OpenAI and training data firm Handshake AI are reportedly requesting that their third-party contractors upload examples of real work completed in past and current roles, raising concerns about intellectual property and data privacy. According to a Wired report, this initiative appears to be part of a broader strategy among AI companies to leverage contractors for generating high-quality training data, with the ultimate goal of automating more white-collar tasks.

OpenAI's request, outlined in a company presentation, asks contractors to detail tasks performed in previous jobs and provide concrete examples of their work, including documents, presentations, spreadsheets, images, and code repositories. The company instructs contractors to remove proprietary and personally identifiable information (PII) before uploading these files, offering access to a "ChatGPT Superstar Scrubbing tool" to assist in this process.

The move highlights the critical role of data in training large language models (LLMs). These models, like OpenAI's GPT series, learn to generate human-quality text by analyzing vast datasets. The quality and relevance of this training data directly impact the model's performance and capabilities. By using real-world examples of professional work, AI companies aim to improve the accuracy and effectiveness of their models in automating complex tasks.

However, the practice raises significant legal and ethical questions. Intellectual property lawyer Evan Brown told Wired that this approach poses a considerable risk for AI labs, as it relies heavily on contractors to accurately determine what constitutes confidential information. "Any AI lab taking this approach is putting itself at great risk with an approach that requires a lot of trust in its contractors to decide what is and isn't confidential," Brown stated.

An OpenAI spokesperson declined to comment on the specific initiative.

The use of contractor-provided data reflects a growing trend in the AI industry. As AI models become more sophisticated, the demand for high-quality, real-world training data increases. Companies are exploring various methods to obtain this data, including synthetic data generation, web scraping, and partnerships with data providers. The reliance on contractors, however, introduces unique challenges related to data security, privacy, and intellectual property rights.

The long-term implications of this data collection strategy are still unfolding. If successful, it could accelerate the automation of white-collar jobs, potentially impacting employment across various industries. Furthermore, the use of real-world data raises concerns about bias and fairness in AI systems. If the training data reflects existing societal biases, the resulting AI models may perpetuate and amplify these biases.

The current status of OpenAI's data collection initiative remains unclear. It is unknown how many contractors have participated or the volume of data that has been collected. As AI companies continue to pursue this strategy, it is likely that regulatory scrutiny and public debate will intensify, focusing on the need for clear guidelines and safeguards to protect intellectual property and individual privacy.