[guided]We first build the map on the independent bases, because the bases are the only non-model-theoretic data distinguishing $M_I$ from $M_J$. Since $I$ and $J$ have the same cardinality, there is a bijection
\begin{align*}
f_0:I &\to J.
\end{align*}
The sets $I$ and $J$ are [independent](/page/Independence) subsets of the $A$-definable [strongly minimal set](/page/Strongly%20Minimal%20Set) $D$, so independence over $A$ is being taken in $D \setminus A$. Hence $I \cap A=\varnothing$ and $J \cap A=\varnothing$. We may therefore extend this map to $A \cup I$ without any conflict by declaring that it fixes the parameter set $A$ pointwise: for every $a \in A$, set $f_0(a)=a$.
The point of the hypothesis on finite independent tuples is precisely that this arbitrary bijection is already elementary on finite pieces. Let $i_1,\dots,i_n \in I$ be distinct, with $n \in \mathbb{N}$. Because $I$ is independent over $A$, the tuple $(i_1,\dots,i_n)$ is independent over $A$. Because $f_0$ is injective and its image lies in the independent set $J$, the tuple $(f_0(i_1),\dots,f_0(i_n))$ is also independent over $A$. The assumption therefore gives
\begin{align*}
\operatorname{tp}(i_1,\dots,i_n/A)
=
\operatorname{tp}(f_0(i_1),\dots,f_0(i_n)/A).
\end{align*}
This equality of complete [types](/page/Type) over $A$ says exactly that every formula with parameters from $A$ true of $(i_1,\dots,i_n)$ is true of $(f_0(i_1),\dots,f_0(i_n))$, and conversely. Hence the restriction of $f_0$ to $A \cup \{i_1,\dots,i_n\}$ is an [elementary map](/page/Elementary%20Map) over $A$. Since the finite tuple was arbitrary, every finite restriction of $f_0$ is elementary over $A$.[/guided]