Eh? He was just bringing up one of the standard counterexamples to the kind of simplistic utilitarianism that Harris is advocating. I don't know what the "your idea kills children argument" is, but he certainly wasn't making any such argument.
>It would require statistical modeling, and some notion of community-accepted confidence thresholds, to come to a conclusion in a full utilitarian framework.
That's the rub. It doesn't require any of those fancy methods to know that it would be wrong to kill the child. Hence the implausibility of full-on utilitarianism as a reasonable moral philosophy.
(Of course, one can reject full-on utilitarianism without denying that the consequences of our actions for human happiness play a very big role in determining whether they're right or wrong.)
This sounds suspiciously like the "Your idea kills children" argument. I call strawman.
The problem with this scenario is that it vastly oversimplifies utilitarianism, presuming that only deaths matter and that a utilitarian model of ethics would demand the death of the child; an emotionally repugnant act.
In fact, the quantifiable harm from society intentionally killing a child due to its potential risk could be far greater than the eventual potential deaths from letting that child survive. It would require statistical modeling, and some notion of community-accepted confidence thresholds, to come to a conclusion in a full utilitarian framework.