UK Start-up Develops ‘Uncertainty-Aware AI’ to Combat Hallucinations

TL;DR: DigiLab has developed the Uncertainty Engine, an AI assistant that provides confidence ratings instead of hallucinations. The technology is being tested for maritime route planning and is already supporting nuclear reactor design, with potential applications in healthcare and shipping efficiency.

UK start-up digiLab has developed a new generation of artificial intelligence that addresses one of AI’s most significant safety concerns—hallucinations. The “uncertainty-aware AI” system, called the Uncertainty Engine, provides confidence ratings alongside its answers rather than generating fictitious responses when faced with incomplete data.

Context and Background

The technology is currently being tested aboard a high-performance sailing yacht with the Next Step Racing team, where it assists with complex route planning by processing wind conditions, sea states, boat performance data, and competitor positions. Co-skipper Joss Creswell noted that AI’s ability to quantify and reduce vast amounts of information provides crucial competitive advantages.

DigiLab is already collaborating with a major manufacturer on small-scale nuclear reactor design, demonstrating the technology’s applicability to safety-critical situations. Founder Tim Dodwell emphasises that as AI adoption expands into transformative sectors like energy, maritime safety, and healthcare, uncertainty quantification becomes essential for establishing trust where human lives are at stake.

Looking Forward

The maritime industry presents significant opportunities for the technology. DigiLab projects that capturing uncertainty in weather conditions and shipping lane availability could achieve 1–5% fuel savings for oil tankers and cargo vessels. For UK shipping, this translates to 600,000 tonnes of reduced carbon emissions annually—equivalent to removing 300,000 cars from roads.

The development addresses a fundamental challenge in AI deployment: establishing trust in algorithms that influence safety-critical decisions. By explicitly acknowledging when confidence is low, the Uncertainty Engine enables human operators to assess answer reliability and seek additional information where necessary, potentially accelerating AI adoption across regulated industries.

Source Attribution:

Share this article