably well and predicts that cycle 24 will be the weakest in a century.