If you’re imagining a dystopian dictatorship, then sure; a single agent can be in “charge of government and political decisions”. Perhaps that would work better in countries that are currently dictatorships by removing human selfishness and corruption from the equation. I doubt much empathy already exists in those governments, so the replacement seems like an improvement. But, dictatorships exist for those who have taken power, not for the people. Would they benevolently relinquish control to anyone else including a machine? Unlikely.
If you’re imagining a democracy, then by definition this is not going to be the case.
AI is a tool, like any technology. It can be a powerful tool helping make things more efficient, providing creative solutions, and supporting decisions with large amounts of real world data rather than guessing. But, even a human-level AI would not be given so much power to turn a democracy into an autocracy. Democracies work (as best as they can!) by bringing in different perspectives, experiences, and cultures. An AI would simply bring one perspective to the table. AIs could have a voice in government and political decisions, but that would be shared with other groups in a true democracy.
There are other ways that AIs can be used in government aside from leadership. Imagine AIs assigned to groups of people, providing them with the type of representation equivalent to the best senators or congresspersons. Or, annually optimizing tax laws for fairness and to balance budgets. Or, optimizing natural resource allocations. Or, in drafting fairer laws without loopholes. The Affordable Care Act was originally over 2,000 pages and has expanded to over 80,000 by 2016 (How many pages was the Affordable Care Act when it went to a vote and how many legislators likely read every word of it?). That’s a tough weekend read for most people. If ethical human-level AIs with legal degrees and expertise in healthcare were employed to read that document, one could expect huge improvements. Combine that with empathy provided by humans in government, iterate, and perhaps agreement can grow between opposing viewpoints.
AIs can be quite helpful in government and politics without having to become our overlords.