James Duez is the CEO and co-founder of Rainbird.AI, a decision intelligence business focused on the automation of complex human decision-making. James has over 30 years’ experience building and investing in technology companies. He’s worked with Global 250 organisations and state departments, and is one of Grant Thornton’s ‘Faces of a Vibrant Economy’. James is also a member of the NextMed faculty and the Forbes Technology Council.
In this post, James discusses the future of generative AI. While GenAI is a powerful prediction tool, it still commits errors because it lacks reasoning and causal explanations. James explains how integrating GenAI with retrieval-augmented reasoning (RAR) architecture is the key to creating a trustworthy system with decision intelligence capabilities:
Every organisation in the world is focused on leveraging artificial intelligence (AI) and data, and this is only accelerating since the advancement of generative AI and techniques like retrievalaugmented generation (RAG).
Generative AI is at the top of the hype cycle and expectations remain high. If used carefully there is much potential to aid the efficiency of experts, but generative AI alone cannot evaluate problems logically and in context, nor produce answers that come with a causal chain of reasoning that delivers certainty that every answer is 100% explainable and free from the risk of error.
The quest for systems that not only provide answers but also explain their reasoning in a transparent and trustworthy manner is becoming paramount, at least where AI is being used to make critical decisions.
When Air Canada’s chatbot gave incorrect information to a traveller, the airline argued its chatbot was ‘responsible for its own actions’ but quickly lost its case, making it clear that organisations cannot hide behind their AIpowered chatbots when they make mistakes.
What went wrong is clear. Generative AI is a machine learning technology that creates compelling predictions, but is not capable of human-like reasoning and cannot provide causal explanations.
While many are stunned by the bright lights of generative AI, it’s best leveraged as one piece of a bigger architecture. As highlighted in the previous cover issue, it is a Neurosymbolic approach to decision intelligence that holds the greatest value – to deliver solutions that start with a clear focus on outcomes and work back from that to a hybrid or composite use of AI. Decision intelligence leverages AI models in a configuration that is grounded in trust and transparency, avoiding the perils of noise, bias and hallucination.
This thinking is now validated by numerous analysts, including Gartner, who convey decision intelligence as being of equivalent importance to generative AI in terms of both its transformational potential and the timescale to mainstream adoption.
What’s more, the issue of accuracy and explainability to drive trust is becoming recognised as a critical component, with knowledge graphs accepted as the primary grounding technology for generative AI and therefore, a key enabler to its adoption.
Organisations have the ambition but also the responsibility to uncover and leverage ways of using AI responsibility in a world that is only becoming more regulated.
As generative AI continues to rapidly evolve, more and more focus is falling on the importance of trust.
The quest for systems that not only provide answers but also explain their reasoning in a transparent and trustworthy manner is becoming paramount, at least where AI is being used to make critical decisions.
AI leaders have continued to grapple with the challenges of leveraging generative AI, specifically large language models (LLMs) which risk outputting incorrect results (known as hallucinations) and lack formal explainability in their answers.
One architecture that has gained momentum is that of retrieval-augmented generation (RAG). It uses well-understood mathematical techniques to identify similar snippets from a reference source of documents, and then injects those snippets into an LLM to help inform outputs.
Although RAG still risks hallucinations and lacks explainability, it has the advantage of being able to make content predictions over targeted document sources and can point at the parts of the document that it used when generating its predicted outputs.
However, RAG is inherently limited by its exploratory nature, which focuses on accumulating knowledge or facts that are then summarised, without a deep understanding of the context or any ability to provide logical reasoning.
LLMs alone are poor at reasoning as was pointed out recently by Yann LeCun, one of the godfathers of AI and Head of AI at Meta. Regardless of whether you ask a simple question, a complex question or even an impossible question – the amount of computing power expended to create each block of generated content (or token) is the same.
This is not the way real-world reasoning works. When humans are presented with complex problems we apply more effort to reasoning over complexity. With LLMs alone, the information generated may look convincing but could be completely false and demand that the user then spends lots of time checking the veracity of the output.
Architectures are evolving to try and improve the performance of LLMs, including retrieval-augmented thought (RAT) to pull in external data, semantic rails to try and keep LLMs on topic and prompt chaining to turn LLM outputs into new LLM inputs. All of these are designed to diminish risks but cannot eliminate them.
But a new architecture is taking hold, that of retrieval-augmented reasoning (RAR), an innovative approach that transcends the limitations of RAG by integrating a more sophisticated method of interaction with information sources.
Unlike RAG, RAR doesn’t just seek to inform a decision by generating text; it actively and logically reasons like a human would, engaging in a dialogue with sources and users to gather context, then employing logical reasoning to produce answers accompanied by a logical rationale. It requires a symbolic reasoning engine and uses a very high-level knowledge graph to work.
The distinction between RAG and RAR is not merely technical but fundamentally changes how AI systems can be applied to solve real-world problems. It’s comprehensive, accurate and represents the ultimate in guardrailing, so it cannot hallucinate.
We all now know that LLMs produce answers based on their training from the public internet. Risks of hallucination are high and the lack of explainability is a serious problem.
RAG’s approach, while useful for exploratory queries, falls short when faced with the need to understand the nuances of specific situations or to provide answers that are not only accurate but also logically sound and fully explainable. While it can point to sources for its predictions, it can still hallucinate and cannot explain its reasoning.
RAR however addresses all these challenges head-on by enabling a more interactive and iterative process of knowledge retrieval, consultation, and causal reasoning.
For example, RAR can enable lawyers to interact with legislation and complex case law in the context of their case, and obtain a causal chain of reasoning for worked answers. RAR can power tax solutions, reasoning over large amounts of regulation to find appropriate tax treatments at a transaction level.
These use cases all represent the ability to reason in a way that is contextually relevant, free from hallucination, and backed by a clear chain of reasoning. It’s sector agnostic and enables the rapid creation of digital assistants in any domain.
But there is yet another benefit of the RAR architecture.
Because it uses a symbolic reasoning engine and a knowledge graph to navigate document sources, the graph itself can be extended to incorporate human expertise. This enables models to leverage both documented regulation, policy or operating procedure and human tribal knowledge, enhancing its contextual decision-making capabilities.
RAR is particularly valuable in regulated markets, where evidence-based rationales are crucial to trust and therefore to adoption. By providing answers that come with source references and logical rationales, RAR fosters a level of trust and transparency that is essential in today’s data-driven world.
While RAG has served as a valuable tool in the AI toolkit, the advent of RAR represents a significant leap forward in our ability to harness the power of AI for complex decision-making.
By offering nuanced answers that are grounded in logical reasoning and contextual understanding, RAR opens up new possibilities for AI applications that require a high degree of trust and explainability.
As we continue to navigate the challenges and opportunities presented by AI, approaches like RAR will be instrumental in ensuring that our technology not only answers our questions but does so in a way that we can understand and trust.
