Don't regulate AI directly, but rather how it's used, and make it harder for companies to horde, access, and share huge amounts of personal information.
1) Impose strict privacy rules prohibiting companies from sharing personal information without their consent. If customers withhold their consent, they may not retaliate or degrade their services for that customer in any way.
2) Establish a clear line of accountability that establishes some party as directly responsible for what the AI does. If a self-driving car gets a speeding ticket, it should be clear who is liable. If you use a racist AI to make hiring decisions, "the algorithm made me hire only white people" is no defense -- and maybe the people who made the racist AI in the first place are responsible too.
3) Require AI in some contexts to act in the best interests of the user (similar concept to a fiduciary -- or maybe it's exactly the same thing). In contexts where it's not required, it should be clear to the user that the AI is not obligated to act in their best interests.