We are dancing around a conversation about accountability. Automated processes (or algorithms, romantically) currently diffuse accountability. This is a problem

We are dancing around a conversation about accountability. Automated processes (or algorithms, romantically) currently diffuse accountability. This is a problem because we rely on rule of law to keep the peace. We would like corporations profiting off the algorithms to be held accountable, but this is hard today and lawmakers have moved too slowly on a solution. One alternative is to outlaw algorithms operating at a scale larger than our current accountability framework can handle. The reason this works is because it forces a human back into the chain of accountability. The corp will pass the potato and fire the employee who approved the ad. The employee's incentives are slightly more nuanced than you wrote - they want to maximize dollars while avoiding being fired. Since the employee is a regular human and not a corp with superlawyers and deep pockets, there's a better chance of criminal or civil followup targeting the former employee. With this accountable-scale bottleneck in place, we now have incentive for lawmakers and corporations to legislate the original problem and remove the bottleneck. That's why I think GP's proposal is a step forward. Even though the net harm may be the same at automation-scale and human-scale, the accountability envelope is closer to the desired end state.