Future Brain
Non-Invasive Neuromorphic Computing (NINC)
I’m always on the lookout for the next technology set. In 2017, I went to MIT for their AI Strategy program and it opened up a new world for me, writing my first book for business in 2019. In 2021, I launched the Quantum Strategy Institute to support the acceleration of the adoption of quantum technologies. In 2025, I’m thinking about the advance of robotics and how humans can keep pace leveraging AI and quantum.
Enter neuromorphic computing. Not necessarily the Cyborg kind (the kind where wires and tubes enter your body). Rather, the kind you wear and store away. The kind that let’s you still be fully human if that’s your choice or you have that option. I call it Non-Invasive Neuromorphic Computing or NINC.
Image source: Brian Lenahan/Midjourney
Non-invasive neuromorphic computing, as a specific term, isn't widely established in the literature, but the concept can be interpreted as neuromorphic computing applied in non-invasive ways, particularly in contexts like brain-computer interfaces (BCIs) or systems that interact with biological neural systems without physical intrusion.
What is Neuromorphic Computing?
Neuromorphic computing involves hardware and software designed to mimic the brain's neural architecture, using spiking neural networks or analog circuits to process information efficiently. It’s typically used for tasks like pattern recognition, sensory processing, or low-power computing in devices.
Why Non-Invasive
As I mentioned, the ‘Borg’ in Start Trek (TM) to me seemed like they were being controlled - yes collectively - yet still controlled. If the tech were non-invasive, one could remove it at will. The term "non-invasive" is more common in BCIs, where non-invasive methods (e.g., EEG, fMRI) measure brain activity externally without surgical implants. Applying "non-invasive" to neuromorphic computing could imply systems that:
Interface with the brain or biological systems without physical intrusion.
Operate externally to process neural-like data in a brain-inspired way.
Use neuromorphic principles for non-intrusive applications, like wearable devices or ambient computing.
Provide virtual links to traditional and quantum computers.
Research exists on combining neuromorphic computing with non-invasive BCIs. For example, EEG-based BCIs capture brain signals externally, and neuromorphic chips (like Intel’s Loihi) can process these signals efficiently due to their ability to handle sparse, event-driven data like neural spikes. This is non-invasive as it requires no implants, only scalp sensors.
Emerging wearable devices use neuromorphic processors for tasks like gesture recognition or vital sign monitoring, processing data from external sensors (e.g., skin-based or optical sensors) without penetrating the body. Studies (e.g., from IEEE or Nature journals) explore neuromorphic chips for processing EEG signals in real-time for applications like prosthetic control or epilepsy monitoring. These systems are non-invasive as they rely on external sensors. Companies like BrainChip are developing neuromorphic systems for edge computing, which could be applied to non-invasive health monitoring or IoT devices. Non-invasive approaches often face lower signal resolution (e.g., EEG vs. implanted electrodes), which neuromorphic systems can mitigate by efficiently handling noisy, sparse data. However, the field is still developing, and most neuromorphic applications remain in research or early commercial stages.
Feasibility and Future
Non-invasive neuromorphic computing is feasible and already emerging in specific niches, particularly where low-power, brain-inspired processing meets external sensing technologies. Future advancements could see neuromorphic chips integrated into AR glasses, smart wearables, or medical diagnostics, processing data from non-invasive sensors like EEG, EMG, or even optical brain imaging.
Adding Post-Quantum Cryptography (PQC):
PQC refers to cryptographic algorithms designed to resist attacks from quantum computers, which could break traditional public-key cryptography (e.g., RSA, ECC) using algorithms like Shor’s. PQC methods include lattice-based cryptography (e.g., Kyber), code-based cryptography (e.g., McEliece), hash-based signatures (e.g., XMSS), and others, which are believed to be quantum-resistant due to their reliance on mathematical problems quantum computers struggle to solve efficiently. PQC can run on classical computers, making it compatible with existing infrastructure, unlike quantum cryptography (e.g., QKD), which requires quantum hardware.
PQC could secure data transmitted between GPT and neuromorphic systems or external devices. For example, lattice-based cryptography (e.g., CRYSTALS-Kyber) could encrypt EEG data or user inputs to protect against quantum attacks, ensuring confidentiality and integrity. Research suggests neuromorphic systems could enhance PQC. For instance, a 2024 study proposed a PQC-based neural network that maps code-based PQC to a neural structure, using non-linear activation functions and random ciphertext perturbations to increase security. Such an approach could be adapted for neuromorphic hardware, leveraging its parallel, event-driven nature to perform cryptographic tasks efficiently. For example, spiking neural networks (SNNs) could implement lightweight PQC algorithms for constrained devices, like wearables interfacing with Grok.
Integrating a GPT like Grok
Grok currently operates on GPU-based architectures within xAI’s Colossus supercomputer, designed for large-scale language model training and inference. Having a GPT to answers questions through a NINC would be handy wouldn’t it? There’s no public evidence that Grok uses neuromorphic hardware or PQC directly but I selected Grok because of its performance in 2025 with select examples below.
Artificial Analysis Intelligence Index (AAII): Grok 4 topped the AAII with a score of 73 points, slightly ahead of Gemini 2.5 Pro and OpenAI’s o4-mini-high.
