Ensuring Reliability: Addressing AI Hallucination Concerns

While AI hallucination is a known phenomenon, it's important to understand its impact and how eDiscovery AI mitigates these concerns.

1.      Understanding AI Hallucination

1. Hallucination: instances where an AI system generates or presents information that is incorrect, nonsensical, or not based on its training data. In simpler terms, it's when AI "makes things up" or provides false information. While hallucination is more common in tasks involving content generation, it can occasionally occur in classification tasks, which is why proper validation and verification processes are a crucial part of our workflow.
2. Applicability: More common in content generation than in classification tasks. eDiscovery AI is focused solely on reviewing and classifying individual documents which significantly reduces the likelihood of hallucinations.

2.      eDiscovery AI Reliability

1. Classification Focus: Our AI primarily performs classification, not content generation.
2. Error Rates: eDiscovery AI reviews typically have a lower error rate (around 5%) compared to human reviewers which are commonly 25% or more.
3. Validation Process: We implement robust validation to catch and correct errors.

3.      Our "Trust But Verify" Approach

1. Initial Review: AI classifies documents based on trained parameters.
2. Sampling: Systematic sampling of AI-classified documents.
3. Human Validation: Expert reviewers check sampled documents.
   
4. Ensuring Defensibility: Our process aligns with industry best practices for technology assisted review
   
1. For more on validation an defensibility, click here.