Here's what Soren said: "Based on my experience, I kind of knew up-front there's no way to make the A.I. use that appropriately," Johnson said. "When a human player looks at another human player's HQ, there's something they'll notice that's a little off. I see a steel mill, but they're not arranged in a super-optimal way. So it makes me suspect that one of those is a hologram. For an A.I. to know that, it almost has to be pattern-recognition, visual imagery software, which is way beyond the scope of the A.I. in this game."
This is just not true. There are simple algorithms for finding and placing holograms. I'll explicitly give one for Spies. You can calculate the "hologram-value" of someone else's production by deleting each not-yet-revealed tile, one at a time, and then calculating the worth of his remaining production. This gives a $ value for each tile. The ones that result in less $ loss are more likely to be hologrammed. Tiles with Goon Squads and colony adjacency get bonuses. Amongst all the available Spy locations, cull ones that reveal strictly less tiles than other Spy locations. Then, assign a probability to each remaining Spy location, and calculate Bonus/(hologram-value times the probability of uncovering). Sum all these terms, and choose the probability distribution that minimizes this sum. This distribution can be found by simple linear programming, shuffling 1% probability from one Spy bucket to another and recalculating the total sum until equilibrium is reached.
Hologram placement uses the same function in reverse.
To determine when to use holograms, the AI can estimate the power of the BM by comparing each sabotage's effect to the upfront resource and construction-time cost of the advanced building. When this power is large, the AI buys a hologram. This calculation should affect its tendency to buy the advanced building, since the advanced building is less likely to be hit with a hologram on.
To determine when to place spies, the AI counts the number of holograms bought, and then uses them there is a clear enemy, such as in 1v1, or FFA with a leader whose strength is comparable to everyone else combined.
Holograms that are not revealed by Spies are opaque to the AI. It's still possible for the AI to throw BM at suspected Hologrammed locations, since it can see the number of holograms on the field, and it can reuse the above calculations to guess where the holograms are.