Your idea is ok. But true, that it needs some 'formalization'. We suppose $a_n<b$ (or else nothing to show). You may write in base 4 expansion $$4^b \leq \sum_i 4^{a_i}=\sum_{k=0}^{b-1} p_k 4^k+n_b 4^b$$ with each $0\leq p_k\leq 3$ and $n_b\geq 0$. We will construct by induction coarser and coarser (sub-) partitions of the set of $a_i$'s. In the beginning we set $S_0=([a_1],...,[a_n])$ in which each $a_i$ is a one element partition. The value of each element of a partition is the sum of the 4th power of its elements. In $S_0$ each element therefore has a value of the form $4^k$, with $0\leq k<b$.
If $0<p_0$ (remember $p_0$ is not greater than 3) then there must be $p_0$ elements of the partition $S_0$ of value $1$ (i.e. having $a_i=0$). You put those elements aside. The rest you regroup into the smallest possible partition elements to form multiples of 4. Call $S_1$ this new partition. The value of each element is then of the form $4^k$ with $1\leq k<b$.
If $0<p_1 (\leq 3)$ there must be $p_1$ of the elements in $S_1$ with value $4$ which you may set aside. The rest you regroup into a partition $S_2$ where each element of the partition has a value of the form $4^k$ with $2 \leq k<b$, etc...
In $S_{b-1}$ each partition element must have value $4^{b-1}$. If you have 4 of such elements then their union constitute a solution to your problem. If not then there are at most 3 such elements and adding the values of all you have set aside you get
$$ \sum_i 4^{a_i} \leq \sum_{k=0}^{b-1} 3 \times 4^k < 4^b$$
contradicting the hypothesis.
(Making a drawing of such partitions may help getting the idea).