Machine learning algorithms can help to find patterns (coherence) in the "ingredients" (data). As long as the conditions aren 't changed, taking these patterns as it 's starting point, it can extrapolate the ingredients to the (near) future, showing some future developmental possibilities (again, as long as the composition of all ingredients together and the many contexts remains largely unchanged, which in real life is hardly ever the case). So AI doesn 't really "predict" anything, but it 's just a smartly designed reversed kitchen blender for data. Garbage in, garbage out, strawberries in, strawberries out. Again, predicting is only possible if you know exactly what you put in and if nothing changes during the ride.