Capps suggests an alternative AI framework, instance-based learning (IBL), which allows users to trace every decision back to the training data used. Unlike black-box AI, IBL does not generate an abstract model of the data but makes decisions directly from the data itself, allowing for greater transparency and accountability. This makes IBL a potentially valuable tool for companies, governments, and other regulated entities looking to deploy AI in a trustworthy, explainable, and auditable way, particularly in areas prone to bias allegations such as hiring, college admissions, and legal cases.
Key takeaways:
- Major AI platforms operate on black-box models, which are untrustworthy as they cannot be held accountable for their actions due to their lack of transparency.
- These black-box AI platforms are built on a technology framework called a “neural network,” which makes predictions based on abstract representations of data, not actual data.
- An alternative AI framework, instance-based learning (IBL), is gaining prominence as it allows users to trust, audit, and explain AI decisions, tracing every decision back to the training data used.
- IBL AI could be used by companies, governments, and other regulated entities to meet regulatory and compliance standards, and is particularly useful for applications where bias allegations are rampant.