Make machines think

Cognitive computing is all about making machines think.

Intelligence is a human thing and the only way for a machine to become smarter is to be able to use and process knowledge in the same way.

Providing help, assisting people, find solutions to problems requires that a given situation is understood. And this is complex since many cognitive processes and psychological functions are involved. The challenge is to put these human things inside something immaterial, inside a machine so that this machine can act like a human.

As you might imagine, studying what’s happening in a person’s mind is not always the easiest thing to do. It is what cognitive psychologists are trying to understand. Cognitive psychology is the area of psychology that focuses on internal mental processes. Such processes include thinking, decision-making, problem-solving, language, attention, and memory. The core focus of cognitive psychology is on how people acquire, store, and process information.

To study the human mind, cognitive psychologists have developed different models to represent how thinking works. One of the most popular of these is the information-processing approach. In this approach, the mind is thought of much like a computer. Thoughts and memories are broken down into smaller processing units performing a specific task. From a psychological point of view, the challenge is to propose a functional framework, a kind of cognitive architecture, explaining how these mental processes are linked together, how information is exchanged, how each processing unit is working. For this framework to be valid, it should have a biological analog counterpart, and this is the reason why the main hypothesis that cognitive psychologists took was that those processing units have a real existence inside the human brain.

As information enters the mind through the senses, it is then manipulated by the brain which then determines what to do with the information. Some information triggers an immediate response. Other units of information are transferred into long-term memory for future use. Many key cognitive processes such as reasoning, decision-making and problem-solving, are using information available within the long-term memory. In addition, most psychological functions are also linked with the long-term memory, such as sentiments, emotions or some of our needs. Knowing how the long-term memory is storing information is one of the main concerns. If we know how the information is structured inside the brain, we will then gain confidence of how related cognitive processes are accessing and processing this knowledge. We know that there is somewhere in the brain a processing unit able to store information but where is it and how is it structured?

A monitoring method, called electroencephalography (EEG), can measure and record the electrical activity of the brain. While performing some cognitive tasks, this monitoring method can highlight which area of the brain is activated so that a clear link can be established between the nature of the cognitive task that is performed and the areas of the brain that are involved. Those areas of the brain can then be seen as a processing unit for this cognitive process. With this monitoring method, some of these processing units have been found while others remained unlocated. But why?

The main issue with EEG is that it poorly measures neural activity that occurs below the upper layers of the brain (the cortex). Because of this, some key cognitive processing units could not be found meaning that, for those cognitive processes, the processing unit must be located somewhere deeper inside the brain. One of these unit is the one that is storing information and our semantic knowledge.

The brain is the most complex organ playing a key role in attention, perception, awareness, thought, memory, language, and consciousness. It is made of several components where each have a specific function. One of the major components of the brain is called the Cerebrum or Telencephalon. It is the largest part of the brain.

Neurons are usually considered the most important cells in the Cerebrum. The property that makes neurons unique is their ability to send signals to specific target cells over long distances. They send these signals by means of an axon, which is a thin protoplasmic fiber that extends from the cell body and projects, usually with numerous branches, to other areas, sometimes nearby, sometimes in distant parts of the brain or body. Axons transmit signals to other neurons by means of specialized junctions called synapses. A single axon may make as many as several thousand synaptic connections with other cells. When an action potential, traveling along an axon, arrives at a synapse, it causes a chemical called a neurotransmitter to be released. The neurotransmitter binds to receptor molecules in the membrane of the target cell. Synapses are the key functional elements of the brain. The essential function of the brain is cell-to-cell communication, and synapses are the points at which communication occurs.

Multiple layers of neurons are structuring the surface of the Cerebrum, forming the cerebral cortex which is divided into multiple functional areas (hubs). Mainly made of neuronal cell bodies, dendrites, unmyelinated axons and synapses, known as the grey matter, each hub of the cerebral cortex can communicate a message to another hub. Since there is a longer distance to cover to convey the message between those hubs, neurons with longer axons are involved. Those neurons have a fatty substance, called myelin, surrounding their axon, acting as an insulator, allowing the electrical signal to jump rather than coursing along the axon which is increasing the speed of transmission. Since myelin is a fatty substance with a relatively light appearance, the inner part of the Cerebrum is called the white matter. Since the white matter is only present under the surface of the cerebral cortex, the processing units that could not have been located elsewhere at the surface of the Cerebrum should be located somewhere inside this white matter, deeper in the brain.

