Managing Cognitive Risks in Agentic AI

# Illustrative sketch (not from the paper)
import torch
# Define cognitive levels
cognitive_levels = ["physical", "social", "self-referential"]
# Map each level to risk categories
risks = {
    "physical": ["loss of control over actuators"],
    "social": ["manipulation of human autonomy"],
    "self-referential": ["erosion of human agency"]
}
# Map each level to mitigation strategies
mitigations = {
    "physical": ["sandbox containment", "AI generation detection"],
    "social": ["depersonalization", "human-in-the-loop oversight"],
    "self-referential": ["human-in-the-loop oversight", "sandbox containment"]
}
# Load a placeholder LLM (e.g., GPT‑like) – here a dummy torch.nn.Module
class DummyLLM(torch.nn.Module):
    def forward(self, x):
        return x  # placeholder
model = DummyLLM()
# Simulate risk assessment loop
for level in cognitive_levels:
    # In practice, evaluate model outputs relevant to the level
    print(f"Assessing {level} cognition:")
    print("  Risks:", risks[level])
    print("  Mitigations:", mitigations[level])
# End of sketch

// Illustrative sketch (not from the paper)
const torch = require('torch-js'); // placeholder for torch in JS
// Define cognitive levels
const cognitiveLevels = ["physical", "social", "self-referential"];
// Risks per level
const risks = {
  physical: ["loss of control over actuators"],
  social: ["manipulation of human autonomy"],
  "self-referential": ["erosion of human agency"]
};
// Mitigations per level
const mitigations = {
  physical: ["sandbox containment", "AI generation detection"],
  social: ["depersonalization", "human-in-the-loop oversight"],
  "self-referential": ["human-in-the-loop oversight", "sandbox containment"]
};
// Dummy LLM class
class DummyLLM {
  forward(input) {
    return input; // placeholder
  }
}
const model = new DummyLLM();
// Simulate assessment loop
cognitiveLevels.forEach(level => {
  console.log(`Assessing ${level} cognition:`);
  console.log("  Risks:", risks[level]);
  console.log("  Mitigations:", mitigations[level]);
});
// End of sketch