Skylark Labs unveils brain-inspired AI that learns and forgets like humans

Remember when artificial intelligence (AI) needed humans to learn anything new? That era might be ending. In a major leap for AI, Skylark Labs’ brain-inspired AI can continuously evolve on its own through a revolutionary automated mechanism that works exactly like the human brain, without any human guidance.

The making of an innovation

In a quiet Palo Alto office near Stanford, Skylark Labs founder and CEO Dr. Amarjot Singh demonstrates a standard security camera connected to a simple black box. What happens next, however, is anything but standard.

"Watch what happens when we introduce something the system has never seen before," says Dr. Singh, recipient of the 2025 Space and Defense Innovator Award, and an MIT 35 Under 35 honoree.

As an unfamiliar drone flies into the camera's view, the system does not just detect it — it begins to identify and learn this new trajectory completely on its own. No engineers are rushing to collect data, no developers are writing new code, and no datasets are needed. The AI teaches itself, resembling how a human brain would respond to novelty.

"This is what we've been working toward since day one," Dr. Singh explains. "AI that truly learns on the edge without datasets or retraining, an AI that works like a human brain, deciding for itself when to learn, what to learn, and how to integrate that knowledge."

Solving the limitations of traditional AI

As AI grows exponentially, a fundamental limitation has persistently held back practical applications: the rigid deployment cycle, drift, and retraining. Traditional AI systems handle new situations poorly: issues must be manually flagged, updates require costly retraining, and there’s no guarantee the update will ever be needed again.

This broken process has created persistent headaches across industries, and high false-positive rates often plague conventional systems. Skylark Labs’ breakthrough addresses these challenges through continuous adaptation, breaking the static model paradigm dominating machine learning. 

Its AI autonomously learns when engineers deploy it, without massive datasets or disruptive retraining cycles. It intelligently distinguishes between genuine threats and benign anomalies, dramatically reducing alert fatigue and focusing on what truly matters.

The brain-inspired architecture hybrid AI that learns and forgets like humans

What makes Skylark Labs’ system truly revolutionary is its brain-inspired design, which works like human memory. The AI can instantly spot what is important in a sea of information, learning what matters and forgetting what does not. 

Unlike traditional AI, which requires massive datasets and constant retraining, Skylark Labs’ brain-inspired hybrid AI learns on-device without needing training or datasets on what to remember. This is true for both the short and long term, just like human brains do. It mimics the human brain with specialized components that combine supervised and unsupervised learning. 

From capturing simple features in early layers to building complex patterns in the middle and forming high-level concepts in the final layers, the system processes information like the human brain.

"Our system mimics how the neocortex and hippocampus work together," explains Singh. "It's not processing data blindly; it's self-aware of what needs learning. When encountering something new, it immediately learns from that single example and recognizes similar instances later. But here's the biological twist that changes everything: the system autonomously decides what memories to strengthen and what to forget based on frequency and recency."

Skylark Labs’ AI adapts on the fly, learning directly on-device without massive datasets or heavy compute. New patterns are held in a fluid state, kept in short-term memory if recently seen, then gradually forgotten if they don’t reappear. But if a pattern returns often, it’s reinforced and moved to long-term memory, expanding the system’s knowledge without overwriting what it already knows.

This method solves a critical problem that plagues AI: the inability to determine what information matters amid the noise. The system automatically manages both short-term and long-term memory and intelligently forgets irrelevant patterns, preventing memory bloat. It focuses resources only on what's truly relevant, much like the human brain's efficiency in memory management.

"Our architecture mirrors the brain's remarkable efficiency," Dr. Singh explains. "Just as your brain consolidates important memories during sleep while discarding trivial details, our system continuously evaluates and organizes knowledge, strengthening what matters and releasing what doesn't." 

As a result, the system instantly identifies threats and anomalies without needing the massive computational infrastructure that traditional AI requires, making it extraordinarily resource-efficient. 

Dr. Singh continues, "Either invest millions in retraining for anomalies that might never reappear, or accept the risk of missing critical events your system isn't prepared to handle. Our perspective eliminates this dilemma entirely." 

Real people, real impact: Seeing self-evolution save lives

The actual test of any technology is not in the lab but in the field. Skylark Labs’ flagship Kepler platform has a brain-inspired hybrid AI architecture that fuses unsupervised and supervised learning across multiple data pipelines. 

For Col. PK Dubey, who oversees drone security at a sensitive military installation, Kepler has fundamentally transformed threat detection capabilities. "Last month, we detected an unidentified drone using a completely novel process to avoid standard detection methods," Dubey explains. 

"Traditional systems would have missed it entirely until we manually updated them, which could take weeks. Kepler identified it as anomalous using its flight trajectory and adapted its detection parameters on the spot. By the second pass, it was tracking the drone with precision."

Similar to how Skylark Labs’ fixed systems adapt to evolving threats, the Tracer AI Mobile Platform brings that same intelligence on the move. Now deployed in runway safety operations, Tracer detects and tracks Foreign Object Debris (FOD) in real time as it learns directly from its environment. 

Col. (Ret.) Karir, who leads a major airbase’s runway safety project, explains, “FOD changes constantly—plastic wrap, loose bolts, torn rubber. Tracer adapts on the fly without retraining, identifying new debris and adjusting in real time.” 

Powered by Skylark Labs’ brain-inspired AI, Tracer runs on mobile patrol vehicles. It processes data on-device, filters false positives, and instantly shares updates across the network. With each pass, every sweep becomes smarter, safer, and more responsive.

The potential applications for Kepler's continuously adaptive AI extend beyond security to numerous domains where conditions change rapidly. In transportation, when self-driving vehicles encounter unexpected road closures, the system can immediately update environmental models and navigation policies based on sensor data, without waiting for fleet-wide updates from engineers.

"When seconds count, you can't wait for someone to retrain an AI system," says Dr. Singh. "Lives literally depend on technology that can adapt in the moment, just like a human first responder would."

Beyond the hype: What self-evolving AI means for our future

Skylark Labs’ achievements stand out in a tech terrain filled with exaggerated claims but little measurable impact. According to Dr. Andrea Soltoggio from DARPA's Lifelong Learning Machines Program, "What makes Skylark Labs’ perspective revolutionary is solving the stability-plasticity dilemma that has frustrated researchers for decades. Their system continuously adapts without forgetting critical knowledge or becoming unstable."

For businesses, this breakthrough changes the economics of AI deployment. Traditional systems demand constant attention—data collection, model retraining, update rollouts—driving up costs and causing disruption.

“Think about upgrading your phone,” says Dr. Singh. “Everything stops during installation. We’ve built AI that keeps improving quietly in the background, without interruption.”

Industry experts see Skylark Labs’ approach as a path to more robust AI, cutting retraining costs and unlocking deployment in environments once considered too unpredictable. This could accelerate adoption in sectors where downtime isn’t an option and conditions shift constantly: healthcare adapting to new diseases, financial systems countering emerging fraud, or humanitarian tools responding to fast-changing disasters.

By enabling AI that learns continuously and independently, Skylark Labs is laying the groundwork for truly autonomous systems—not just smarter, but self-reliant. As Dr. Singh puts it, “The future of AI isn’t just intelligence—it’s independence.”