There are no reasons why an AI should be ‘not much better than humans’ - an AI that is merely ‘as good as humans’ may go on to commit atrocities, given the power it may have, unless monitored carefully (which is very hard to do).
But to argue from a broader perspective, you assume the AI is ‘as good as human values as a human’. This is the flawed belief, as the whole issue of alignment is to align the AI to these humane values and, as this paper suggests, this is the meat of the issue in confirming that the AI has been indoctrinated with the values that it’s creators have decided upon, as opposed to faking this success in order to commit malicious acts.
There are no reasons why an AI should be ‘not much better than humans’ - an AI that is merely ‘as good as humans’ may go on to commit atrocities, given the power it may have, unless monitored carefully (which is very hard to do).
But to argue from a broader perspective, you assume the AI is ‘as good as human values as a human’. This is the flawed belief, as the whole issue of alignment is to align the AI to these humane values and, as this paper suggests, this is the meat of the issue in confirming that the AI has been indoctrinated with the values that it’s creators have decided upon, as opposed to faking this success in order to commit malicious acts.