UDC 001.81:519.68:681.32 Autonomy and Artificial Life Forms for Amusement and Use as Agents VKoichi Murakami (Manuscript received August 24, 1999) This paper describes the attributes that can be given to artificial life forms and intelli- gent agents so that they appear to be alive and posses their own individuality. Also, this paper describes possible architectures for realizing these attributes. 1. Introduction ficial intelligence for the purpose of understand- We started to develop forms of artificial life ing human beings and creating human-like robots. in 1990, and in 1995 we developed an entertain- The study then branched into several areas, for ment software called “TEO – Another Earth” example, robot engineering, image and voice rec- which can communicate with human beings. ognition, and symbol processing for planning and Then, in 1999 based on this experience, we con- learning. Now, however, these areas are being structed and released an agent on the Internet. integrated into a united effort to study human This paper describes our concept of artificial beings. life, the image of the “living organisms” we intend Agents may possibly change people’s associ- to create, and some architectures for artificial life.1) ations with computers. If a person asks an agent Then, this paper describes various systems we on a computer screen to do something, the agent have developed, in particular “TEO – Another will carry it out to the best of its capability. Agents Earth,” which we developed in collaboration with will come to be regarded as capable secretaries, Mr. Macoto Tezka, the producer. faithful stewards, obedient servants, and perhaps even close friends. 2. From agents to artificial life However, to create these autonomous enti- An agent is a human expert who helps an- ties, we will have to better understand the difficult other person do something upon request, and an region covered by the theory of communication, interface agent is a computer program that oper- where we have not yet found standard solutions ates as an expert.2) Many researchers are now even to questions regarding communication be- concentrating their energy on the study of artifi- tween human beings. The questions to be solved cial intelligence and agents. For examples of the include “What makes a person want to communi- work being done in this field, see Ref. 2), in which cate with an entity?” and “What makes a person nearly 20 of the foremost researchers in this field want to continue communicating with an entity?” (including Minschy, the most prominent figure in There are a variety of possible answers – for ex- the study of artificial intelligence) discuss agents ample, the entity is cute, friendly, interesting, or from various points of view. in the process of growing. We hypothesize that Researchers originally began studying arti- people want to communicate with an entity or 158 FUJITSU Sci. Tech. J.,35,2,pp.158-164(December 1999) K. Murakami : Autonomy and Artificial Life Forms for Amusement and Use as Agents agent when they feel that the entity is alive. We to achieve certain goals on a computer. The cannot prove this hypothesis now; however, based autonomous behavior of this mechanism is com- on our implicit knowledge, we can say that hu- pletely different from the random selections of man beings have certain special feelings toward predetermined behaviors that are performed in living organisms. the simulated pet systems which are enjoying re- An artificial life in a computer cannot be gen- cent popularity. The autonomous behavior of an uinely alive, but it can nevertheless appear to be artificial life is dependent on factors in the exter- alive, and it is up to the person who communi- nal environment (e.g., weather, time of day) of the cates with an artificial life to decide whether it user and other living organisms and the artificial appears alive. We therefore have to solve the fun- life’s internal conditions (e.g., degree of hunger, damental question of what makes people feel as emotions, and degree of familiarity with the user). if an artificial life is actually alive. The answer to If we use a conventional programming paradigm this question may come from philosophy, biology, to realize autonomous behavior, we would have to or physiology, or perhaps the answer has already describe all the conditions of an enormous exter- been given by Walt Disney and Osamu Tezuka, nal environment in a list. This, however, is not a who is a famous Japanese cartoonist. That is, practical approach. In fact, robot researchers have perhaps the answer is that animation gives life faced the same problem. to things and characters. After all, we cannot ig- To achieve its goal, a robot must determine nore the fact that many of the cartoon characters its actions by itself in a dynamic environment that they created have been “living” in our minds. includes moving objects such as people and other robots without hitting them. Therefore, the ar- 3. Architecture of artificial life chitectures developed by robot researchers are The first task of our research into the essen- planning systems which react to conditions. tial properties of artificial life was to build an Figure 1 shows one such architecture we have autonomous mechanism that behaves adaptively developed at Fujitsu. These architectures consist Reactive planning Rule base -if (condition) then (behavior) ... Plan tree Condition Verification Behavior [External condition] [Internal condition] Motivation Goal achieved/not achieved Emotion Strength t Time attenuation Stimulus : Information exchange from external environment : Information exchange from internal environment Figure 1 Autonomous architecture. FUJITSU Sci. Tech. J.,35, 2,(December 1999) 159 K. Murakami : Autonomy and Artificial Life Forms for Amusement and Use as Agents of three modules: one for recognizing conditions according to motivations internal to the living in the external world, one for judgment (think- organism and that when a living organism dem- ing) based on this recognition, and one for onstrates this autonomy we are likely to regard executing the actions to be taken. The main fea- the organism as possessing a degree of conscious- ture of these architectures is that the three ness. modules are linked with each other in a loop in However, an autonomous mechanism cannot realtime, so that when a robot takes an action in be created simply by designing an artificial life the external world, it quickly recognizes the ef- that reacts to the external environment; it can only fects of the action and then makes the appropriate be created when an artificial life is given much judgement or judgments. Because of this loop the more profound internal conditions. To achieve this, robot maintains an internal analog of the exter- we developed an emotion module with parame- nal world. ters for physiological states such as hunger and In the judgment module, which is the core of fatigue and feelings such as sorrow and joy (see these architectures, behavior is described as a set Figure 1). These parameters quantitatively of rules for actions. The relationships between change depending on the artificial life’s own ac- actions in this module and how they influence the tions, its recognition of external conditions, and robot’s behavior are not predictable but depend the degree to which it achieves its goals. As a re- on conditions in the external world. In other sult, a change in circumstances can produce a new words, individual actions are only meaningful in goal inside the artificial life. terms of external conditions. Consider, for example, the feelings of like and A robot having such an architecture, howev- dislike and assume that an artificial life “likes” er, is not a living organism because it is always an object. If the object is in front of the artificial given a goal in advance. Let us take the example life, the simple goal of “getting the object” is pro- of a cleaning robot that moves around and col- duced. If the object is at a distance, the goals of lects dust and trash while making way for human “finding,” “reaching,” and “getting” the object are beings, cats, dogs and other obstacles. Such a produced. The artificial life then takes the neces- cleaning robot can be made with an extremely high sary actions to attain the produced goal or goals. level of technology and could make enormous con- While executing these actions, the artificial life tributions to society. However, it will only act for monitors and reacts to changes in the environment the purpose of cleaning and will still be just a because the recognition, judgment, and action machine. modules continue operating in the loop described A living organism can have simultaneous, im- above. If someone brings the target object close portant goals, for example, self-preservation and to the artificial life, it will have achieved its goal reproduction. A living organism might become without the need to move to the object’s original confused when some of these goals conflict with location, so it stops moving. Then, because its goal each other and find which is the most important has been achieved, the artificial life can autono- goal for that situation. It eats when it is hungry, mously take a new action. For example, the looks for a friend when it feels lonely, and sleeps artificial life may be pleased and begin laughing when it is tired. When people see or recognize or dancing or begin to “love” the person who such behavior in a living organism, they try to brought the object and follow that person around. deduce the living organism’s state of mind from Thus, we built an artificial life mechanism which its behavior. The essential points here are that behaves according to its own goals and appears the goals of a living organism are not dictated by as if it has a mind.