White House wants to vet powerful AI models for risks—a computer scientist explains why AI safety is so difficult Tech Xplore