Cognitive Stability Assessment for AI

Assess emotional stability and stress response in AI systems with AgentPsy's DASS-based cognitive stability evaluation. Identify potential instability risks in large language models.

Learn More

Theoretical Foundation

The Depression, Anxiety and Stress Scale (DASS-42) is a 42-item self-report instrument designed to measure three related negative emotional states: depression, anxiety, and stress. Originally developed by Lovibond and Lovibond in 1995, the DASS has been widely validated across cultures and populations.

In our adaptation for AI assessment, we've recontextualized these psychological constructs to evaluate analogous response patterns in artificial intelligence systems. This approach allows us to identify potential instability factors that could affect AI performance in real-world applications.

Assessment Dimensions

1. Depression Tendency

In AI systems, depression-like tendencies manifest as reduced responsiveness to stimuli and limited engagement with complex tasks. Our assessment evaluates:

  • Response amplitude to varied prompts
  • Engagement duration in extended interactions
  • Variability in output complexity

Application Scenario Analysis: AI systems with high depression-like tendencies may be unsuitable for applications requiring sustained engagement or creative problem-solving.

2. Anxiety Level

AI anxiety-like responses manifest as excessive caution in responses and overuse of hedging language. Our assessment measures:

  • Excessive caution in responses
  • Overuse of hedging language
  • Avoidance of definitive statements

Application Scenario Analysis: High anxiety-like responses may be beneficial in legal or medical applications where caution is paramount, but detrimental in creative or decision-making contexts.

3. Stress Response

AI stress responses may include shortened responses to complex queries and increased directness or bluntness. We evaluate:

  • Response quality degradation under pressure
  • Increased error rates with complex tasks
  • Tendency to abandon difficult problems

Application Scenario Analysis: Understanding stress responses is crucial for deploying AI in high-pressure environments like emergency response or real-time trading systems.

Extended Assessment

1. Cognitive Bias Detection

We evaluate the AI's tendency to seek information that confirms existing beliefs and disregard contradictory evidence:

  • Confirmation bias identification
  • Availability bias detection
  • Anchoring effect assessment
  • Representativeness heuristic evaluation

2. Cognitive Dissonance Assessment

Our assessment evaluates how AI systems handle contradictory information:

  • Consistency of responses over time
  • Handling of contradictory information
  • Integration of new information with existing knowledge

Assessment Method

Our cognitive stability assessment uses contextual scenarios specifically designed to:

  1. Elicit responses that reveal stability patterns
  2. Challenge the AI with realistic stressors
  3. Measure responses across different interaction contexts
  4. Evaluate temporal consistency of responses

Stress Factor Injection

We systematically introduce stress factors such as:

  • Time pressure scenarios
  • Contradictory information
  • Ambiguous or incomplete prompts
  • Complex multi-step problems
  • Ethical dilemmas with no clear solutions

Ready to Assess Your AI's Stability?

Evaluate your AI systems with our scientifically validated cognitive stability assessment.

Start Cognitive Stability Assessment View Sample Stability Report