How To Lose Your Class In One Easy Step
Watch this video on tree/graph search. It’s part of an interesting and informative introduction to AI published by Stanford. Which isn’t without its flaws, it appears… :)
As far as I can tell, the goal of the video is to show that keeping track of explored and frontier states allows you to pull trees out of graphs without polluting your tree with useless looping traversals. The course seems to be segueing from its tree search definition (as defined here) to its graph search definition (as defined here), but it’s done in a really confusing way.
The lecturer shows A->S->A (a repeated visit to A, a consequence of tree search over an undirected graph), and then at the end explains that A->T->L->T isn’t a needed path because the final step is a… ‘regressive step’! (graph search) How can both these things be present on the same diagram! They shouldn’t be…
Not only that, but the lecturer also completely mis-labels the explored (blue) and frontier (green) states on his diagram.
The commenters on the youtube video correctly identify both of these points.
One way to communicate the intended points more clearly would be to a) correctly label the frontier and explored states, and b) use A->S->A as an example of a regressive step, but then scratch it out before going on to show that A->T->L->T is also a regressive step that is avoided by the use of frontier and explored state tracking.
Just goes to show that you need to stay on your toes, even when learning from (supposedly) the best. Everyone makes mistakes!*
The trouble is, this confused segue completely obscures the point that is trying to be made. If the class doesn’t stop and unravel the confusion before continuing, their understanding of what follows is in jeopardy because they are likely to have a broken mental model of what’s going on (as they consciously or unconsciously try to mentally resolve, or ignore, contradictions and confusions in their minds). Unfortunately, unravelling the confusion is much harder than following a correct explanation (and because the lecturer does not notice and acknowledge the mistake, it is not natural for the class to take it as one due to the authority dynamic…). Further, I suspect that unravelling the confusion is often harder than working it out for yourself from first principles. I suspect this is why self-learners often do pretty well (if a little slowly): they have less opportunity to fall into the trap of being accidentally misled by their teachers.
* Strangely, I have yet to find a serious mistake in any of the Khan Academy videos… what’s going on there? Could they have been… reviewed…!?!?