One of the main hub of the Cerebrum is called the Fusiform Gyrus. In a recent study of the State Key Laboratory of Cognitive Neuroscience and Learning and IDG/McGovern Institute for Brain Research in Beijing, Yan Cheng and his team published in January 2020 their ?ndings, showing that this hub (the Fusiform Gyrus) worked in concert with nine other hubs in the semantic memory network for general semantic processing. They revealed the existence of a neuroanatomical white matter structure able to process semantic knowledge which means that our long-term memory is deeper inside our brain, contained within the white matter of our Cerebrum. It is made of neurons, connected to each other, structured in a way so that it is possible to process semantic knowledge, receiving signals from several hubs to acquire and process new information.

As cognitive psychologists, one of the key processing unit to which all others are linked is the one that is storing knowledge. No matter which kind of cognitive architecture is defined, it has to consider core mental activities, cognitive processes and psychological functions and link all those processing units together to propose a cognitive framework. Within such a framework, the long-term memory is a key processing unit to which all other processes are linked to. This is why it is essential to define how the information is structured within this processing unit to ensure that all linked processing units will operate. Recent findings highlighted that there is a neurobiological analog structure within the white matter that is dedicated to store semantic knowledge so that the structure of how information is stored within the long-term memory is now confirmed. This finding is key to be able to define the right data structure to put inside the machine and make the machine think.

To make it easier to understand all this, let us take a metaphor. It’s like making a soup. To make a soup, what do you need ? Ingredients, such as tomatoes, celery, salt, pepper, olive oil. All these ingredients once acquired needs to be prepared and mixed together in a bowl. All those ingredients are coming from different areas or places and are ? meeting ? each other within the bowl to prepare the soup.

We have different processing units (the ingredients of the soup), either dedicated to mental activities, cognitive processes or psychological functions and all these units are getting or providing information to our long term memory (the bowl) so that all the available information can be used.

Marvin Minsky, Cognitive Scientist, father of AI, founder and Director of the MIT AI Lab, said “ If you understand something in only one way, then you don’t really understand it at all. The secret of what anything means to us depends on how we’ve connected it to all other things we know. Well-connected representations let you turn ideas around in your mind, to envision things from many perspectives until you find one that works for you. And that’s what we mean by thinking! ”.

Here again, the main idea of Marvin Minsky is that representations (our knowledge, information) are connected, all available at a given moment (like ingredients in a bowl). Intelligence is a human thing. It can be defined by our ability to reason, make decisions and solve problems. These cognitive processes are linked to psychological functions such as emotions, sentiments or needs (the ingredients). These mental states are providing additional context related key insights which once added to other available inputs will create a conglomerate of information (the bowl), providing us with a mental representation of the situation, enabling our awareness and adapting our behaviors and actions.

At MacAnima, our intent is to find a recipe to put these human things inside something immaterial, inside a machine so that this machine is able to act like a human. Knowing this recipe will empower us to create and explain how to create cognitive computing solutions to help companies during their digital transformation journey. It is with this intent that MacAnima was founded.

All began in 1994, as Beatrice & Mark were studying cognitive psychology in Paris. Both, very interested in computer software and programming languages, founded an association called Beamak (for Beatrice & Mark).

During two years, Beamak created about 20 different free software which were distributed by the press. In 1996, Beamak received an award for the first ever made database which could be fed with unstructured information. It was just as simple as talking to a friend to input or search information. This software featured an innovative search algorithm made with fuzzy logic.

As Cognitive Psychologists, they both worked for the French National Center for Scientific Research and the Computer Science Laboratory for Artificial Intelligence before moving their career forward within the IT & Consulting industry, working for over 20 years for companies such as Andersen Consulting, Accenture, Cap Gemini or Hewlett Packard Enterprise, managing over 200 projects for the biggest European companies.

In 2016, Beamak became a company aiming to create artificial intelligent software products based on cognition and featuring psychological functions. For over 3 years spent on Research & Development, Beamak was able to specify a cognitive framework linking together mental activities, cognitive processes and psychological functions and adapted this framework so that it can be put inside a machine to create cognitive computing solutions. To do this, multiple innovations have been patented such as the SmartNeuron, a software component processing information in a natural way like chemical neurons do and the VirtualBrain, a specific data structure were the information is stored like it is in our long term memory. To showcase that this is working, a software product called Hector was created. For this work, Beamak received multiple awards (AI Startup Challenge, PwC & Slush in 2018 / Semi-Finalist of the Let’s Go France Trophee) and was shortlisted as one of the top 10 cognitive computing provider in Europe in 2019. Hector’s features were demonstrated to companies such as Amazon, Airbus, Orange, Tessi and PwC.

In 2020, we founded MacAnima with the intent to provide dedicated cognitive computing consulting services & software solutions.

We would be very happy to help you soon on your journey.

Best,

Béatrice & Mark