Yoshua Bengio: Obtaining Safety Guarantees to avoid AI Catastrophic Risks