Sensepoint AI
Providing Trusted Autonomy Through Neuro-Symbolic AI

An AI model typically receives inputs in the form of images, text, or sound and produces a classification or prediction along with a statistical measure of confidence. This reflects the model’s internal calculation of how closely the input matches what it encountered and labeled during training. The problem is that a model’s internal confidence does not account for contextual knowledge, whether that involves lighting conditions for a visual model, agreement across different models running on separate devices, or alignment with other sources of knowledge.

Failure to ground model outputs in contextual knowledge is the primary cause of the wildly inaccurate results that sometimes occur in both AI/ML and GenAI systems, commonly referred to as hallucinations. Retrieval-Augmented Generation (RAG) methods can provide domain-specific, ad hoc protections for large language models. However, AI-driven solutions typically span multiple domains. For organizations, it is difficult to trust AI to act autonomously without foolproof guardrails to prevent hallucinations.

Sensepoint AI was founded to solve this problem in a principled and scalable way. The company delivers a framework that uses contextual knowledge to transform internal confidence scores into transparent measures of warranted belief that can be compared and aggregated across time, devices, and even models—and then improved where needed. The highest level of warranted belief is trust.

“You cannot act on what you do not trust. We are not just labeling data; we are teaching machines to verify, reconcile, and ultimately decide when they can trust what they perceive, the same way humans do,” says Erik Thomsen, Chief Science Officer.

From Confidence to Warranted Belief

At its foundation is a neuro-symbolic framework that combines statistical outputs from machine learning with structured reasoning about context. Within this framework sits the company’s ontology fabric, an adaptive model that surrounds machine learning processes and provides the contextual knowledge and controls needed to convert model confidence into warranted belief. Outputs are no longer treated in isolation but are anchored to the circumstances of capture, the readings of other sensors, the outputs of other devices or people, and other relevant knowledge.

You cannot act on what you do not trust. We are not just labeling data; we are teaching machines to verify, reconcile, and ultimately decide when they can trust what they perceive, the same way humans do.

The framework also applies what Sensepoint calls principled multi-channel reconciliation. This process compares warranted beliefs from different sensors, analytic processes, and stored data to determine whether they align or conflict. When conflicts arise, the system can generate and execute plans to resolve them and strengthen warranted belief by gathering additional data, shifting perspective, or requesting inputs from other knowledge sources. Only when the resulting warranted belief surpasses a safe threshold is the result used for autonomous action or further decision-making. This transforms AI from a passive generator of outputs into an active system for validating and improving its own knowledge.

Real-World Impact and the Future

The company’s impact is most clearly seen in its embodied applications. For example, Sensepoint AI’s technology was deployed on a fleet of autonomous watercraft and drones tasked with surveying a harbor and identifying, with human-level precision, what kinds of vessels were arriving and departing while monitoring for vessels in distress. Traditional object detection algorithms running on boats and drones typically report detections such as “sailboat” or “trawler,” along with a confidence estimate, for example, 70 percent. The problem is that this estimate fails to incorporate contextual knowledge that could either increase or decrease statistical confidence.

”You cannot afford to be wrong when a false positive could launch a helicopter or cause a missed rescue. What we are building is not just AI that sees, but AI that understands its own uncertainty and knows what to do about it.”

To address this, the Sensepoint system operated on top of the traditional object detection algorithms on each autonomous vessel and drone. When a detection occurred, Sensepoint factored in environmental conditions such as lighting, sea motion, sensor angle, prior sightings, and existing knowledge of vessels to calculate a more realistic measure of warranted belief. It then autonomously queried nearby vessels and drones to determine whether they observed the same object and, if so, calculated a measure of aggregate belief. If that belief was not yet high enough to trust, the system either moved itself or requested that other units reposition to obtain a better view, enabling a trustworthy identification.

This shift from static confidence scores to context-sensitive models of warranted belief reduces false alarms and enables faster, safer, and more reliable coordination among autonomous systems.

“You cannot afford to be wrong when a false positive could launch a helicopter or cause a missed rescue. What we are building is not just AI that sees, but AI that understands its own uncertainty and knows what to do about it,” says Joss Stubblefield, co-founder.

Now, the company is expanding its reach. Having achieved breakthroughs in embodied AI through drones, rovers, and autonomous watercraft, Sensepoint AI is shifting its focus to the digital world.

Sensepoint AI’s future lies in providing trustworthy, collaborative AI to organizations. Its approach allows organizations to start small by applying the technology to a specific department or function, and then scale up as trust builds.

In a data-driven future, the most valuable signal will be trust. Sensepoint AI delivers that trust in a form organizations can measure, improve, and scale.