What is intelligence? How do we, as people, know what we know, and how is it stored in our brain? How can we develop machines to echo the structure and functioning of the mammalian brain to further developments in science, business operations, uses of technology, etc.? Why is it even important to develop machines in such a fashion?
Machine learning (ML) is a branch of artificial intelligence (AI) in which data and algorithms work to imitate the way humans learn. Accuracy is tested to see how close machines can function to get to a desired result.
The difference between machines and the human brain is perception. The human brain learns based on anatomy, behavior, and neurophysiology while machines learn from simulation, computation, and algorithms. A biologically inspired machine learning model is trying to combine the two. Neuroscientists have developed a model where they create better brain models which lead to better simulations creating better new neuroscience hypotheses and new experiments which lead to new principles to be used in biologically feasible computing.
Recently, data and neuroscientists alike are combining forces to further the development ML and AI. Research has been focused on the neocortex which is the outer layer of the cerebrum that if stretched out flat would be the size of a dinner napkin. This layer of the brain in mammals is responsible for sensory perception, cognition, generation of motor commands, spatial reasoning, and language (partially). The neocortex is relevant to machine learning based on the interaction of pyramidal neurons which was modeled by the hierarchical temporal memory (HTM) model. HTM at the core is responsible for learning algorithms that can store, learn, infer and recall high-order sequences as well as learn time-based patterns in unlabeled data on a continuous basis. HTM is also robust as it can learn multiple patterns simultaneously. This is well suited for predictions, anomaly detection classification and ultimately sensorimotor applications. Even though the HTM model can be very complex it is one of the ways being modeled after the naturally occurring neural network observed in the brain.
Currently at Massachusetts Institute of Technology (MIT), researchers are shedding light on how the brain processes language. In the past few years, artificial intelligence (AI) was used to predict simple function like the next word someone may use when imputing a Google search. Then AI was further developed to understand the meaning of language which moved away from simple pattern recognition. Currently at MIT, researchers are trying to further develop AI to resemble the language-processing centers in the human brain. Nancy Kanwisher, a member of the MIT’s McGovern Institute for Brain Research and Center for Brains, Minds, and Machines (CBMM), states, “The better the model is at predicting the next word, the more closely it fits the human brain.” Their research continues to confirm simply how complex the human brain is. Again, their model reflects that of the deep neural networks. It contains computational “nodes” that form connections of varying strength and layers that pass information between each other in prescribed ways. In the past, scientists have structured models after the primate visual which has helped AI models in a visual capacity of object recognition. Currently, the team is taking a similar approach taking language-processing models and training them with different language tasks. Each system was compared to the activity in the human brain derived from human datasets that included functional magnetic resonance (fMRI) data and intracranial electrocorticographic measurements taken in people undergoing brain surgery for epilepsy. Evident to say, there was a high correlation between the best performing models and those that resembled the brain.
Now, how could this be important to your business? Think about if you could predict the clients that might “no show” based on previous behavior. This would prevent you from wasting company resources and time. Or what if you could predict and reach the customers that would be most interested in the products that you sell. This would again save resources and time. Companies like Google, Facebook, Instagram, and LinkedIn (just to name a few) are trying to perfect their biologically inspired computing algorithms to get your ads to the right consumers. But there are other methods that data can be used to predict various business questions such as those above. Just a few methods are shown below:
Figure 1- Decision Trees with R: The figure below shows a decision tree generated by R (a programming language for statistical computing) used to determine car seat sales. In the dataset, we took several variables which included competitor price, income, advertising, regional population, price, shelving location, age, education level at location of store, if the store was located at an urban location, and if the store was in the US. Using these factors, the data was trained and tested to see what factors where the largest contributors to car seat sales. In the decision tree below, we can see how different factors contribute to sales of car seats. With this type of predictive modeling, it would allow a customer to tailor their business operations to make their practices most effective and optimal. The tree was also “pruned” which improved the confidence and eliminated some prediction errors. Along with the visual “tree”, other calculated factors were given such as accuracy and precision (yes, the two mean different things in statistics) that help us confidently predict solutions.
Figure 2 – Decision Trees with Python: Python is another high-level, general-purpose programming language that can run decision trees. Taking from another data set, this data set monitored the purchases of a customer based on age, visits to a particular store, and how many items were purchased. Again, the model was tested, and trained and the results are displayed in the tree below. This tree is different than the previous as it shows key factors as you move down the tree.
Decisions trees are one of the most popular algorithms to run in machine learning. Random forests build off the model of the decision trees and it builds different samples and takes the majority vote for classification and average. However, there are several models used to predict outcomes of various dilemmas.
To conclude, machine learning and artificial intelligence are benefiting by developments in neuroscience as we aspire to train machines to think like people. Additionally, people and businesses are also benefiting as we continue to develop algorithms to solve many unanswered questions. This is just the beginning of how neuroscience and artificial intelligence will combine forces to improve intelligence and functioning.
Bibliography
Anne Trafton | MIT News Office. (n.d.). Artificial Intelligence Sheds light on how the brain processes language. MIT News | Massachusetts Institute of Technology. Retrieved September 15, 2022, from https://news.mit.edu/2021/artificial-intelligence-brain-language-1025
HAWKINS, J. E. F. F. (2022). Thousand brains: A new theory of intelligence. BASIC BOOKS.
Malik, M. (2018, May 8). Machine learning of human brain. Medium. Retrieved September 15, 2022, from https://towardsdatascience.com/machine-learning-of-human-brain-739ab0419612
TEDxTalks. “Machine Learning + Neuroscience = Biologically Feasible Computing | Benjamin Migliori | Tedxsandiego.” YouTube, YouTube, 3 Jan. 2018, https://www.youtube.com/watch?v=iPdKMs9cEAs.