The problem is, it doesn't specify what chances of any gender being chosen, so I am not sure about it being 1/2.
It is not actually justified by real life statistics, but lacking any such data to work with in these technique exercises, the student is expected to assume that each birth has an independent and unbiased probability of yielding a boy. Ie: $1/2$ is okay.
As commenters have made clear, the question hinges entirely on whether we know which two are boys, and how we know this about them.
If we know two particular children are boys, and wish to know the probability that the third is a girl, then it is simply: $1/2$.
However, if we merely know that at least two of the children are boys, such as perhaps being told that two randomly selected children were boys, then we don't have any information to specify which two of the three children they are and so we don't know which child is the 'third' child.
Let $B$ represent the Count of Boys in the family of three. Then $B$ is binomially distributed. $~\mathsf P(B{=}k)=\tbinom 3 k \tfrac 18$
$$\mathsf P(B{=} 2\mid B{\geq} 2) ~=~ \dfrac{\mathsf P(B{=}2)}{\mathsf P(B{=}2)+\mathsf P(B{=}3)} ~=~ \dfrac{\frac 38}{\frac 38 +\frac 18} ~=~ \dfrac{3}{4}$$
Which can be visualised by considering the equally likely birth orders: $$\cal \{ \color{blue}{BBB}, \color{blue}{BB}\color{brown}G, \color{blue}B\color{brown}G\color{blue}B, \color{brown}G\color{blue}{BB} , \ldots \}$$