Unlocking the secrets of the human brain might be the key to creating conscious AI, but can technology truly replicate the complexity of human awareness?
Research: Is artificial consciousness achievable? Lessons from the human brain
In a paper published in the journal Neural Networks, researchers conducted a comprehensive analysis of the development of artificial consciousness from an evolutionary perspective, using the human brain as a benchmark.
They argued that while artificial intelligence (AI) may face intrinsic and extrinsic limitations in emulating human consciousness, drawing upon the brain's complex structural and functional characteristics could be a promising approach.
They also suggested that AI might have partial or alternative forms of consciousness that are qualitatively different from human experience.
Finally, they recommended adopting caution when discussing artificial consciousness, advocating for greater clarity in defining AI's level and type of consciousness.
Background
Past work has established the brain as a physico-chemical system, with consciousness being one of its most complex features.
Researchers explored the theoretical possibility of emulating consciousness through artificial methods, noting the challenges as analogous to creating life in a lab.
The increasing interest in artificial consciousness raises social, ethical, and technical challenges, urging a multidisciplinary and nuanced approach.
Key challenges include defining and replicating the subjective experience central to consciousness and addressing concerns about the potential societal impact of conscious AI systems, such as unintended consequences or ethical dilemmas.
Brain-Inspired AI
AI has been inspired by a biological understanding of the brain through computational models and artificial neural networks (ANNs), though these models have a biological understanding of the brain's architecture.
While functionalism suggests mental capacities could be realized in different physical systems, theories like integrated information theory (IIT) question this view, emphasizing the importance of physical structure for consciousness.
Despite their success in applications like deep learning, ANNs still fall short of replicating the complexity of biological neurons fully.
The debate continues on whether AI can achieve human-like conscious processing or only functional emulation.
Artificial vs. Biological Consciousness
Traditional computers' hardware, built from semiconductor transistors and neuromorphic substrates, operates much faster than the brain but lacks the biochemical diversity and intricate organization crucial for conscious processing.
The brain's structure, composed of various proteins, neurons, and nested levels of organization, enables conscious processing through complex biochemical interactions, including the modulation of neurotransmitters and regulatory circuits.
While capable of simulating some brain functions, AI systems lack the chemical diversity and experiential depth essential for true consciousness.
Despite progress in neuromorphic computing and parallel processing, AI is limited by its architecture and speed, which differ from the brain's hierarchical, multilevel structure.
Consciousness in AI might eventually emerge, but current models still need to fully replicate the brain's organizational complexity, significantly influencing its cognitive and conscious capabilities.
Brain vs. AI
The evolution of the human brain, shaped by Darwinian processes, contrasts with the rigid, prewired architecture of AI systems. Brain development involves genetic conservation across species and epigenetic evolution, where synapse selection allows for learning and cultural adaptation.
This variability, influenced by individual experiences and environmental interactions, fundamentally distinguishes human consciousness from AI, which lacks such evolutionary and experiential development.
While AI systems, like deep reinforcement learning (DRL), can learn from experience, they do not possess the complex, hierarchical, and nested epigenetic processes that characterize human consciousness.
Due to its limited evolutionary and epigenetic capabilities, AI may reach basic or recursive consciousness but lack self-awareness or reflective consciousness. However, future advancements might bridge these gaps, allowing AI to develop more human-like cognitive processes.
AI vs. Consciousness
Current AI systems, predominantly functioning in an input-output mode, starkly contrast with the human brain, which operates in a predictive mode, continually testing hypotheses about the world and itself.
This predictive mode is underpinned by the brain's intrinsic spontaneous activity, a baseline function independent of external stimuli, crucial for conscious perception and creativity.
Unlike AI, which treats spontaneous activity as noise, the brain's spontaneous activity is structured, globally distributed, and linked to reward and emotional systems, enabling flexible and creative responses.
While neuromorphic and robotic AI approaches attempt to replicate aspects of these functions, they still lack the multidimensional and emotionally integrated depth inherent in human consciousness.
This limitation in AI highlights the challenges in achieving a genuinely conscious artificial system, as the current models lack the embodiment and emotional context crucial for human-like awareness.
Consciousness Distinctions Explored
The distinction between conscious and non-conscious processing, as described by Descartes' res cogitans and res extensa, highlights the physiological differences between these states.
Conscious processing, marked by "ignition" and increased neural activity around 200 to 300 milliseconds after stimulus onset, contrasts with the subdued activity of non-conscious processing.
This intense activity underpins complex reasoning, symbolic understanding, and language generation, which current AI, including large language models, continues to lack in semantic depth and emotional context. AI systems also fail to replicate human creativity and reasoning supported by intrinsic rewards and neural plasticity.
Future AI research should address these gaps and refine conceptual frameworks to better reflect advances in artificial consciousness.
Conclusion
To sum up, the review highlighted several key brain features crucial for human consciousness, such as hierarchical organization, reward sensitivity, and spontaneous activity, suggesting they could enhance artificial consciousness research.
Current AI limitations include insufficient brain data, limited hardware diversity, and a lack of emotionally relevant value attribution.
Although developing human-like artificial consciousness remains uncertain, recognizing these brain features could significantly advance AI systems.
The authors recommended that human and artificial consciousness be clearly differentiated and that the dimensions of consciousness artificial systems might exhibit be explicitly clarified.