An End to Hallucinations: NetMind.AI and King’s College London Partner to Tackle a Key Issue in Natural Language Generation

2024-09-24|NetMind
blob

Natural Language Generation (NLG) has transformed our daily lives, from the helpful voice assistants that order groceries to chatbots that write everything from poems to complaint letters. NLG is everywhere, simplifying tasks and improving productivity. However, one critical challenge remains: filtering out false or irrelevant information – known as “AI hallucinations.”

At NetMind.AI, we’re excited to announce a multi-year partnership with King’s College London, led by the esteemed Dr. Zheng Yuan. Together, we aim to take a giant leap forward in improving the accuracy and reliability of AI-generated content. Our mission? To ensure NLG outputs stay grounded in reality while giving users greater control over the content they create.

The Challenge: AI Hallucinations

So, what exactly is the issue we’re addressing? While AI systems can generate impressive, natural-sounding text, they often fall short by producing inaccurate or misleading information – AI hallucination. These hallucinations can be difficult to detect because the AI can present the information confidently, making fact-checking challenging. The consequences can be serious, especially in fields like journalism, healthcare, and academia, where factual accuracy is critical.

Dr. Zheng Yuan explains: “The public has access to amazing tools with generative AI, [using them for] everything from checking grammar to polishing academic writing and business emails. This is the power of natural language processing – however, we notice they tend to ‘hallucinate.’"

One of the causes of these hallucinations is that AI systems rely on vast data but are not always connected to verified up-to-date sources. As a result, users can be misled, even when the AI output appears reliable. Dr Yuan’s team will address this issue by integrating advanced fact-checking protocols and hallucination mitigation strategies, making NLG systems more accurate and trustworthy.

A New Era in AI Development

Our partnership with King’s College London enables researchers to directly tackle this issue. As part of a multi-year gift agreement, Dr Yuan’s work will focus on developing algorithms that detect and mitigate AI hallucinations in real time.

The goal is to create next-generation evaluation systems that not only assess the accuracy of AI-generated text but also identify “factual hallucinations” – true but irrelevant information. The system will suggest effective mitigation strategies, providing users with more control over the content they receive and ensuring greater reliability across industries.

At NetMind.AI, we’re supporting this initiative with our cutting-edge NetMind Power platform, offering the computational power needed to train sophisticated models capable of detecting hallucinations at scale.

The Road to Better AI

Addressing AI hallucinations is crucial to ensuring that NLG models remain both useful and trustworthy. Dr. Yuan notes that existing evaluation systems were developed before the rise of advanced NLG models, leaving gaps in how we assess AI output. Her team is closing these gaps by identifying which models are better at preserving factual accuracy and integrating this knowledge into their hallucination detection system.

The project’s first phase will focus on detecting hallucinations and classifying them as factual or non-factual. (Factual hallucinations occur when information is actually true but irrelevant or unexpected) “Once the type of hallucination is identified, we can propose different mitigation strategies depending on the level of hallucination,” Dr. Yuan explains. This system will allow users to either remove or correct non-factual hallucinations and choose to keep or filter out factual ones.

Once operational, the system will be scaled for broader applications. For example, in translation, these algorithms could ensure that the true meaning of the original text is preserved. Similarly, summarisation tools could align more closely with source information, providing more accurate summaries.

The Broader Impact

By addressing one of the most pressing challenges in generative AI – factual accuracy – this partnership aims to set new global standards for AI development, impacting industries from journalism to medical reporting.

We’re proud to be part of this journey, providing both financial support and the computational backbone needed to tackle such a complex issue. As AI becomes more integrated into everyday life, collaborations like this will help ensure this groundbreaking technology serves humanity ethically and meaningfully.

Stay tuned for updates on our progressed!

Read more:

https://www.kcl.ac.uk/news/kings-researcher-partners-with-ai-start-up-to-build-natural-language-generation-models