"'Difficult' problems are 'difficult' because their small number (usually very obvious) solutions are all unpleasant to someone." Or in the words of Jean-Claude Juncker, "We all know what to do, we just don't know how to get re-elected after we've done it."
...
So the problems of government are rarely just a wizard solution away from pleasing everyone, but at Matthew Parris puts it "have our cake or eat it" problems, where the dilemma comes hard-wired, guaranteed to annoy someone no matter what machine learning you throw at it.
...
It turns out economics, politics, management—they are all about people. Consequently, political skill is far less about selecting for genius than handling the enemies, knaves and fools.

None of this is news to a political adviser, and Cummings is a far more accomplished one than I am. So while he writes that his putative recruits will be having conversations about "Computational rationality: A converging paradigm for intelligence in brains, minds, and machines," I suspect he knows that, in the words of a colleague, the day job will revolve around "political and policy trade-offs, persuasion, compromise, coalition-building, arm-twisting and finely balanced judgment calls." He may threaten to "bin" anyone found playing office politics, but Downing Street is a political office, and that is what people play—and the people who play it best get things done.

Indeed, from my time there, the most effective official was never much of a science whizz, but a genius at the art of wringing progress out of the machine: brilliant at knowing the right person, crafting the best email, plotting the way through the chancellor in the right bar with the right Treasury official—a Machiavellian able to convince anyone he was on their side. He'd be a superb adornment to No 10. What a pity, I fear the rascal studied history.