Based on observing this and other slow motion videos of cloud-earth lightning I observe the following:
Some low-intensity "pilot lightning" starts in the cloud, usually only at one point or at most a few points, and propagates in many directions away from that point, branching out much like a river delta or a tree branch.
The general direction of propagation is towards the ground, but not very strongly. The most consistent trait of the movement is away from the origin, but not even that is absolute (there are a few loops and upward branches).
Once one of the branches touches or gets close to the ground an intense arc forms between that point on the ground and the origin of the "pilot lighning". There is no clear indication of a direction, partly because of the unavoidable camera overload.
The pilot lightning starts where the gradient of the inhomogeneous electric field is steepest, somewhere close to the concentrated collection of charge in the cloud. It does not start on the ground because the ground conducts comparatively well so that the charges dissipate, lowering the electric field strength.
The pilot lightning's propagation is driven by two factors: The electric field and a feedback mechanism. The feedback happens because the plasma corridor created by the emerging lightning bolt conducts electricity very well and thus facilitates propagation once it has started: Electrons and ions are accelerated and extend the corridor at its front. Apparently there is a chaotic element to this, possibly because turning air into plasma is an explosive process. This multi-directional explosive expansion is also responsible for the branching.
That the lightning roughly follows the field gradient explains the general direction. The feedback sustains the lightning once it has started. The explosive expansion of the heated air explains its chaotic parts.
When one of the pilot lightning branches gets close enough to the ground that an arc to the ground is completed there is a completely conductive connection between the separated charges which will lead to the runaway discharge known as a lightning bolt. (It's a runaway process because higher currents increase the arc's conductivity by turning more air into plasma, so the current grows until the charges are exhausted.)
I do not think that the eventual main lightning bolt has a different "direction" than the pilot lightning, and I cannot find observational evidence. Even at 100,000 frames per second the intensifying appears instantaneous, and the main bolt is so bright that cameras are overloaded.
It may appear to the human perception as a "back strike" because it starts once the pilot lightning touches ground, much like a bounce-back event. But I think it's just that suddenly a lot of charges flow where only few charges were flowing a millisecond earlier, in the same direction.