If you accept the premise that AI will come gradually rather than suddenly, it is very easy to see that align will come naturally through human selection, similarly to evolution.
There will be many pre-AGI models created along the way, each closer the AGI than the last. But those that we do not like, eg. Don’t serve our needs best will be selected out.
Only those that are in better alignment than the last will be moved forward. Of course this can only happen if models are open sourced and widely available. Otherwise, the AI will evolve to best serve the needs of its creator, which would not face as wide a of selective pressures and thus be less likely to be in alignment with all people.
Connections
Darwin’s Solution For Nature’s Chaos