An alternative way to look at it:
It is enough to find an injective map
$$\mathbb{Q}[u_1, \ldots, u_n]\to \mathbb{C}$$
Consider the canonical map from the ring of polynomial in $n$ variables $x_i$, $i=1,n$ to our domain
$$\mathbb{Q}[x_1, \ldots, x_n]\to \mathbb{Q}[u_1, \ldots, u_n]$$
This map has a kernel $\mathcal{P}$, a prime ideal of $\mathbb{Q}[x_1, \ldots, x_n]$. Now, a map $\mathbb{Q}[u_1, \ldots, u_n]\to \mathbb{C}$ will be given by an assignment $u_i \to t_i$, $t_i \in \mathbb{C}$ such that $t=(t_1, \ldots, t_n)$ satisfies all of the equations $P(t)=0$, where $P\in \mathbb{P}$. Note that since $\mathbb{P}$ is finitely generated ( Hilbert), we may consider only finitely many equation $P_1$, $\ldots$, $P_m$. In other words, a map $\mathbb{Q}[u_1, \ldots, u_n]\to \mathbb{C}$ is given by a choice of a point $t$ in the zero set $Z$ in $\mathbb{C}^n$ of the ideal $\mathcal{P}$. Now, when is the map given by $t$ injective? If and only if $t$ satisfies no other polynomial equations over $\mathbb{Q}$ than the ones in $\mathbb{P}$. In other words, an injective map $\mathbb{Q}[u_1, \ldots, u_n]\to \mathbb{C}$ is given by choosing a "generic point" in the zero set $Z$. Now, if we take a "random point" in $Z$, how often will that be generic? Almost always, and this could be argued with measure theory, since non-generic points lie on a countably union of lower-dimensional subvarieties of $Z$, and thus form a subset of $0$ measure.