How partial truths are a threat to democracy | De Kai | TEDxKlagenfurt

Summary of YouTube Video: "The Most Insidious Form of AI Misinformation"

Key Concepts: Information disorder, misinformation, disinformation, malinformation, neginformation (context omission), cognitive biases, fundamental attribution error, confirmation bias, reactive devaluation bias, omission bias, AI manipulation, unconscious decision-making.

1. Introduction: The Problem of Omitted Context

• The video begins by illustrating how omitting context can drastically alter the perception of an event. A video of people being beaten is shown, initially eliciting sympathy for the victims. However, when the preceding context of the "victims" firing guns at an unarmed crowd is revealed, the perception shifts.
• The speaker argues that the focus on "fakes" and "bad actors" in discussions about misinformation overshadows the more insidious problem of omitted context.
• The speaker, an AI professor who pioneered machine translation technology, expresses concern about AI being used to increase division and hate instead of fostering understanding. He describes this realization as his "Oppenheimer moment."

2. Defining Information Disorder and Its Limitations

• The term "information disorder" is used to encompass the various ways AI algorithms contribute to societal dysfunction.
• The standard categorization of information disorder into misinformation (false information shared without intent to harm), disinformation (false information shared with intent to harm), and malinformation (truth or partial truth shared to cause harm) is critiqued. This framework, popularized by Claire Wardle and Hussein Derakhshan, is considered incomplete.
• The speaker argues that the "rule of threes," a cognitive bias that makes information presented in groups of three more readily accepted, contributes to the uncritical acceptance of this three-part categorization.

3. Introducing Neginformation: The Missing Piece

• The speaker proposes a 2x2 grid to analyze information disorder, with axes representing the intent of the propagator (malicious vs. negligent) and the nature of the information (outright falsehood vs. partial truth with omitted context).
• This grid reveals a "glaring hole" in the analysis: neginformation, defined as partial truths with selectively omitted crucial context negligently propagated by ordinary people.
• Neginformation is argued to be the most dangerous form of information disorder because it is more believable than outright falsehoods and is spread more widely by well-meaning individuals.
• Example: Uncle Mike resharing misleading information without checking the context.

4. The Power of Neginformation and Cognitive Biases

• Neginformation is more misleading than falsehoods because it is harder to spot and allows unconscious biases to fill in the missing context.
• AI algorithms exploit human cognitive biases to manipulate perceptions through neginformation.
• Examples of Cognitive Biases Exploited by AI:
  • Fundamental Attribution Error: Ascribing negative behavior of other groups to inherent flaws.
  • Confirmation Bias: Interpreting information to confirm existing beliefs.
  • Selective Perception Bias: AI predicts how liking or disliking a group influences perception of context omissions.
  • Semise Reflex Bias: AI chooses partial truths that omit new evidence to fit existing likes and dislikes.
  • Subjective Validation: AI chooses partial truths that align with one's identity.
  • Conservatism Bias: Resistance to revising beliefs even with new evidence.
  • Backfire Effect Bias: Reacting to contradictory evidence by strengthening existing beliefs.
  • Compassion Fade Bias: Showing more compassion for identifiable victims than anonymous ones.
  • Egocentric Biases: Overestimating the accuracy of one's own perspective.
  • Illusion of Validity: Overestimating the accuracy of one's own judgments.
  • Overconfidence Effect Bias: Maintaining excessive confidence in one's own answers, even when wrong.
• The speaker emphasizes that AI algorithms, operating unconsciously, can manipulate human unconscious biases without triggering fact-checking or bad actor detection.

5. Addressing the Misguided Focus on Falsehoods and Bad Actors

• The video critiques the current emphasis on detecting bad actors and false information as a misdirection caused by cognitive biases within the field of information disorder research itself.
• Reactive Devaluation Bias: Devaluing ideas simply because they come from a perceived adversary. The speaker argues that the validity of an idea should not depend on its source.
• Omission Bias: Judging harmful actions as ethically worse than omissions that lead to equal or greater harm. The speaker argues that misleading neginformation can be more harmful than outright falsehoods.

6. Conclusion: A Call for Awareness and Action

• Humanity can no longer afford to focus solely on falsehoods and bad actors while ignoring the more significant problem of negligently propagated partial truths (neginformation).
• Continuing to ignore neginformation will lead to AI driving democracies into disaster.
• The speaker urges a shift in focus towards understanding and addressing the insidious effects of context omission in AI-amplified information disorder.
• Humanity will destroy humanity before AI gets a chance to if we keep getting distracted by only falsehoods and bad actors.