AI Origins · Lesson 11

Robot Repair Clinic

Five broken robots need a diagnosis. Use IF-THEN rules to find each fault — the fewer clues you need, the better your reasoning.

Round 1 / 5 Score: 0 4 inspects left

🤖 Loading…

Inspect an area (each costs 1 token)

Logic Probe Hover or click a robot subsystem to scan for clues.
Chest: power values · Head: diagnostics · Wheels: motion clues

Clues found so far

    AI Rule Engine — live analysis

    No diagnosis yet Gather evidence to raise confidence.

    Inspect the robot to gather evidence. The AI will update here.

    The engine fires IF-THEN rules as each clue arrives.

    Your diagnosis

    What This Trains

    System Mechanics

    • Symbolic LogicKnowledge encoded as explicit IF-THEN rules
    • State MachinesFacts update in deterministic forward passes
    • Weighted InferenceConfidence factors model uncertainty beyond true/false
    • ExplainabilityEvery troubleshooting guess can be traced to fired rules

    Conceptual Questions

    • TrustWho authored the rules and where can bias hide?
    • ConflictWhat should happen when rules disagree?
    • LimitsWhen does explicit logic stop scaling well?
    • BridgeHow does this compare to neural-network learning?

    Historical Context

    This lab represents the Symbolic AI / Expert Systems era (1970s–1980s).

    Expert Systems Era

    • ApproachHuman knowledge formalized as transparent rules
    • MYCIN-styleCertainty-factor methods showed how rule confidence can combine
    • StrengthAuditable reasoning path for each recommendation
    • WeaknessRule authoring burden and brittle domain transfer

    AI Timeline Placement

    1. Perceptron era begins neural-symbolic split
    2. ELIZA popularizes scripted language interaction
    3. Expert systems (including MYCIN-style ideas) scale symbolic AI
    4. Deep learning re-centers data-driven representation learning

    Worked Example: Forward Chaining & Evidence Combination

    Trace how an inference engine fires production rules from observed sensor facts and calculates cumulative diagnostic certainty.

    Sample Problem: Diagnosing a Mobile Robot Arm Fault

    Suppose our mobile robot fails to grasp a sample. Diagnostic telemetry loads the following initial facts into working memory: battery_voltage = normal, gripper_current = 0.0A, and bus_timeout = true. The knowledge base contains two diagnostic rules pointing to a severed harness:

    • Rule 101: IF battery_voltage IS normal AND gripper_current IS 0.0A THEN actuator_bus_severed [CF = 0.70]
    • Rule 204: IF bus_timeout IS true THEN actuator_bus_severed [CF = 0.60]
    1. Step 1: Match Premises against Working Memory.
      The inference engine scans all rules. Rule 101 checks if battery_voltage is normal and gripper_current is 0.0A. Because both conditions exist in working memory, Rule 101 fires.
    2. Step 2: Assert Intermediate Conclusion with Certainty Factor \(CF_1\).
      Rule 101 adds the hypothesis actuator_bus_severed with certainty factor \(CF_1 = 0.70\).
    3. Step 3: Evaluate Rule 204 (Forward Chaining Pass 2).
      Rule 204 examines working memory for bus_timeout = true. This condition is also satisfied, which means Rule 204 fires independently and concludes actuator_bus_severed with certainty factor \(CF_2 = 0.60\).
    4. Step 4: Combine Evidence Using the MYCIN Formula.
      When two independent rules support the exact same conclusion, the engine combines their certainty factors according to Shortliffe's classic formula:
      \(CF_{\text{combined}} = CF_1 + CF_2(1 - CF_1) = 0.70 + 0.60(1 - 0.70) = 0.70 + 0.18 = 0.88\).
      Mechanism: Because multiple independent diagnostic symptoms corroborate the same fault, the system's certainty increases from 70% to 88%. This explainable chain of evidence allows engineers to verify exactly why the diagnosis was made.

    Self-Check Question

    If a third rule later confirms actuator_bus_severed with \(CF_3 = 0.50\), what is the new combined certainty factor?

    Show Answer & Derivation

    Answer: \(CF_{\text{new}} = CF_{\text{prev}} + CF_3(1 - CF_{\text{prev}}) = 0.88 + 0.50(1 - 0.88) = 0.88 + 0.06 = 0.94\) (94% certainty). Each corroborating piece of evidence reduces the remaining uncertainty by half, bringing the total confidence asymptotically toward 1.0.

    Sources & Further Reading

    Primary literature on knowledge engineering, rule-based inference, and certainty factor mathematics.

    • 1976 MYCIN Landmark Shortliffe, Edward H. Computer-Based Medical Consultations: MYCIN. Elsevier/North-Holland, New York (1976). Established the mathematical foundation for managing uncertainty in rule-based diagnostic systems.
    • 1984 Knowledge Engineering Buchanan, Bruce G., & Shortliffe, Edward H. Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project. Addison-Wesley (1984). Comprehensive retrospective on production rules, conflict resolution, and explanation generation.
    • 1977 Foundational Survey Feigenbaum, Edward A. “The art of artificial intelligence: Themes and case studies of knowledge engineering.” Proceedings of the 5th International Joint Conference on Artificial Intelligence (IJCAI-77), 1014–1029 (1977). Introduced the principle that real-world AI power derives from domain-specific knowledge rather than generic inference mechanisms.

    Lesson Outputs

    Each team should publish a short reflection and demo link at the end of class.