We could start by not ignoring recommendation/amplification algorithms, especially when they clearly amplify bad arguments and obscure better arguments.