AI startup Counterintuitive has set out to build “inference-native computing” that allows machines to understand rather than just imitate. Such breakthroughs have the potential to move AI from pattern recognition to true understanding, paving the way for systems that can think and make decisions, or in other words, systems that are more “human-like.”
Counterintuitive chairman Gerald Rego spoke about what the company calls the “twin trap” problem facing AI, saying the company’s first goal is to solve two key problems that limit current AI systems and prevent even the largest AI systems from becoming stable, efficient, and truly intelligent.
The first trap highlights that today’s AI systems are built on outdated mathematical foundations and lack a reliable and repeatable numerical foundation. An example is floating-point arithmetic, which was designed decades ago to speed up tasks such as games and graphics. Therefore, it lacks precision and consistency.
In numerical systems, each mathematical operation introduces small rounding errors that can accumulate over time. Because of this, running the same AI model twice can give different results, leading to non-determinism. These discrepancies make it difficult to verify, reproduce, and audit AI decisions, especially in fields such as law, finance, and medicine. When AI outputs cannot be clearly explained or proven, they become “hallucinations.” This is a term coined to describe the “lack of provability” in AI.
Modern AI has a fundamental struggle with accuracy that lacks truth, creating an invisible wall. This flaw has become a severe limitation, impacting overall performance, increasing cost, and wasting energy on noise correction.
Modern AI struggles with accuracy without truth, creating an invisible wall. This flaw is a severe limitation, impacting performance, increasing cost, and wasting energy on computational noise correction.
The second trap is in architecture. Current AI models have no memory. Instead, it predicts the next frame or token without doing any inference to help achieve the prediction. The company says it’s just predictive text on steroids. Once a modern model outputs something, it does not retain the reasons why it made such decisions and cannot reconsider or build on its own inferences. AI may appear rational, but it’s only mimicking reasoning, not truly understanding how it reaches its conclusions.
“Counterintuitive is a veteran of the world’s leading research labs and technology companies and has built a world-class team of mathematicians, computer scientists, physicists, and engineers who understand and solve the fundamentals of the twin trap,” Rego said.
Rego’s team has more than 80 patents pending, spanning deterministic inference hardware, causal memory systems, and software frameworks. These patents are thought to have the potential to “define the next generation of computing based on inference, not imitation.”
Counterintuitive’s inference-native computing research aims to create the first inference chips and software inference stacks that push AI beyond current limits.
The company’s Artificial Reasoning Unit (ARU) is a new type of computing rather than a processor that, unlike a GPU, focuses on memory-driven inference and performs causal logic in silicon. “Our ARU stack is more than a new chip category in development. It’s a complete departure from probabilistic computing,” said Syam Appala, co-founder of Counterintuitive.
“ARU ushered in the next era of computing, redefining intelligence from imitation to understanding, powering applications impacting the most important sectors of the economy without the need for large hardware, data centers or energy budgets.”
Counterintuitive aims to develop more reliable and auditable systems by integrating memory-driven causal logic into both hardware and software. This marks a shift from traditional speed-oriented, probabilistic AI black-box models to more transparent and responsible reasoning.
(Image source: “Abacus” by blaahhi is licensed under CC BY 2.0.)
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
