Also, in my child’s case her math capability far exceeds her fine motor skills. So she finds the interfaces for some games really difficult (not to mention we’re using my Mac which has a mouse suited for my fat hands, not her dainty fingers). That said, she gleefully exclaims “dad, move the 3 beads next to the 5 beads and that makes 8 which is 1 bead less than the 9 we just had”. So to me it seems that the UI must adapt to kids who have the concepts ok, but lack the dexterity to express that “in silico”.

And in general, parents need to be the primary mentor — the game shouldn’t try to be the authority, regardless of how sophisticated its neural net back-propagation. So it should facilitate (even encourage) parental guidance (vs. direct parent answers which of course is counterproductive.)

Perhaps you should treat this as a feature opportunity — suggest questions to parents when the child is struggling. For example, many of your games simply repeat the instructions over and over if the child is taking time to respond. Why not detect the delay and prompt parents with appropriate questions? Then parents are learning how to guide their child appropriately!

In summary, while I think the learning algorithms are geeky cool, the reality is that parents need to learn how to teach their kids more than we need an impressive Bayesian classifier. Change the learning algorithm —- not the parent role as lifelong math mentor. Let it adapt to parental guidance (even alert us if it suspects we are being over zealous)…”Hey Dad, Junior’s last 5 answers have been 100% correct and really fast…are we helping too much??”