Back to ISTQB Labs

Traditional vs AI Testing

Understand the fundamental differences between testing traditional software and AI-based systems

Traditional Approach

Deterministic

Testing a rule-based login system

InputExpected OutputDeterministic
Valid username + valid passwordLogin successYes
Valid username + invalid passwordError: Invalid passwordYes
Invalid usernameError: User not foundYes
Empty fieldsError: Required fieldsYes
SQL injection attemptError: Invalid inputYes

Characteristics

  • Deterministic outputs
  • Clear pass/fail criteria
  • Specification-based expected results
  • Same input always gives same output
  • Binary outcomes (success/failure)

AI Approach

Probabilistic

Testing an AI-based fraud detection login system

InputExpected OutputDeterministic
Valid credentials + normal behaviorLogin success (95% confidence)No
Valid credentials + unusual locationAdditional verification (72% risk)No
Valid credentials + new deviceMaybe blocked (60% confidence)No
Valid credentials + unusual timeVaries by risk modelNo
Previously seen attack patternLikely blocked (85% confidence)No

Characteristics

  • Probabilistic outputs
  • No single "correct" answer
  • Context-dependent behavior
  • Output may vary between runs
  • Confidence scores instead of binary

New Testing Challenges

  • How do you define "correct" when output is probabilistic?
  • What confidence threshold is acceptable?
  • How do you handle different results for the same input?
  • How do you test for bias in the risk assessment?
  • How do you regression test when outputs naturally vary?