The thing you have to remember is that you, as the player, can ALWAYS play for the long game. Research the techs that don't make an impact until years later; the Philosophy line, the +literacy line, Medicine, etc. You can do that because if something goes terribly wrong in your game, you can just quit and restart. The AI has to play every game like it's the only shot it gets. With that mindset, it makes A LOT more sense to grab RGO and military techs instead of Philosophy and +literacy.
Philosophy is a no brainer that pays for itself within 3-5 years, maybe apart from very few certain circumstances, like starting colonisation or plannung an offensive war with tech advantage. So the AI should really research it asap (apart from 1870). Education techs are another thing, esp. for large nations like Russia.
The problem is that you can only weight techs, so there is ALWAYS a chance that the AI will chose something else. What might help to solve the problems, though, would be a tech check. So for example, if the AI does NOT have the tier II power tech, it has a much smaller chance to research other tier III techs from the tree.
Another problem is, that the player can exploit the AI if it has a too obvious research pattern. In HoI2 you could assign different building templates to nations, so for example in one game GER did built a submarine fleet, in another it built a surface fleet, in another one there was the chance it would build a CV fleet and in a another game it didn't build any navy at all - so no game was always the same. However, the AI in AHD is quite predictable - and if the AI would focus on the strongest techs too much, it would become even worse.
The ideal thing would be a reactive AI that "learns" during the game, but that is probably extremely difficult to script with the given system. For example, most of the time as Chile, you can sit back cause nobody will attack you. You can even disband most of the army and forgo army techs, giving you a huge early game advantage that will become an even greater late game advantage. Now if the Argentinan AI would become smarter after it has seen Chile sitting around with 3 brigades for 20 years, it would start teching some land techs, build a few more units and grab that ripe Chilean territory.