The Question Of Neural Nativism – Dr. Joost Huizinga, Evolving Artificial Intelligence Laboratory, University Of Wyoming – Neural Networks And The Challenge To Improve Their Connections, Structures, And Autodidactic Capabilities

Dr. Joost Huizinga, a Ph.D. from the Evolving Artificial Intelligence Laboratory at the University of Wyoming, provides an informative overview of the advances and challenges in artificial intelligence (AI) learning and training.

Dr. Huizinga's interest and expertise are primarily focused in the areas of evolutionary robotics and algorithms, neural networks, learning algorithms and other elements of artificial intelligence. A portion of Huizinga's work involves neural networks and the challenge to improve them and make them more similar to the human brain. As neural networks are trained to classify images according to different categories, and learning algorithms then take command, the information about the images is still, unfortunately, scattered all throughout the neural networks. In contrast, the human brain has specific areas for various functions such as the visual cortex and motor cortex, etc., which provides a solid structural organization.

According to Dr. Huizinga, to have a fully connected neural network within the human brain, every neuron would need to be connected to every other neuron, which would necessitate a larger brain than normal in order to handle the vast network. Additionally, every connection would need the essential proteins that are required to build these types of connections. The Ph.D. explains one of his team's goals during their experimentation process—to have an evolutionary algorithm build the network itself, optimizing for not only performance, but also for a reduction in length, and number, of the connections within the network.

As a result, these artificial neural networks approach a more modular state and begin to functionally specialize. Huizinga explains that this structured specialization allows for higher efficiency in learning and thus more closely approximates a human's neural network organization. In regard to these applications as they are applied to robotics, Dr. Huizinga hopes to push neural networks to create a more advanced hierarchical structure for improved robotic functionality.

The AI Ph.D. provides an interesting comparative discussion on the similarities and differences between human brain processing and AI network processing, including how each response to external stimuli, processes the information and reacts to it. As Huizinga explains, when humans need to react quickly to external stimuli or an event or action, the human brain often accesses learned information from past events, which allows for more immediate, and appropriate, responses.

Doctor Huizinga states that the future of AI will be heavily based in teaching the AI to develop its own structures and essentially acquire the information and processes it needs to teach itself how to learn. Further, Huizinga stresses the importance of using AI to influence and train networks to exercise caution before making decisions, as human brains do, before rushing confidently into an action. This subtle training will allow for greater safety as autonomous functional products and robotic entities become mainstream.