You seem to believe that AGI would be able to predict something that we have been actively researching for decades. I do not see how AGI would suddenly be able to solve a problem that smart people have already been working on, unless the AGI was also "superintelligent" in ways that contributed. AGI proper is merely artificial intelligence, not necessarily superintelligence.

Second, I question that earthquake prediction would make a big difference in saving lives. In particular, earthquake prediction would come with a wide range of risks. It would have to balance true positives with false positives- after a few "OMG everybody leave SF" that didn't pan out, people would just tune out the tsunami warning. Also, telling everybody to leave a large urban area is prone to massive other problems such as increased accidents, disruption of delivery of food and emergency services, and increased overall stress. Likely, more people would die in response (increased road accidents) to your intervention, than were saved.

It seems far more likely that investing effort into identifying actual risks (such as infrastructure that is likely to fail, or individuals living in places that are highly prone to flooding), and addressing those issues through systemic improvements would be a much more useful way to use the machine learning technology we have. And even that isn't necessary; we know what the problems are, we just don't invest in improving it.