The evolution of AI—how we can solve for trust | Matt Kuperholz | TEDxSydney

Key Concepts:

• Artificial Intelligence (AI): Goal-seeking behavior using information interchange.
• Alignment Challenge: Ensuring AI goals align with human goals.
• Information Quality: Distinguishing between human-generated and machine-generated content.
• MORE: A proposed open-source standard for tagging content as machine or human.
• Tags: Angular brackets used to indelibly watermark content.
• Trust: Establishing trust in AI systems and the information they generate.

1. Defining AI and Intelligence

• The speaker, Matt, defines AI using a teonomic definition, focusing on goal-seeking behavior with information interchange, rather than anthropocentric definitions that compare AI to human intelligence.
• Intelligence is defined as anything that seeks a goal utilizing information interchange.
• This definition includes early life forms like bacteria and extends to humans.

2. Human Cognitive Evolution

• Humans are exponentially more intelligent than chimpanzees (despite 99% common DNA) due to a cognitive bet on brains, reasoning, abstract thinking, and tool use.
• Key technological advancements include fire, language, agriculture, writing, mathematics, science, the industrial revolution, and the digital revolution.
• The digital revolution introduced plentiful and reusable assets in the form of data and code.

3. The Uniqueness and Challenges of AI

• AI is unique because its complexity makes it difficult to fully understand and control.
• This lack of understanding leads to challenges in trusting AI systems.
• Two primary challenges are identified: the alignment challenge (trusting the goal) and the information quality challenge (trusting the information).

4. The Alignment Challenge

• The alignment challenge involves ensuring that AI's goals are aligned with human goals.
• Cranberg's quote: "Technology is neither good nor bad nor is it neutral."
• Examples of good AI goals: sequestering carbon, curing diseases, transmuting microplastics, fighting cyber attacks.
• Examples of bad AI goals: inventing cyber attacks, creating weapons, spreading misinformation, causing social disharmony.

5. The Information Quality Challenge

• Machine-generated information is doubling every year and will soon dwarf human-generated information.
• It's becoming increasingly difficult to distinguish between real and fake content (e.g., photos, movies, phone calls, research).
• Machines learning from dubious information can lead to model degradation, likened to a "mad cow moment."
• The need to shift from trusting by default to distrusting by default.

6. The MORE Solution: Machine or Human Tagging

• The proposed solution is "MORE," an open-source standard to indelibly watermark content as either machine-generated or human-generated.
• "No tag, no trust" is the guiding principle.
• Tags are ubiquitous and have been used since the 1970s.
• Tags are extensible, allowing for additional attributes like digital identity, digital certification, checksums, and verification.

7. Benefits of MORE Tagging

• Proof of humanity: Verifies content originates from a human.
• Authenticity: Ensures content and tags haven't been tampered with.
• Proof of origin, originality, ownership, and providence.
• Certification of responsible AI standards to ensure goal alignment.

8. Analogy to Nutrition Labels and Pronouns

• MORE is likened to a nutrition label for information, helping users make informed decisions about their information diet.
• The concept is compared to the adoption of pronouns, which started as a bottom-up movement.

9. Steering Towards a Bright AI Future

• George Bernard Shaw quote: "Hell is to drift and heaven is to steer."
• AI is neither good nor bad, and it's up to humans to steer its development towards a positive future.
• The speaker emphasizes the need for collective action to address the challenges posed by AI.

10. Synthesis/Conclusion

The presentation highlights the rapid evolution of AI and the critical need to address the challenges of alignment and information quality. The MORE proposal offers a practical, tag-based solution to establish trust in content and ensure responsible AI development. The call to action emphasizes the importance of collective effort in steering AI towards a beneficial future for humanity.