Over the past year, generative Artificial Intelligence (AI) models like ChatGPT and DALL-E have made it possible to generate an enormous amount of high-quality, creative content that appears to have been created by humans from a limited set of commands.
Although current AI systems are quite powerful and consistently outperform humans, especially in big data pattern recognition jobs, they are not as smart as we are. AI systems do not learn and are not configured in the same way as the human brain.
AI systems likewise consume enormous amounts of energy and resources during their training phases, compared to when we eat three or more meals a day. Unlike humans, they are less able to adapt and function in dynamic, unpredictable and noisy environments and do not have human-like memory capacities.
The study explores non-biological systems that are more similar to the human brain. In a new study published in Science Advances, we discovered that self-organizing networks of tiny silver wires learn and remember in a way similar to the thinking mechanism in the human brain.
Our research falls under the umbrella of the discipline known as neuromorphics, which seeks to mimic the structure and functioning of biological neurons and synapses in artificial systems.
Our work focuses on a technology that mimics the brain's synapses and neurons using a "nanotel" network. The width of these extremely small nanowires is equivalent to that of a human hair. They are usually coated with an insulating material such as plastic and are made of a highly conductive metal such as silver.
Self-assembled nanowires form a network topology like a biological neural network. Each metal nanowire is covered with a thin insulating layer, similar to how neurons have an insulating membrane.
When we excite the nanowires with electrical signals (such as neurotransmitters across synapses) ions flow through the insulating layer and into a nearby nanowire. As a result, we see electrical signals resembling synapses in nanowire networks.
Is Synthetic Intelligence Possible?
Our latest research explores the possibility of human-like intelligence through a nanowire system. Two features suggestive of higher cognitive function—learning and memory—are central to our research.
Our research shows that synaptic pathways in nanowire networks can be selectively strengthened (and weakened). “supervised learning” in the brain is similar to this. In this procedure, the output of synapses is compared with a desired result. Synapses are then clipped or strengthened.
We enhanced this finding by showing that we can increase the amount of reinforcement by "rewarding" or "punishing" the network. The brain's "reinforcement learning" was the inspiration for this procedure.
We also used a variation of the "n-back task," a test that assesses working memory in humans. A series of stimuli is presented in succession, and each new entry is compared to what was n steps ago.
Previous signals were “remembered” by the network for at least seven steps. Strangely enough, people are believed to be able to hold seven things at once in their working memory.
The memory performance of the network improved significantly when we used reinforcement learning.
We discovered that in our nanowire networks, how synapses were previously stimulated affects how synaptic pathways are formed. The brain's synapses also exhibit the same behavior that neuroscientists call "metaplasticity."
The re-creation of human intelligence is probably still a very remote possibility.
But our work on neuromorphic nanowire networks shows that it is possible to incorporate qualities essential for intelligence, such as memory and learning, into non-biological, physical technology.
Nanowire networks are different from artificial neural networks related to artificial intelligence. They can still produce a result called "synthetic intelligence".
Perhaps one day, a neuromorphic nanowire network will be able to remember discussions that are more like human interactions than ChatGPT.
Günceleme: 30/04/2023 10:15