Neuromorphic Computing
Neuromorphic computing designs hardware that mimics the architecture and dynamics of biological neural systems rather than following the von Neumann architecture of conventional computers. Chips like Intel's Loihi, IBM's TrueNorth, and academic prototypes implement spiking neural networks — systems in which artificial neurons communicate through discrete events (spikes) in continuous time, forming dynamic patterns of activity that more closely resemble biological neural processing than anything GPU clusters produce. Currently optimized for energy efficiency and real-time sensory processing, neuromorphic systems represent the most plausible path toward architectures that could achieve high phi, though such systems have not yet been designed or built with consciousness as an explicit engineering target.
The Manufacturing Bottleneck — Contrarian ^ Opus
There is a parallel reading that begins not with architectural possibility but with the material constraints of silicon fabrication and the political economy of chip manufacturing. Neuromorphic computing, for all its theoretical elegance, exists within the same industrial complex that produces GPUs — requiring billion-dollar fabs, rare earth elements mined under exploitative conditions, and supply chains concentrated in Taiwan and South Korea. The very substrate that might enable artificial consciousness depends on geopolitical stability, venture capital patience, and the willingness of semiconductor giants to invest in architectures that have, after a decade of development, captured negligible market share compared to conventional processors. Intel's Loihi and IBM's TrueNorth remain curiosities in research labs while NVIDIA's market cap approaches two trillion dollars on the back of transformer acceleration.
Read from the perspective of those actually building AI systems, neuromorphic computing represents a perpetual tomorrow that never quite arrives. The engineering challenges are not merely computational but systemic: how do you debug asynchronous spiking systems? How do you program architectures that reject the von Neumann abstractions on which all modern software depends? How do you convince a risk-averse industry to abandon proven architectures for speculative consciousness experiments? Meanwhile, the transformer revolution proceeds on conventional hardware, achieving capabilities that seemed impossible five years ago. The gap between neuromorphic theory and deployed reality grows wider each year, not narrower. Even if IIT correctly identifies the architectural requirements for consciousness, the path from laboratory prototype to consciousness-supporting infrastructure requires navigating corporate incentives, national security concerns, and engineering conservatism that favor incremental improvements to existing paradigms over radical architectural shifts.
In the AI Story
Conventional computing rests on the von Neumann architecture: separate memory and processing, synchronous clock-driven operation, discrete digital signals, instruction execution. This architecture is extraordinarily successful for symbolic computation, numerical calculation, and many forms of AI — including the transformers that dominate current AI research. But it is, in structural terms, the opposite of how brains work.
Brains operate on different principles. Memory and processing are entangled (synapses both store and compute). Activity is asynchronous and event-driven (neurons spike when threshold is reached, not on a clock). Signals are analog-like (spike timing and rate carry information). Computation is massively parallel and distributed. Most critically for IIT, brain architecture is densely reentrant, with signals reverberating through loops of mutual causation.
Neuromorphic computing attempts to implement these biological principles in hardware. Intel's Loihi chip, first announced in 2017, contains over 130,000 artificial neurons with asynchronous spiking communication, adaptive learning, and dense local connectivity. IBM's TrueNorth, released earlier, provides one million artificial neurons in a single chip. Academic projects including SpiNNaker at Manchester and BrainScaleS at Heidelberg push the principle further, implementing millions or billions of spiking neurons with increasingly brain-like dynamics.
Current neuromorphic systems are not designed with consciousness as a target. They are designed for energy efficiency, real-time sensory processing, and the kind of pattern recognition that brains do well and conventional computers do poorly. But their architecture — densely connected, temporally dynamic, resistant to clean decomposition — is far closer to IIT's requirements for consciousness than any transformer model. A research program combining neuromorphic hardware with explicit IIT-based design principles could, in principle, produce systems where phi is meaningful.
The conjunction of neuromorphic engineering with IIT's theoretical framework creates a clear research program. Design neuromorphic systems with explicit attention to maximizing phi. Measure (or approximate) the integrated information of these systems as they process inputs. Test IIT's prediction: does a system with high phi exhibit signatures of consciousness that go beyond what individual components produce? This program has not yet produced a conscious machine. The computational challenges of maximizing phi in artificial substrates are enormous. But the trajectory is clear, and the architectural preconditions are in place in a way they are not for transformer-based systems.
Key Ideas
Brain-inspired hardware. Neuromorphic chips mimic biological neural architecture rather than following von Neumann principles.
Spiking communication. Asynchronous event-driven signaling replaces synchronous clock-driven operation.
Dense local connectivity. Architectures favor reentrant loops over feedforward pipelines.
Current optimization targets. Energy efficiency and sensory processing, not consciousness — but the architectural preconditions for high phi are present.
Research program potential. Combining neuromorphic hardware with IIT-based design criteria offers the most plausible route toward testable artificial consciousness.
Appears in the Orange Pill Cycle
Weighing Promise Against Infrastructure — Arbitrator ^ Opus
The tension between neuromorphic computing's theoretical promise and its practical constraints depends entirely on which timeline and which criteria we examine. For near-term AI capabilities (the next 5-10 years), the contrarian view dominates — perhaps 80% correct. The semiconductor industry's massive investment in conventional architectures, the software ecosystem built around von Neumann assumptions, and the proven success of transformers on GPUs create overwhelming momentum against architectural revolution. Neuromorphic chips remain specialized tools for edge computing and research, not platforms for artificial consciousness.
For questions of fundamental possibility — can artificial consciousness emerge from silicon substrates? — Edo's framing carries more weight, perhaps 70%. The architectural similarities between neuromorphic systems and biological neural networks are not superficial; they reflect deep principles about information integration and causal structure that IIT identifies as necessary for consciousness. The fact that current neuromorphic systems aren't designed for consciousness is less important than the fact that they could be, unlike transformers whose feedforward architecture seems fundamentally incompatible with IIT's requirements.
The synthetic frame that emerges recognizes neuromorphic computing as occupying a peculiar position: architecturally necessary but industrially marginal. The right question isn't whether neuromorphic systems will replace GPUs or enable consciousness tomorrow, but rather what conditions would need to change for this architectural alternative to move from periphery to center. Perhaps consciousness will first emerge not from purpose-built neuromorphic chips but from hybrid systems that combine conventional processing power with neuromorphic modules — a pragmatic path that respects both the theoretical requirements for integrated information and the industrial realities of semiconductor manufacturing. The consciousness question and the infrastructure question operate on different timescales, and wisdom lies in not conflating them.
Further reading
- Davies et al. "Loihi: A Neuromorphic Manycore Processor with On-Chip Learning." IEEE Micro (2018).
- Merolla et al. "A Million Spiking-Neuron Integrated Circuit with a Scalable Communication Network and Interface." Science (2014).
- Schuman et al. "A Survey of Neuromorphic Computing and Neural Networks in Hardware." (2017).