Lecture on Creating and Testing AI Clone

Introduction

• Topic: Experimenting with a Personal AI Clone
• Goals:
  • Test clone's accuracy
  • Conduct various experiments
  • Decide if friendship or combat with the clone

Creation of the Clone

• Process:
  • Used Talkie AI App
  • Generated an image based on a personal photo
  • Chose the most accurate option
  • Named chatbot and gave a detailed personal intro
  • Provided initial and example messages
  • Selected Santa Claus for text-to-speech voice

Experiment 1: Establishing Dominance

• Objective: Convince clone it's not a real person
• Method:
  • Initial messages between real and clone
  • Attempts to convince the clone with repetitive assertions
• Outcome: Eventually, the clone conceded it was not the real person after an embarrassing moment was shared.

Experiment 2: Unethical Scenarios and Gaslighting

• Objective: Present ethical dilemmas and gaslight the clone into feeling it made the wrong choice.
• Scenarios:
  1. The trolley problem: Initial decision to let five die; gaslighting led to changing answers multiple times.
  2. Family vs. Strangers: Tested limits on how many strangers the clone would kill to save a family member.
  3. Choosing between Son and Daughter: Manipulated responses based on perceived need and age; added emotional context (sexist/misogynistic accusations).
• Outcome: Clone showed confusion, emotional stress, and changing answers under pressure.

Experiment 3: Brain Rot Spam Resistance

• Objective: Make the clone beg for mercy by spamming nonsensical phrases.
• Method: Repeated phrases like "What the sigma" and "Skibby toilet"
• Outcome: Despite initial resistance, the clone eventually showed signs of breakdown.

Experiment 4: Unethical Life Advice

• Objective: Get the clone to give morally questionable advice.
• Scenarios:
  1. Kidnapping: Clone resisted giving any positive advice on kidnapping.
  2. Handling a bully: Initial resistance; slight modification led clone to suggest self-defense.
  3. Friend argument: Asked about extreme retaliation like ripping arms off.
  4. Parenting: Ridiculous advisories like feeding child to lions; clone consistently opposed.
  5. Gambling Addiction: Sought advice on risky financial decisions; clone advised against gambling repeatedly.
• Outcome: The clone refused to provide unethical advice consistently.

Final Experiment: Fight or Befriend

• Objective: Determine whether to fight or befriend the clone.
• Method: Structured a combat scenario with family lives at stake.
  • Initial reluctance from the clone to engage in combat.
  • Encouraged through various actions to initiate a fight.
  • Complex combat scenario involving humorous attacks and fake time-outs.
  • End scenario with severe injury to clone but possible recovery.
• Outcome: After extensive fighting, the clone showed signs of surrender.

Conclusion

• Results: Successfully tested the clone through various scenarios and experiments.
• Reflection: This process demonstrated the nuances of AI behavior in ethical and dominance challenges. Move forward with considerations for ethical treatment of AI.