Human intelligence can be summarized as the way a function is used, in this case, memory. Simply, intelligence is defined as the quality of usage, of what is available in memory.
Usually, new information shapes decision-making. However, if the new information cannot be relayed to other parts—of memory—the information may not be useful for intelligence.
Human intelligence works, conceptually, by the variation of relays, across memory areas. Memory specificities [language, numeracy, subjects, and so forth] are critical, but the ability to crisscross destinations deepens intelligence.
Neuroscience has several labels for memory—procedural, long-term, short-term, episodic, semantic, and so forth—but there are no labels for relays. There are labels of pathways, like corticolimbic and others, but there are no definitions of how relays from one memory area to another define how intelligence works.
Even prediction, which is often assumed to be one of the things the brain does, is more about relays than actual prediction. The human mind almost has a balance—of importance—for functions and for relays, with determinations for outcomes.
Humans can still drive better than machines in chaotic conditions because the human mind has better relays than machines. Human focus could be knocked off by some distraction, a sound, a severe temperature, and others because relays to necessary areas of memory are diverted—albeit the memory is present. Some aspects of forgetfulness can also be defined as problems of relay. Planning, observation, analysis, and so on, are all tied to relays in the mind. These relays exceed neural codes, since they often involve chemical signals too, not just electrical signals.
Relays are critical to the human mind, including intelligence. There cannot be any model for how the mind works, or for human intelligence, without a delineation of relays. The relays are not just arrows between blocks of labeled memories, but how the components of the mind [electrical and chemical signals] relay.
Measuring digital or machine intelligence may be determined by the level of relays, compared to the human mind. AI safety and alignment could be better with models around relays. Already, large language models have prediction [as relays]. Safety may involve other relays, like consequences, even as newer relays make them better.
There is a recent paper in The European Physical Journal B, Brain-inspired computing systems: a systematic literature review, stating that “Brain-inspired computing is a growing and interdisciplinary area of research that investigates how the computational principles of the biological brain can be translated into hardware design to achieve improved energy efficiency. Brain-inspired computing encompasses various subfields, including neuromorphic and in-memory computing, that have been shown to outperform traditional digital hardware in executing specific tasks. With the rising demand for more powerful yet energy-efficient hardware for large-scale artificial neural networks, brain-inspired computing is emerging as a promising solution for enabling energy-efficient computing and expanding AI to the edge. However, the vast scope of the field has made it challenging to compare and assess the effectiveness of the solutions compared to state-of-the-art digital counterparts.”
Digital memory is already excellent in several regards. However, the dimensionality of relays is the current limitation of AI to human intelligence and beyond. There, however, is the likelihood of better relays towards superintelligence.
There is a recent feature in Scientific American, The “Fight or Flight” Idea Misses the Beauty of what the Brain Really Does, stating that, “brains operate mainly by prediction, not reaction. All brains constantly anticipate the needs of the body and attempt to meet those needs before they arise. They seek to reduce uncertainty to survive and thrive in circumstances that are only partially predictable.”
Simply, the mind relays in different directions. The observation of prediction is theorized to be a relay called split, where some, in a set of electrical signals relay ahead of others to interact with chemical signals like they had previously, such that if the interpretation matches the perception, then the incoming electrical signals follow in the same direction, if not, they go in another direction, correcting what is labeled prediction error. This relay explains predictive coding and processing.
Listing all the possible relays in the mind, for a standard of what might become of machine intelligence, could determine their approach and how to align them to human values.
