Artificial Intelligence (AI) Agents Checks, Rules, Context

Examples of ways to control agent behaviour

• Rules, system instructions
• Context
  • Ontology it knows about, follows
  • Past behaviour instructions and input from humans
• Layers: each layer is focused on specific areas like doctrine, critic, ontology mapping
• Human alerts, flags
• Adversarial AI
  • Criticizes other AI output
  • Makes suggestions
• Run agent in restricted environment with specific permissions and account access