Transformer AI: Imagination & Human-Like Thought
- Researchers have developed a new artificial intelligence (AI) model, Co4, that mimics human thought processes by replicating the dual-input, state-dependent mechanisms found in the human neocortex. This innovative...
- Ahsan Adeel, an associate professor at the University of Stirling, spearheaded the study, which explores how AI models can reproduce higher mental states.
- The new transformer architecture, detailed in Adeel's paper published on arXiv, is inspired by interactions between external and internal information processing in layer 5 pyramidal two-point neurons (TPNs).
A groundbreaking new AI model, Co4, is making waves by emulating human thought processes with a novel transformer architecture, and News Directory 3 has the scoop.This cutting-edge AI aims too replicate the dual-input, state-dependent mechanisms found in the human neocortex. By drawing inspiration from the human brain, Co4 could achieve faster learning and lower computational demands compared to existing AI. The model uses triadic modulation loops to mimic problem-solving, with promising results in learning, vision, and language tasks, possibly revolutionizing how we approach artificial intelligence. Learn how researchers are pre-selecting relevant information, mirroring how humans filter information. Discover what’s next for this exciting technology.
New AI Model Emulates Human Mental States with Novel architecture
Updated May 29, 2025
Researchers have developed a new artificial intelligence (AI) model, Co4, that mimics human thought processes by replicating the dual-input, state-dependent mechanisms found in the human neocortex. This innovative approach to AI development could lead to faster learning and reduced computational demands.
Ahsan Adeel, an associate professor at the University of Stirling, spearheaded the study, which explores how AI models can reproduce higher mental states. The goal is to accelerate learning and reduce computational load by emulating perceptual reasoning and imaginative states.
The new transformer architecture, detailed in Adeel’s paper published on arXiv, is inspired by interactions between external and internal information processing in layer 5 pyramidal two-point neurons (TPNs). Current AI algorithms, while inspired by the human brain, frequently enough fall short of replicating high-level perceptual processing and creativity.
adeel’s work focuses on pre-selecting relevant information before applying full attention, mirroring how humans filter and prioritize information. This approach uses a reasoning pattern based on questions,clues,and hypotheses,adapting thinking processes over time to solve problems.
Attending to what is relevant is fundamental to both the mammalian brain and modern machine learning models such as transformers. Yet determining relevance remains a core challenge, traditionally offloaded to learning algorithms like backpropagation. Inspired by recent cellular neurobiological evidence linking neocortical pyramidal cells to distinct mental states, this work shows how models (e.g., transformers) can emulate high-level perceptual processing and awake thought (imagination) states to pre-select relevant information before applying attention.
The Co4 model employs triadic neuronal-level modulation loops among questions, clues, and hypotheses, enabling parallel reasoning chains. This leads to faster learning with reduced computational demand, operating at an approximate cost of O(N), where N is the number of input tokens.The model has shown promise in reinforcement learning, computer vision, and natural language question answering.
triadic neuronal-level modulation loops among questions (Q), clues (keys, K), and hypotheses (values, V) enable diverse, deep, parallel reasoning chains at the portrayal level and allow a rapid shift from initial biases to refined understanding. This leads to orders-of-magnitude faster learning with substantially reduced computational demand (e.g., fewer heads, layers, and tokens), at an approximate cost of O(N), where N is the number of input tokens. Results span reinforcement learning (e.g., CarRacing in a high-dimensional visual setup), computer vision, and natural language question answering.
Evaluations of the adapted transformer architecture in learning,computer vision,and language processing tasks have been promising. These results suggest that this new mechanism could significantly advance the reasoning skills of AI models, bringing them closer to human-like cognition.
What’s next
Future research will focus on expanding the capabilities of Co4 and exploring its potential applications in various fields. Adeel believes that emulating the cellular foundations of higher mental states could be a crucial step toward creating cognitively meaningful machine intelligence, shifting AI systems from mere information processing to contextual reasoning and real understanding.