Abstract
Model of artificial mind discussed in this and the following two chapters considers critical elements of the mind operation. The question is whether we can propose artificial brain structures in machines that will be able to create the basis of intelligence and consciousness. Wanting to build an artificial brain, we propose what properties it should have, and how it should be organized. The chapter begins with presenting embodiment of the mind as the part of the environment that is under control of the mind. Perceiving and identifying with one's own body depends on observing the body's actions in the environment. The embodiment must communicate with the brain through channels that ensure the perception of the environment. The use of body dynamics facilitates control, planning, and decision making. Conditions that exist in the real world create a framework for proper action and reflect the compatibility of agent's competencies with the environment. In a conscious embodied mind, representations are created and used for actions. Higher level consciousness can be treated as an abstract version of the coordination of perception and action. Conscious states are triggered by externally supplied signals from the environment and by internally generated mental states. Self-consciousness requires distinguishing oneself from the environment. The definition of embodied intelligence adopted in this book is aimed at building an intelligent and conscious machine. The authors have recognized the ability to learn as the most important feature of intelligence, which is why they consider beings that do not learn anything as not intelligent. Machines will not have the same needs as people but must have needs whose fulfillment is a measure of success. Meeting these needs will require physical and mental effort, and the development of useful skills will be associated with the development of intelligence. The agent treats unmet needs as a signal to act. Using the analogy to pain, these signals representing unmet needs will be called the pain signals. Strength of these signals can be measured and compared with each other. Various pain signals not only provide motivation for action but also control the learning process. Finally, they discuss the role of feelings and emotions and their importance in the agent's learning process. In particular, they discuss their role in creation of conscious sensations. They explain the source of feelings as associated with but different than reward or punishment signals. The signals provided by the senses to anticipate reward or punishment are related to the physical properties of the observed objects, which are directly related to feelings. Pleasure is the promise of meeting a real need. Feelings will fuel emotions. They relate emotions to subconscious reactions to what happened. They also discuss why we may need to build emotional machines and how artificial emotions can be created in machines.
TopThe Embodied Mind
The notion of embodied mind was introduced in part I, proving that the embodiment of the mind is essential for the intelligent system to interpret simple sensory impressions and turn them into qualia. Discussing the natural mind, we drew attention to the importance of the notion of an embodied mind in the shaping of speech, and we even quoted the concept of embodied mathematics to explain how mathematical abilities have evolved in human minds. Now we will examine this concept as fundamental to the design of artificial minds and to understanding what qualities can be achieved by an autonomous agent—a robot operating in a natural environment. Moreover, we claim that no intelligence can arise without this embodiment.
The embodied mind is the mind of an autonomous agent who possesses the body. The principles of designing robots using the idea of embodied intelligence were first described by Brooks (1991) and were characterized by several postulates for the development of embodied agents. The first assumption is that agents develop by acquiring experiences in a changing environment in which they act and observe the effects of their actions through their senses. Another important assumption adopted by Brooks is the postulation that there is no need to build an environmental model. Instead, the robot simply can observe the environment in which it works and learns. This approach has revolutionized robotics and has changed the research direction of cognitive systems.
Animals, even those with very simple nervous systems, learn. Eric Kandel of Columbia University studied the learning process in the example of marine gastropods (2007). The sea snail only has about twenty thousand neurons, so it is relatively easy to study its learning process. It was observed that when the snail was gently touched, it withdrew—sensory neurons directly stimulated motor neurons for immediate withdrawal. However, repeated gentle touches caused a gradual habituation to touching, and the withdrawal action was weakened. As a result of learning, the strength of connections between sensory and motor neurons weakened. This is already known to us as the effect of getting used to a stimulus—habituation. However, this was not the case with painful touches. In this case, the sensitivity and strength of the response grew. The snails learned a more radical reaction to an unpleasant stimulus.