Palantir is operating under the idea that "The AI will figure it out once we can access all the data"
This is a mistake because:
1. A lot of the data at big orgs is garbage or only understandable within a certain context by specific people.
2. Internal politics within lots of organizations prevents access to this data.
3. The AI cannot just figure it out. You need tons of humans in the mix which brings you back to #1 and #2.
Word on the street is a couple years back Palantir knocked a good chunk of SOCOM's intelligence capabilities offline for a few weeks.
Intelligence analysts like to use Communities of Interest "COIs" to keep track of things. These are kind of like wikis. A COI on the Middle East may have groups like Al-Qaida and ISIS, and people like MBS, and countries like Saudi Arabia and links between them. They can also have hierarchies, like "planes" split into "prop" and "jet" which in turn splits into "military" and "civilian," think decision trees.
One problem with these is they can take a while stand up. COIs need at least a few hundred objects in them before they can be useful. Palantir came along and said, "well, we can help stand them up with our AI."
They did their work, apparently didn't properly test it and just did things in production (maybe SOCOM doesn't have a dev environment, who knows?), and when they pushed run it created hundreds of thousands of objects throughout the COIs, with levels in the hierarchy that didn't make any sense. Like, imagine "747" falls under "Helicopter" which falls under "Cessna" which is under the "Nation" category.
It took weeks for SOCOM to rebuild things. I'm not sure if they had problems with backups or maybe they didn't properly have things backed up.
This is a mistake because:
1. A lot of the data at big orgs is garbage or only understandable within a certain context by specific people. 2. Internal politics within lots of organizations prevents access to this data. 3. The AI cannot just figure it out. You need tons of humans in the mix which brings you back to #1 and #2.