Let $\{(Y_i,X_i)\}_{i=1}^n$ be a sequence of random pairs satisfying the model $$Y_i=f(X_i)+u_i$$ where $E(u_i)=0$ for all $i\in\{1,\dotsc,n\}$. The regression function is defined as $E(Y_i| X_i)=f(X_i)$ for all $i$.
My question is: without further conditions, is $E(Y_i|(X_i,X_j))=f(X_i), i\neq j,$ true? That is, the expected value of $E(Y_i|X_i)$ is changed when two observations $(X_i,X_j)$ are given? If not, when it is true?
Comments: I know that the sigma-algebra generated by $X_i$ is contained in $\sigma(X_i,X_j)$. So, I suspect that the conditional expectation is changed if there were some dependence structure between the observations.