Curating Data is Key to Safe Artificial Super Intelligence AI Systems

Monday - December 11 2023, 09:52 UTC - 11 months ago

OpenAI has revealed that, in order to create safe ASI (Artificial Super Intelligence) systems, the critical step is curating the data to include ethical and pro-humanity examples. This is laborious yet important in order to create AI that is beneficial to the world, with greater safety in mind.

An OpenAI employee has observed that Large Language Models starting with the same dataset converge to the same point. This would mean curating the data is the critical step in creating safe ASI (Artificial Super Intelligence). If we can front load the AI with the desired ethical and pro-humanity examples, then the resulting AI system should have a core of human aligned safety.The OpenAI research employee presented this finding at the 31st Conference on Neural Information Processing Systems (NeurIPS) .

This observation has been made by analysing a range of models, measuring their accuracy according to the length of the dataset and the size of the model. Their findings showed that longer datasets for the language models generally produced higher performance results.At the core of this process is data curation, which is a laborious and expensive process. Curation is the process by which a dataset is built, and this is the critical step in the production of an AI system .

Curation is currently done manually, often by outside firms who specialize in building datasets. It is a long process and requires the curators to ensure that the data that is collected is accurate and complete, and that it is aligned to the desired outcomes of the AI.AI researchers now have an opportunity to create datasets with a focus on curating for data that enforces ethical and “pro-humanity” norms .

This will allow for the chance to create AI that is more valuable and beneficial to the world, rather than AI that is only profitable. Instead of creating AIs that are designed to help humanity but ultimately run counter to its goals, AIs can be designed with greater safety in mind.In order to truly achieve safe Artificial Super Intelligence (ASI), we must begin with front loading the learning process by curating data that draws upon ethical and inclusive examples .

The researchers at OpenAI have shown us that curating the data is the critical step in creating an AI system that is aligned with human values.