InReason | Neuro-Symbolic AI

Benefits

Explainable AI

InReason is a white-box, explainable AI system. Its internal operations are observable, explainable, verifiable, and auditable. Operators can assess safety, examine the system’s reasoning process, and maintain informed oversight of AI decisions.

Black Box
Neural AI

White Box
Symbolic AI

High Stakes AI

Designed for domains such as medicine, law and finance that require high reliability. Symbolic reasoning reduces hallucinations and enables dependable decision-making in mission-critical applications. Reliability comes from the architecture itself and not from adhoc safeguards.

Strong Reasoning

InReason builds on decades of work in symbolic reasoning from the fields of mathematics and logic. While the neural AI systems (such as LLMs) tend to lose accuracy as problems require deeper reasoning, symbolic reasoning maintains logical consistency even as the reasoning depth increases.

Learning From Small Data

InReason NeuroSymbolic AI is not data hungry and does not depend on the Scaling Laws of LLMs. It learns new concepts from just a few examples. And it updates its knowledge as it encounters new information during operation.

Neural AI: Triumps & Limits

The Deep Learning revolution had produced a series of Neural Network architectures that solved problems previously beyond the reach of computers. Transformers, the architecure behind the LLMs, are the latest in this progression. The Triumphs of the Neural Networks include

Human Perception tasks such as Object Detection & Speech Recognition.
Aggregation tasks such as Sentiment Analysis & Recommendation systems.
Chat systems like ChatGPT that assist with everyday tasks.
Unsupervised learning from internet scale.

However, the following limitations are inherent to these architectures and pose challenges for their deployment in critical applications.

LIMITATION 01

Black-Box Opacity

The internal operations of the neural networks are hidden within the billions of weights, making it impossible to audit or verify high-stakes decisions.

LIMITATION 02

Stochastic Hallucination

Outputs are based on probability and not on reasoning. Without a proper model, the system cannot distinguish truth from statistical correlation.

LIMITATION 03

Reasoning Decay

As the depth of a logical problem increases, the accuracy of neural networks drop sharply. They lack the machinery required to maintain the accuracy.

InReason NeuroSymbolic Paradigm

InReason Architecture

Neuro Symbolic Synergy

InReason brings together the complementary strengths of neural AI and symbolic AI to overcome their respective limitations. AI problems are divided into intutive tasks and reasoning tasks. Neural modules handle the intuitive tasks while the symbolic modules handle the reasoning tasks.

Intutive tasks include the human perception tasks, translation between natural and symbolic domains, predicting intermediate moves during symbolic reasoning. Symbolic tasks include logical reasoning, verification, problem solving and math.

Our Research Focus

Unsupervised learning of symbolic rules.
NeuralAI to translate between natural and symbolic domains.

"LLMs have a very limited understanding of logic... cannot reason in any reasonable definition of the term."

YANN LECUN

"Because scaling sucked out all the air in the room, everyone started to do the same thing. We got to the point where we are in a world where there are more companies than ideas by quite a bit."

ILYA SUTSKEVER

"LLMs work best *in conjunction* with systems that can actually plan or reason -- such as symbolic planners."

FRANÇOIS CHOLLET

"We need to be able to extend deep learning to do things like reasoning, learning causality, and exploring the world in order to learn and acquire information."

GARY MARCUS

"It’s very difficult to make sense of, how can the LLM, on the one hand, do these amazing things, and then on the other hand, repeat itself twice in some situation?"

ILYA SUTSKEVER

"Hallucination… is arguably the biggest hindrance to safely deploying these powerful LLMs into real-world production systems that impact people's lives."

Survey Paper on Hallucination

"LLMs Can't Plan, But Can Help Planning in LLM-Modulo Frameworks."

Subbarao Kambhampati

Contact InReason

Email

connect@inreason.ai

Phone

+91 9740169654

LinkedIn

Updates

Location

Prestige Lakeside Habitat, Gunjur,
Bengaluru, IN