Artificial Cognition Architectures

For more than 80 years, science and science fiction have been addressing the need and significant challenges of achieving truly autonomous machines that can act on their own. From the 1927 depiction of Maria in “The Eyes of the State,” to Gort in “The Day the Earth Stood Still,” to more semi-recent versions, HAL 9000 in “2001: A Space Odyssey,” “Terminator,” and Sonny from “I, Robot,” the world has been fascinated, amazed, amused, and terrified at the thought of robots intertwined in our existence. These creative depictions of artificially intelligent “robots” were supposedly capable of thinking, reasoning, learning, and making decisions. Unfortunately, the notion of a machine, run solely by software, no matter how sophisticated and creatively plausible, has continuously been met with significant theoretical, physical, and social challenges. Some would say, “troublesome” at its very core.

Research for the development of credible solutions within the Information Continuum has been a 17 year journey that began in the mid 1990s when the authors of this book were designing new ways to perform data capture, processing, analysis, and dissemination of high volume, high data rate, streams of information (what today would be called a “big data” problem). Hence, data analysis and lack of quality user interaction within that process are not a new problem. Users have continued to be challenged with keeping up with the vast volumes and multiple streams of data that have had to be analyzed. By the time a user had grabbed a time-slice of data, plotted it, and analyzed it, 100s of Gigabytes of data had passed through the system. In order to provide some semblance of rational attack against the onslaught of data we created what could be likened to a virtual window into the system that allowed the analysts to “walk into the middle” of the data and look at it as it flowed through the system. Analysts could reach out and grab a set of data, rotate it through its axes, and perform automated analysis on the data while remaining within the system data flow. This way analysts could intelligently and rapidly hop between portions of data within multiple data streams to gain pattern and association awareness.

The preceding chapters focused upon introducing the characteristics within an Information Continuum and how they relate to a fully autonomous, learning, reasoning system (analogous to a synthetic brain), and how a SELF must possess constructs in its hardware and software to mimic humanistic processes and subsystems . This chapter will focus more upon designing and implementing these humanistic structures by understanding how they must interact and cooperate in order to form a comprehensive learning system. We employ concepts adapted from the domain of cognitive psychology as inputs into the formation of these interactive humanistic structures, sub-structures, and components. In short, Psychology helps us understand how these structures function within the human brain followed by translation efforts to design and implement these dynamic functions within an analogous synthetic brain. Hence, the foundational building blocks likened to “synthetic consciousness”, comprised of cognition, intuition, and other capabilities that humans possess.

At their very heart, memories involve the acquisition, categorization, classification and storage of information. The purpose of memory is to provide the ability to recall information and knowledge as well as events that have happened to us in the past. We base our current understanding of the world around us on what we have learned and stored in the past and we react to that same environment relying on the memories of what has happened before, and what has been learned in the past. Without our memories, day-to-day living is not manageable. It would require continuous abstract thought and continuous reiteration of the most basic functions, analogous to the symptoms of an Alzheimer patient. Without memories, we wouldn’t be able to drive a car, brush our teeth, or perform any of the things we do “without thinking about them.” Through our abilities of conceptual recollection of past memories we are able to reflect, infer, and even communicate with other people.

To develop “Artificial Consciousness” for a SELF requires investigation and understanding of what it means to be conscious. The textbook definition of consciousness is:

From the moment our brain functions at all, we begin the process of learning. In order for a SELF to be an autonomous system, it must also begin and continue the process of learning throughout its existence. The main goals of the mathematical foundations for SELF learning include:

As explained earlier, the ability to reason within a SELF denotes the ability to infer about information, knowledge, observations, and experiences, and affect internal changes that enable it to perform new tasks previously unknown or to perform tasks already learned more efficiently. The act of reasoning, or inferring, allows a SELF to construct or modify representations of experiencing and learning. Reasoning allows a SELF to fill in skeletal or incomplete information or specification (self-assessment). Hence, this chapter is devoted to architectures and frameworks to enable artificial reasoning within a SELF’s cognitive processes that synthesizes human reasoning. First, we will discuss the various stages and forms of human reasoning. The rest of the chapter is devoted to adapting human reasoning concepts into SELF reasoning architectures.

