The Premise: AI Beyond the Lab Coat
Inherent’s King’s Cross office in London doesn't feel like a typical AI research lab. A sign at the entrance reads: “Recording in progress,” a nod to their philosophy of documenting and sharing their work, even the unconventional parts. Unlike many AI firms that focus on incremental improvements to existing models, Inherent aims to build AI systems that are fundamentally different. Their approach is less about refining current capabilities and more about exploring novel architectures and emergent behaviors. This means their research often leads them down paths that blur the lines between theoretical exploration and practical, albeit sometimes strange, product development.
The lab is backed by a $75m fund from the Deep Future fund, with significant investment from prominent figures like Matt Clifford (CEO of Entrepreneur First), Lak Ananth (ex-CEO of M12), and Alex Iskander (founder of sophisticated AI systems). This funding signals a strong belief in Inherent's ambitious vision, which diverges from the more common pursuit of scaling existing large language models. Instead, Inherent seeks to create AI that exhibits genuine understanding and agency, moving beyond pattern matching to something more akin to reasoning.
Beyond 'Just' LLMs: The Inherent Approach
Inherent's manifesto is built on the idea of 'AI-native' systems. This means designing AI from the ground up, rather than retrofitting existing architectures. They are exploring a range of research areas, including the development of novel neural network architectures, advanced reinforcement learning techniques, and the creation of synthetic data to train these models. A key focus is on building AI that can perform complex, multi-step reasoning tasks, moving beyond the limitations of current LLMs that often struggle with deep logical inference or maintaining long-term context.
One of the most surprising aspects of Inherent's work is their willingness to publicly document their research process. This transparency, while unusual in the competitive AI landscape, reflects a belief that open exploration can accelerate progress. They are not just building AI; they are building a narrative around AI development, inviting scrutiny and collaboration. This approach is akin to a scientific community sharing open-source code and research papers, but applied to the proprietary development of advanced AI systems.
The team is composed of individuals with diverse backgrounds, including former researchers from Google DeepMind and Meta AI. This blend of academic rigor and industry experience allows them to tackle complex problems from multiple angles. The lab’s structure emphasizes interdisciplinary collaboration, bringing together experts in machine learning, neuroscience, and even philosophy to inform their AI designs.
Experimental Products and Emergent Capabilities
Inherent's research isn't confined to academic papers. They are actively developing experimental products that showcase their AI's capabilities. One notable project involves an AI agent designed for complex task execution. This agent is trained to understand and carry out intricate instructions, a significant step beyond current AI assistants. They are also exploring AI systems that can learn and adapt in real-time, a crucial feature for AI operating in dynamic environments.
The lab's work on 'synthetic' data generation is another area of focus. By creating artificial datasets, Inherent can train AI models on scenarios that are rare or impossible to encounter in the real world. This allows them to push the boundaries of AI robustness and generalization. They are also investigating AI architectures that can exhibit more sophisticated forms of learning, such as meta-learning (learning to learn) and few-shot learning (learning from minimal examples).
Inherent's approach to AI development is characterized by a high tolerance for experimentation and a willingness to embrace the unexpected. They are not afraid to pursue ambitious, long-term goals that may not yield immediate commercial results. This experimental ethos is evident in their office environment, which is designed to foster creativity and serendipitous discovery. The company is reportedly building a 15-minute AI, a system that can learn a new task in that timeframe, a feat that would represent a significant leap in AI adaptability.
The 'Weirdness' Factor: Redefining AI's Boundaries
What makes Inherent 'weird'? It's their refusal to be constrained by conventional AI paradigms. They are not just building AI that can answer questions or generate text; they are aiming for AI that can truly understand, reason, and interact with the world in novel ways. This includes exploring AI that can generate its own hypotheses, design experiments, and even develop new scientific theories. Their work touches on fundamental questions about intelligence itself.
The lab is also investing heavily in foundational research, particularly in areas like causal inference and neuro-symbolic AI. Causal inference aims to enable AI to understand cause-and-effect relationships, a critical component of true reasoning. Neuro-symbolic AI seeks to combine the strengths of deep learning (pattern recognition) with symbolic reasoning (logic and knowledge representation). This fusion is seen by many as a key pathway to achieving more general and robust AI capabilities.
Inherent's long-term vision is to create AI that can act as a genuine partner in scientific discovery and problem-solving. They envision AI systems that can collaborate with humans, augment our intelligence, and tackle some of the world's most pressing challenges. This ambition, coupled with their unconventional methods and research directions, positions Inherent as a unique and potentially transformative player in the AI landscape.
The Future: What's Next for Inherent?
The implications of Inherent's work extend beyond the immediate applications of their experimental products. By pushing the boundaries of AI research, they are contributing to a deeper understanding of intelligence itself. Their focus on emergent properties and novel architectures could lead to breakthroughs that redefine what AI is capable of. For developers, this means potential new tools and platforms that offer unprecedented levels of autonomy and reasoning.
For founders, Inherent's approach suggests a shift in the AI investment landscape, moving beyond incremental improvements to funding more fundamental, long-term research. The success of their ambitious projects could signal a new wave of AI companies focused on creating genuinely intelligent systems rather than just optimizing existing ones. The question remains: can Inherent translate their ambitious research into widely adopted technologies, and what unforeseen challenges will arise as their AI systems become more sophisticated and integrated into the real world?
The lab's commitment to open documentation, while a competitive risk, could also be a strategic advantage, fostering a community and attracting top talent. As Inherent continues to explore the frontiers of AI, their journey will undoubtedly be one to watch for anyone interested in the future of artificial intelligence.