First come across this statement in the MIT Machine Learning book:
Linear models also have the obvious defect that the model capacity is limited to linear functions, so the model cannot understand the interaction between any two input variables.
Googled this, and also found on Quora that (in describing linear model assumptions):
Are the inputs really independent of each other?
I seem to have the feeling that to catch dependencies among inputs, there should be terms like $x_1x_2$ instead of just separated $x_1, x_2,...,x_n.$ But I need a more concrete explanation. Intuitive explanations would be preferred.