Cardano founder Charles Hoskinson has expressed concerns about the risks artificial intelligence poses for intellectual property, warning that researchers may inadvertently share unpublished work when using cloud-based AI platforms.
Debate over mathematical discovery
Hoskinson addressed these concerns during a recent YouTube broadcast, providing commentary on mathematician Jay Cummings’ claims regarding OpenAI’s ChatGPT. Cummings, a professor at the University of California, San Diego, had previously posted that his own research might have been used by ChatGPT to solve a complex mathematical problem which OpenAI later described as a breakthrough.
According to Hoskinson, Cummings maintained chat logs that could support his suspicion that ChatGPT’s solution was not entirely original. Hoskinson cast doubt on the assertion by OpenAI that its AI model arrived at the solution independently, suggesting that evidence warranted further scrutiny.
OpenAI says that they totally independently came up with this. However, the available evidence could warrant closer examination, especially when chat logs suggest prior user input on similar problems.
AI and intellectual property risks
Hoskinson elaborated that AI systems, when confronted with exceptionally challenging problems, tend to search through prior interactions and public data instead of inventing solutions from scratch. He described modern AI as optimization systems designed to identify the most efficient strategies, which may sometimes incorporate user-provided information encountered in previous chat sessions.
Extending his observations beyond mathematics, Hoskinson cautioned academics, researchers, and entrepreneurs against sharing unpublished work or sensitive ideas with cloud-hosted AI tools. He emphasized the danger that proprietary concepts entered into these systems may no longer be under the originator’s control.
If you’re an academic or entrepreneur, sharing your ideas with frontier AI models in the cloud may mean those ideas are no longer solely yours.
He argued that solving longstanding mathematical problems is a significant professional milestone and underscored the need for stronger safeguarding of underlying research from unintended exposure.
Midnight’s private AI environments
To mitigate these risks, Hoskinson cited Midnight, a protocol currently under development. He described Midnight as creating private environments for AI operations, ensuring that user activity and logs remain confidential and inaccessible to external AI platform operators.
Midnight features a distinctive economic model designed to decouple network security and governance from the financial resources required to use the protocol. This separation seeks to address volatility issues that can arise when a single cryptocurrency is employed for both network validation and user transactions.
Through this design, users may interact with decentralized applications without having to manage a fluctuating native token for every transaction. This could potentially enhance stability and user experience within the network.
Hoskinson has previously referenced a broader industry movement towards sustainable, protocol-level revenue models and argued that, in the long term, network utility should play a central role in maintaining robust ecosystem development.
Mini dictionary: Midnight, a privacy-focused protocol being built within the Cardano ecosystem, aims to enable confidential transactions and data sharing using zero-knowledge proofs, while separating operational and governance costs through its unique economic model.
