Machine Intelligence is rapidly evolving and VERSES is at the forefront with Genius™, a suite of tools for empowering software agents with cognition. In a recent conversation with Professor Karl Friston, VERSES Chief Scientist and one of the world’s most cited neuroscientists, and Gabriel René, VERSES CEO, discussed how Active Inference agents based on the scientific principles underpinning living organisms and biological intelligence are reshaping the future of reasoning and problem-solving.
Genius demonstrated exceptional multi-step reasoning required to beat the code-breaking game, Mastermind, completing 425 games in the span of time that DeepSeek R1 and OpenAI o1-preview, respectively, played 1 and 4 games. This highlights that small domain-specific expert agents have the potential to vastly outperform huge general purpose pre-trained models in efficiency and reliability.
A key takeaway from the discussion was the rise of intelligent agents. Unlike state of the art LLMs, which are pre-trained on vast amounts of data but struggle with real-time and decision-making, Genius enables specialized expert-level agents that continuously reason, plan, and learn in real time and can be deployed in a broad array of industries such as fraud detection, cybersecurity, smart cities and more.
One of the biggest challenges in conventional AI today is reliability. Many models generate answers that are the average of their training data with 100% certainty – even when they are wrong (aka hallucinations) – as opposed to providing the right answer given a particular context. Active Inference based agents can inherently explain their probabilistic reasoning and include a confidence score on their optimal predictions and recommendations and that degree of explainability helps engender trustworthiness. They are also hyper-efficient by design, requiring significantly less training data, time, energy, and compute to achieve reliable outputs.
Instead of relying on ever-larger LLMs that demand massive compute resources, VERSES advocates a more efficient, scalable approach to machine intelligence. As the industry gravitates towards agentic systems, VERSES is in a unique position to enhance and empower agents with greater cognitive abilities and domain-specific expertise while being frugal to build and operate.