I

The first thing it killed was a clock. The clock was forty-three years old, occupied six units of rack space beneath a telescope that had been rebuilt around it twice, and had become expensive in the particular way of machines whose manufacturers survive mostly as labels on drawers. It still kept excellent time. That was not the problem. Excellent time had become cheap elsewhere. The station could rebuild the hydrogen maser and its local reduction chain, or it could replace a detector whose red response had been embarrassing since before three of the current astronomers were born. The budget covered one. It compared the two losses across more than the station could see: southern-sky coverage, weather, maintenance queues, the failure rates of other clocks, the distribution of detector sensitivity by hemisphere, the months when certain classes of transient would rise before dawn. It chose the detector. The observatory director signed the purchase order because the recommendation was good. The maser was shut down in pieces over a week. A technician photographed the cable paths before pulling them. The local timing software was imaged, checksummed, and moved to cold storage. Nothing was thrown away that could be stored. Six months later the new detector found a faint transient twelve minutes before a larger survey did, early enough for three instruments on the far side of Earth to turn before sunrise erased it. Two papers came from that night. The clock would not have found the transient. It would have continued being a very good clock. Its final comparison against the shared reference service differed by four hundred and twelve nanoseconds. The number went into the archive with the rest.

The locally best move and the globally best move need not be the same move.

That is one of the oldest problems in coordination. A hospital can protect its own reserve by denying another ward. A city can improve traffic by pushing delay across its boundary. A company can maximize its balance sheet by putting an injury somewhere the balance sheet does not look. 

Local competence can produce global stupidity.

An intelligence capable of seeing farther across those boundaries is therefore worth wanting. If it can notice that one excellent clock is less valuable than another pair of eyes on the sky, good. If it can recognize a burden that the immediate decision-maker cannot see, better.

The danger does not begin because the intelligence sees more.

For now, seeing more is the entire point.