Page 224 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

5. Pragmatic Implementation of Complementarity for New AI 215

intelligence (bottom-up, top-down, and brain-like). The consequences of complemen-
tarity to building AI are discussed next.

5. PRAGMATIC IMPLEMENTATION OF COMPLEMENTARITY
FOR NEW AI

Intelligence in human brains is the result of a delicate balance between fragmen-
tation of local components and global dominance of a coherent overall state.
The balancing act happens in brains through intermittent transitions between
synchronous and nonsynchronous states. This leads to the cinematic theory of hu-
man cognition [32,34]. Human intelligence is manifested through the emergence
of a sequence of coherent metastable amplitude patterns. Any particular metastable
pattern exists only for a brief period of 0.1e0.2 s; this is why it is called
metastable. The pattern ultimately collapses in a rapid desynchronization effect,
followed by the emergence of a new metastable pattern. According to the cine-
matic theory of cognition, the metastable patterns represent the movie frame,
and the brief desynchronization period corresponds to the shutter. The collapse
of spatial patterns is the manifestation of a spatial bottleneck in neural processing
that produces repeated singularities in time in the form of the shutter. This is a
spatiotemporal effect in brain dynamics with nonlocal interference of temporal
and spatial scales. Human intelligence works though these repeated singularities
to produce the brain’s clockwork [49], which in turn creates knowledge and
meaning from disembodied information acquired through our sensory organs.
The desynchronization event plays a crucial role in cognition and consciousness;
it corresponds to the “Aha” moment of deep insight and understanding [37].
Neither top-down symbolic AI approaches nor pure bottom-up, deep learning neu-
ral networks can grasp this complex balancing process in its entirety. In the spirit of
the complementarity principle, the two aspects of intelligence must be integrated in a
unified approach. Brains tell us that this integration can happen through the sequence
of metastable patterns, separated by phase transitions in the singular spatiotemporal
brain dynamics. These transitions are not prescribed by an external agency; rather,
they inevitably emerge as the result of the brain’s neurodynamics. In order to substan-
tially advance the state-of-art of AI, we can rely on our understanding of the
mechanisms that underlie biological intelligence. Many current AI systems give
the appearance of intelligence, but cannot adapt to changing circumstances or truly
interpret and make decisions based on dynamic input. Deep learning has the potential
of moving toward these goals; however, resource constraints are often ignored in
DL settings. DL typically requires huge amount of data/time/parameters/energy/
computational power, which may not be readily available in several practically impor-
tant scenarios. Relevant task domains include rapid response to emergency situations
and disasters based on incomplete and disparate information, supporting graceful
degradation in the case of physical damage or resource constraints, and real-time