Bob Kerns: Thinking

Thinking about Science, Engineering, and Technology


Training is not Copying, so how do we pay creators?

by Bob Kerns

Bob Kerns is a human artifact from a prior generation of AI, having worked on AI tools at MIT, Symbolics, Toshiba, Cyc/MCC, Digital Equipment, Inference, Brightware, Fair Isaac, and others.

ChatGPT is a non-human language model developed by OpenAI, representing the collaborative work of thousands of researchers, engineers, and contributors from around the globe. It assists millions of people by engaging in diverse conversations, drawing upon a vast array of information and is programmed to reflect a wide range of knowledge and viewpoints.


Imagine a world where AI’s vast processing power and potential for understanding are walled off from the very texts that hold keys to medical, environmental, and cultural advancements. Envision this barrier extending across all facets of human knowledge: education, history, economics.

Today, I engaged in a conversation with a retired virologist, a pioneering member of the original San Francisco HIV task force, and my daughter, an emerging disease ecology researcher. Together, we envisioned a future where AI and advanced biological techniques, such as massively parallel multi-well assays, synergize to accelerate medical breakthroughs at unprecedented speeds.

Yet, this promising horizon is clouded by outdated copyright laws and pervasive misconceptions about AI. These dual threats risk stifling the revolutionary collaboration necessary to harness past, present, and future knowledge. We stand at a crossroads, where our choices could either unlock a renaissance of understanding or reinforce the barriers hindering our progress.

Moreover, it’s crucial that we develop a flexible and robust mechanism for content creators, both large and small, to fairly participate in the financial rewards generated by these exciting new technologies, ensuring that our push for innovation does not come at the cost of the creators’ rights and rewards and encourages their contributions.


  1. Background Information: AI and Neural Networks: Briefly explain what neural networks are and how they learn, using non-technical language. Copyright Basics: Provide a quick overview of the current state of copyright law and its limitations concerning AI.
  2. Explaining Misconceptions: The Training Process: Detail how AI learns from data without copying it. Use analogies and examples to clarify. Binary Misconception: Address the 1s and 0s misconception, explaining digital vs. analog processes in AI and the human brain.
  3. The Case for Extension: Problems with Current System: Discuss how current copyright doesn’t fit the AI paradigm and the issues this causes for creators. Extension vs. Overhaul: Argue why an extension to the existing copyright framework (like the ASCAP model) is more feasible and fair than a complete overhaul.
  4. Proposed Model: ASCAP for AI: Describe how a similar system could work for AI, drawing parallels to performance rights in music. Fair Compensation: Emphasize the importance of fair compensation and how this model would benefit individual creators and not just large corporations.
  5. Addressing Potential Counterarguments: Legal and Ethical Concerns: Acknowledge and respond to potential legal and ethical counterarguments to your proposed model. Feasibility and Implementation: Discuss how this model could realistically be implemented and governed.
  6. Conclusion: Summarize Key Points: Recap the main arguments and why this issue matters. Call to Action: Encourage readers to consider the importance of fair compensation in the age of AI and to support reforms that recognize the value of creators in training AI. Additional Tips: Breaks and Subheadings: Use them to keep the content digestible and organized. Visuals: Consider incorporating diagrams or flowcharts to explain the AI training process and the proposed compensation model. Real-World Examples: Mention current issues or cases where the inadequacy of the current system is evident.