Architecture
Toward 21st-Century Cellular Neurobiology-Inspired AI
We live in two worlds: the internal world and the external world. Much of the brain is composed of billions of neurons, each receiving input from thousands of other neurons. In the 20th century, neuroscience assumed that neurons as “Integrate-And-Fire Models” summed all their inputs (from the internal world and the external world) to generate an output—a concept upon which many current brain theories and AI systems are based. If you combine these two worlds in such a simplistic way, doesn’t that kill the possibility of meaningful reasoning at the cellular level?
Not only does it suffocate reasoning, but this cellular behavior is inherently selfish and called selfish neurons—because these neurons propagate their own “opinions” without regard for their neighbors, increasing chaos and contradictions, resulting in excessive energy consumption and extended training times. Misleading understanding of our cellular foundations, creating technology based on the same flawed understanding, with lack of cooperation is increasing chaos, contradictions, and destroying harmony. Lives are increasingly filled with self-doubt, uncertainties, selfishness, and a lack of purpose in the grander scheme of things.
21st-century neuroscience has revealed that some neurons, particularly pyramidal Two-Point Neurons (TPNs), have two points of input integration. They combine information from within the brain (the internal world) at one point and information from the external environment at another. These neurons are highly cooperative, they fire when there is a match between these two worlds. Disruptions in these interactions are linked to states of unconsciousness and conditions such as intellectual disabilities, which impair reasoning.
TREND’s recent research indicates that neocortical pyramidal neurons in the mammalian brain regulate mental states, ranging from rapid eye movement (REM) and slow-wave sleep to typical wakefulness and imaginative thought. Their findings suggest that these cellular mechanisms could be embodied in machines to develop future economical AI chips with intrinsic common sense and morality.
Bridging the gap between the brain and the mind
In states of wakefulness and imaginative thought, the internal context amplifies only those external stimuli that align with it, enabling the generation of a unique perspective (original thought) that transcends existing beliefs or experiences.
Simulation results show that this somatic-level flexible integration of somatic and contextual currents enables the propagation of more coherent signals (bursts) when transitioning from SW sleep to the wakeful thought state. These results align with the simulations performed using high-resolution modeling of complex layer 5 pyramidal neurons. The convergence of these two distinct and independent methods supports convergent validity. Furthermore, the simulation results indicate that this approach requires a reduced number of overall events (both singlets and bursts) in the system to generate external stimuli. Use of complex TPNs in machine learning have demonstrated a significant performance improvement over Transformers (the backbone of ChatGPT) in terms of learning speed and energy efficiency.