As discussed in previous chapters, the primary SELF software component is the Cognitron. Each Cognitron type provides different cognitive abilities that, together, form a cognitive ecosystem within an ACNF cognitive framework, implementing intra & inter SELF communication and collaboration. The basic Cognitron is a self-contained discrete functional software codelet comprising one or more loosely coupled software services. For a Cognitron to be of a specific archetype (e.g. Reasoner Cognitron), a set of archetype specific services is defined. Additional services can always be added to extend capabilities of a Cognitron archetype. Figure 10.1 lists the core set of services from which a Cognitron’s capabilities can be defined [198].

As we have discussed throughout the book, a SELF is a hardware/software artificial cognitive system designed to mimic human reasoning, learning, and understanding. The first ten chapters have concentrated on the cognitive side of the software architectures and frameworks to accomplish artificial consciousness and artificial human cognitive skills [9, 10]. However, the next few chapters will focus on the pragmatic computer software and hardware architecture upon which the cognitive software functions will operate and be processed. Computer processing units, electronic memory devices, and information networks are requirements in order for the cognitive software to exist, operate, and function.

As with any electronic information processing system in today’s world of hackers, malware, spyware, etc., security is a major component of the overall operational capabilities of a SELF. All information within a SELF must be protected and kept from corruption (whether accidental or intentional). Accidental corruption of information and knowledge within a SELF is handled through continual cross-checking and self-assessment within the ACNF framework. Continuous communications between Cognitrons within the system and constant refresh of memory information keeps information from being arbitrarily modified (loss of bits) and from corruption due to memory failures and catastrophic interference problems discussed earlier. However, these do not protect the system from intentional corruptions and hackers. Since a SELF is intended to be a fully autonomous, self-evolving, self-learning, reasoning artificial entity, any corruption of information across a SELF’s artificial cognitive framework could have devastating effects on a SELF’s learning, reasoning, memory, and cognitive processes analoagously to what occurs in injured humans (e.g. head injury, Alzheimers). Corruption or incorrect modifications to a SELF’s needs, constraints, goals, memories, or algorithms could cause a SELF to act, evolve, remember, or learn, completely incorrectly and/or out of scope for the intentions of a SELF.

So we have given a SELF the ability to think, reason, adapt, and evolve, as well as Metacognitive and Metamemory capabilities to understand its own abilities and limitations; including cyber security within its cognitive framework. The Cognitrons within the system themselves can learn, adapt, and evolve and can communicate with each other, allowing cognitive collaboration and cognitive economy within a SELF. So if we can actually build the complete system, if a SELF becomes a real-time, fully functioning, autonomous, self-actuating, self-analyzing, self-healing, fully reasoning and adapting system, what do we have and what are the ramifications? In Chap.​ 3 we discussed how people from different cultures might respond to a SELF, and the differences between accepting the system when it looks like a machine versus when it looks like a person. We explored the ramifications of giving a SELF basic emotions and emotional memories. How might its memories and actions be influenced by how people react to it? We also discussed how those reactions might influence how a SELF handles being around people. The overall purpose of the book was to begin to describe the capabilities, methodologies, and subsystems that must be in place in order to create a real-time, autonomous, thinking, reasoning system. We hope we have allayed fears that a SELF is going to decide to take over and eliminate the human race, as Hollywood is so fond of portraying. However, we also are not describing a cute, lovable robot, as depicted in the movie “Wall-E.” There are other questions that need to be explored such as how to create versions a SELF at different levels in its evolutions so as not to have to start over again with each SELF we create, i.e. how do we clone a SELF. We also need to explore the advantages and disadvantages of SELF entities communicating with each other.