Humanity’s Last Exam (HLE): Grok 4 scored 25.4% without tools, improving to 38.6% with tools and 44.4% for Grok 4 Heavy (using multiple AI agents). This compares to Gemini-Pro (26.9% with tools) and OpenAI’s o3 (24.9% with tools). HLE, a 2,500-question benchmark across 100 disciplines, tests academic knowledge and reasoning.
GPQA (Graduate-Level Question Answering): Grok 4 scored 88.9%, the highest recorded, outperforming competitors.
Math Arena: Grok 4 achieved 96.7%, dominating mathematical reasoning benchmarks.
USA Math Olympiad: Grok 4 scored 79.4%, showcasing strong performance in competition-level math.
LiveCodeBench (Coding): Grok 4 is noted as a top-tier coder, with a perfect 100% score in the AI and Machine Learning 2025 Challenge.
VendingBench: Grok 4 excelled in this real-world task simulation, managing inventory and pricing effectively.
MMLU Score: Grok 4 scored 86.6%, indicating high-quality language understanding, though slightly below Grok 3’s 92.7%.
Context Window: Grok 4 has a 260k token context window, smaller than Grok 3’s but still competitive.
Text Arena: Grok 4 tied for #3 overall, ranking #1 in Math, #2 in Coding, and #3 in Hard Prompts based on 4,000+ community votes.
Integration with Non-Invasive Neuromorphic Computing:
Grok could interface with neuromorphic hardware to process data from non-invasive sensors (e.g., EEG for BCIs or wearables for user interaction). For example, a neuromorphic chip could preprocess sparse, event-driven data (like brain signals) before feeding it to Grok for conversational or analytical tasks. This is plausible given xAI’s ties to Neuralink, where Grok has been used to interpret neural signals, albeit from invasive implants.
Neuromorphic Advantages: Neuromorphic systems are energy-efficient and adept at handling noisy, real-time data, which could enhance a GPT’s ability to operate in edge environments (e.g., Tesla’s Optimus or IoT devices). For instance, a neuromorphic chip could process EEG signals for a non-invasive BCI, enabling a GPT to respond to thought-based inputs without implants.
Challenges: Neuromorphic hardware is still nascent, with limited software ecosystems and scalability compared to GPUs. Integrating it with a GPT would require significant engineering to bridge neuromorphic processing with, for example, Grok’s transformer-based architecture. Additionally, non-invasive sensors like EEG have lower signal resolution, which could limit performance compared to invasive methods.
"Unhackable" Potential:
Quantum Resistance: PQC provides robust protection against quantum attacks (e.g., Shor’s algorithm), as algorithms like lattice-based or code-based cryptography rely on problems quantum computers cannot solve efficiently. Integrating PQC with neuromorphic systems could secure data pipelines for a GPT, ensuring quantum-resistant communication between users, sensors, and it’s servers.
Neuromorphic Security Benefits: Neuromorphic systems’ event-driven nature and low power consumption make them less susceptible to certain side-channel attacks (e.g., power analysis), as they generate less predictable electrical signatures. Additionally, their ability to learn and adapt could enable dynamic cryptographic protocols, such as adjusting encryption based on detected threats, though this is speculative and requires further research.
Limitations to "Unhackable":
Implementation Risks: Even PQC is not immune to implementation errors, software bugs, or side-channel attacks (e.g., timing attacks). Neuromorphic systems, being complex and novel, may introduce new vulnerabilities, such as adversarial inputs that mislead their learning processes.
Non-Cryptographic Threats: Hacking often exploits non-cryptographic weaknesses, like social engineering or software vulnerabilities. Grok’s integration with neuromorphic systems could introduce new attack surfaces, especially if the software stack for neuromorphic hardware is immature.
Quantum Cryptography Alternative: Quantum key distribution (QKD) offers theoretically unhackable encryption using quantum mechanics (e.g., entangled photons), but it requires specialized hardware and is not feasible for non-invasive neuromorphic systems or Grok’s current infrastructure. PQC is more practical but not provably unhackable.
Summary
Non-invasive neuromorphic computing is merely a concept today yet could enable AI to process data from external sensors (e.g., EEG for BCIs) efficiently, enhancing applications like thought-based interaction in the human robot competition. PQC could secure these data flows against quantum attacks, as could QKD. The combination could approach "unhackable" status by resisting quantum cryptanalysis and reducing certain side-channel risks, but no system is truly unhackable due to implementation flaws and non-cryptographic vulnerabilities. Certainly the "unhackable" concept is aspirational, as PQC provides quantum resistance but not absolute security, and neuromorphic systems introduce new complexities. One can dream however as humans strive to compete with robots while maintaining their humanity in the future.
Brian Lenahan is founder and chair of the Quantum Strategy Institute, author of seven Amazon published books on quantum technologies and artificial intelligence and a Substack Top 100 Rising in Technology. Brian’s focus on the practical side of technology ensures you will get the guidance and inspiration you need to gain value from quantum now and into the future. Brian does not purport to be an expert in each field or subfield for which he provides science communication.
Brian’s books are available on Amazon. Quantum Strategy for Business course is available on the QURECA platform.
Copyright © 2025 Aquitaine Innovation Advisors