I do not think that a closed form could exist and, for sure, as uou did, a numerical method will be required.
Instead of looking for the zero of function
$$f(x)=x^{x^{x^x}}-2020$$ which is varying so fast, search for the zero of
$$g(x)=\log \left(\log \left(\log \left(x^{x^{x^x}}\right)\right)\right)-\log \left(\log \left(\log \left(2020\right)\right)\right)$$ which is quite smooth. A plot of $g(x)$ reveals that the solution is close to $x=2$.
Use one iteration of Newton method with $x_0=2$ and you will need to solve for $x$
$$\frac{(x-2) \left(8+64 \log ^3(2)+64 \log ^2(2)+32 \log (2)\right)}{\log (65536) \log
(\log (65536))}+\log (\log (\log (65536)))=\log \left(\log \left(\log \left(2020\right)\right)\right)$$ Then, you have a nasty expression for $x_1$ which is $1.9448$. But continuing iterations, the pathe to solution will be
$$\left(
\begin{array}{cc}
n & x_n \\
0 & 2.000000000000000000000000 \\
1 & \color{red}{1.94}4787317029677396906107 \\
2 & \color{red}{1.9466}19903004071435441212 \\
3 & \color{red}{1.9466223879}75658343003294 \\
4 & \color{red}{1.9466223879801945531843}44 \\
5 & \color{red}{1.946622387980194553184359}
\end{array}
\right)$$
Let us see what higher order methods (Halley and Householder) would give
$$\left(
\begin{array}{cccc}
n & \text{Newton} & \text{Halley} & \text{Householder} \\
0 & 2.0000000000000000000 & 2.0000000000000000000 & 2.0000000000000000000 \\
1 & 1.9447873170296773969 & 1.9465058701098434768 & 1.9466160957701974640 \\
2 & 1.9466199030040714354 & 1.9466223879816314805 & 1.9466223879801945532 \\
3 & 1.9466223879756583430 & 1.9466223879801945532 & \\
4 & 1.9466223879801945532 & &
\end{array}
\right)$$
Inverse symbolic calculators do not identify the solution but the funny is that it is quite close to the reciprocal of the real root of the cubic
$$39149 x^3-72556 x^2+71412 x-22845=0$$ which would give $20$ exact significant figures.