We seek $x\in\Bbb{R}$ to minimize the distance between $(0,b)$ and $(x,x^{2})$. This amounts to minimizing the value of $$\sqrt{(x-0)^{2}+(x^{2}-b)^{2}},$$ and this gives the same value as the problem of minimizing $$x^{2}+(x^{2}-b)^{2}.$$ To do this, we set the derivative equal to zero and solve:
\begin{eqnarray*}
2x + 4x(x^{2}-b) & = & 0,\\
\implies 2x[2x^{2}-2b+1]=0,\\
\implies x=0\:\text{ or }\:x^{2}=b-\tfrac{1}{2}.
\end{eqnarray*}
Of course, the latter is only valid for $b\geq 1/2$.
Therefore if $b<1/2$, then the minimum distance is $\sqrt{0+(0-b)^{2}}=|b|$. If $b\geq 1/2$, then the minimum distance is either $|b|$ or is $\sqrt{b-\tfrac{1}{4}}$, whichever is smaller, which by inspection is the latter.
As a check, note that $\lim_{b\to1/2^{-}}|b| = 1/2 = \lim_{b\to1/2^{+}}\sqrt{b-\frac{1}{4}}$, which satisfies our intuitions.
It is worth asking what the relevance is of the solution $x=0$ in the case $b\geq1/2$. Some thought (and perhaps a good diagram) shows that $x=0$ in fact gives a local maximum for the distance.
In $\mathbb{R}^{3}$, when we have $z=x^{2}+y^{2}$ and $(0,0,b)$, we can reduce the problem to the two dimensional one due to the rotational symmetry of the paraboloid and the fact that the point $(0,0,b)$ lies on the axis of symmetry. The major difference between the two situations is that instead of two points giving the same minimum distance ($x=\pm\sqrt{b-\frac{1}{2}}$ above), we instead have infinitely many (in fact, the points of minimum distance will form a circle, the intersection between some cone with vertex at $(0,0,b)$ and the paraboloid $z=x^{2}+y^{2}$).