How to ensure Al is a force for good | Bill Welser | TEDxManhattanBeach

AI's Breaking Point: A Summary

Key Concepts: Artificial Intelligence (AI), Autonomy, Machine Learning, Non-biological Systems, Cognitive Capacity, Data Explosion, Control of AI, Risk Assessment, Revenue Sharing, Legislation, User License Agreements.

The AI Breaking Point

The core argument is that humanity's relationship with technology, specifically AI, is approaching a "Breaking Point" where we risk losing control. This point is visualized as the intersection of two lines on a graph: the exponential growth of AI capabilities (driven by data and computing power) and the relatively static cognitive capacity of humans.

• AI Definition: An AI system is defined by three key features:
  • Autonomy: It can operate independently.
  • Learning: It uses algorithms to learn from large datasets.
  • Non-biological: It is human-made, meaning we define its parameters and limitations.
• Data Explosion: The speaker highlights the "Big Data movement" and the massive increase in data collection and storage. This data fuels AI development.
• Cognitive Capacity: Human cognitive capacity is portrayed as slow to evolve compared to the rapid advancement of AI.
• Losing Control: The "Breaking Point" signifies a shift from humans being in control of AI to AI potentially exceeding our ability to fully understand and manage it, leading to companies and governments losing control. This doesn't necessarily mean a dystopian scenario, but rather a loss of comprehensive oversight.

Examples: Control vs. Loss of Control

• Below the Line (Control): Recommendation engines on streaming platforms are given as an example of AI that is currently under control. These systems use data to suggest content based on viewing history.
• Above the Line (Potential Loss of Control): The autonomous vehicle dilemma is presented as a complex ethical scenario. The car must make a split-second decision in a situation where any choice results in potential harm (child, senior citizens, passengers). The difficulty lies in understanding the AI's decision-making process after the fact, even with the best intentions and software engineering. The numerous variables (speed, curb depth, tree size) make it nearly impossible to fully reconstruct the AI's reasoning.

Actions to Stave Off the Breaking Point

The speaker proposes three key actions to mitigate the risks associated with AI and slow down the progression towards the "Breaking Point":

1. Demand More from Elected Officials:
   • Legislators need a deeper understanding of the technologies they are regulating.
   • The speaker cites the Future of Life Institute's call for a 6-month pause on training systems like GPT-4, noting that while many tech leaders signed the open letter, there was no actual slowdown in AI development. Sam Altman from OpenAI is specifically mentioned as someone who didn't sign it.
   • The argument is that creators cannot be solely responsible for self-regulation; knowledgeable government oversight is crucial.
2. Establish a Risk Assessment Framework for AI:
   • Analogies are drawn to risk assessment systems for natural disasters like hurricanes and earthquakes.
   • The current user license agreements are criticized for being dense, complex, and burying risk information in "small print."
   • A "grocery store analogy" is used: consumers expect clear labeling (nutrition facts, origin, price) on products. Similar transparency is needed for technology.
   • Specific information that should be disclosed includes:
     • When are experts in the loop?
     • What kind of personal information is being used?
     • How is the data anonymized?
     • What is the value gained by the company from the data?
3. Implement Revenue Sharing:
   • Tech companies are generating trillions of dollars in market capitalization from user data.
   • The speaker argues that users should receive a portion of this revenue.
   • The idea is to give data a "dollarized value" and purchasing power.

Conclusion

The speaker expresses hope that through better legislation, risk assessment, transparency, and revenue sharing, humanity can slow down or even prevent the negative consequences of unchecked AI development. However, this requires proactive action from individuals and a collective effort to hold technology companies accountable. The responsibility lies with everyone to ensure that technology serves humanity's best interests. The key takeaway is the awareness of the "Breaking Point" and the need to act before it's too late.