dell'Informazione, Universit degli Studi di Parma, pp. 37-40. Experiments with Helping Age

Published in D. D'Aloisi, C. Ferrari, A. Poggi (eds.), Atti del Gruppo di Lavoro su Intelligenza Artificiale Distribuita, IV Convegno dell'Associazione Italiana per l'Intelligenza Artificiale (AI*IA). Dip.to di Ingegneria dell'Informazione, Università degli Studi di Parma, pp. 37-40.Experiments with Helping AgentsAmedeo Cesta, Maria Miceli and Paola RizzoIP-CNRNational Research Council of ItalyViale Marx 15, I-00137Rome, Italy{amedeo|maria|paola}@r.itAbstract - In this work, we explore a simple domain suitable to test experimentally some aspects of our previous analysis about helping behavior. A simulative scenario is described, which is populated by simple agents who need to look for food and eat in order to survive, and can depend on, and interact with, each other. Two types of agents have been defined: "lonely" agents, who just ignore one another, and "social" ones, who, in case of danger (low energetic level), look for help, while in normal conditions they give help to needy agents. In the paper some preliminary results about an interesting interaction between type of agent (lonely vs. social) and food choice function (random vs. nearest), are described and discussed.1. A Simulative Approach to the Study of Helping BehaviorWe are interested in exploring a computationally simple domain suitable to test experimentally some of our analysis about the helping behavior presented in [MIC94, RIZ94]. Therefore we constructed a simulative scenario consisting of a two-dimensional grid where some food is randomly located; this world is populated by simple agents that need to look for food and eat in order to survive, and that can interact with one another. The dependence relationships among agents consist in the differences among their powers, which dinamically change as a side-effect of their actions. Currently our agents have a set of built-in characteristics which make the helping behavior possible without resorting to a complex belief system; anyway, in the future we are interested in augmenting the agents' capabilites, in order to make their behavior more complex.In the following the agent architecture is explained and some preliminary results are illustrated.1. 1. Agent ArchitectureThe agent architecture is quite simply composed of a visual sensor, a set of effectors, a goal generator, and a planning module. The sensor lets the agent perceive food and the other agents within a limited sensorial area. The goal generator chooses a goal to pursue based on the sensorial information and the agent's internal state; the latter is related to the agent's energetic level, which ranges with integer values from 0 to 100, and has a lower threshold at 20 and an upper threshold at 60. Actually the internal states are symbolic labels attached to the various intervals of the energetic values. The relationship between the energetic levels and the internal states can be represented as follows:06010020Lower Threshold Upper ThresholdEnergetic LevelInternal State Danger Hunger Normality Figure 1. Relationship between Energetic Level and Internal StateThe planner is charged with the tasks of producing a plan suitable to pursue the agent's goal, and of controlling the successful execution of the plan. At present, the planning module limits itself to choose the right plan from a set of pre-established ones. The plans currently available to reach the agent's goals are the following:Goal Find food Look for help Give helpPlan look for foodchoose foodgo toward foodeat food signal needwait for receiving foodeat foodlook for foodchoose foodgo toward foodtake foodgo to recipientgive food to recipientFigure 2 - Plans descriptionFinally, the effectors can execute the following elementary actions: moving (one location at a time), taking, giving, and eating (one food-element at a time), and signalling a needy state to the other agents (by changing one's appearance). Each action affects the agent's internal state by lowering the energetic level in a specified amount, except for the action of eating which increases it; "being still" (which occurs for example when an agent waits for being helped), though it does not imply the operation of any effector, is also considered to be an action because it decreases the energetic level.Two types of agents have been defined:•"lonely" agents, that just ignore one another, so there is no interaction among them; their goal is always to individually find food;•"social" agents, that have different goals when they find themselves in particular conditions; more precisely, in case of dangerous internal state, their goal generator activates the goal of looking for help, while in case of normal state, if there are any visible needy agents, the goal of giving help is activated; otherwise, they go on looking for food.The relationships among internal states, type of agents and goals is summarized in the following diagram:Type of AgentInternal StateLonely Social DangerHunger Find foodLook for helpFind food Normal(if needy agents:)Give help(if no needy agents:)Find foodFigure 3 - Relationships among Internal State, Type of Agent and GoalThe simulation is implemented in Common Lisp and uses the MICE testbed [MON90], which allows to create bidimensional worlds populated by user-defined agents. Our world is a 15 x 15 grid that contains 60 food units and 30 agents, all randomly located. The food units keep constant till the end of the simulation by randomly reappearing on the grid each time one or more agents eat some food. The initial energetic value of each agent is50; their visual sensor has a 3 x 3 range (which means that each agent can perceive the world within 3 grid-elements from himself in every direction); they are allowed to move only in vertical or horizontal direction; so their diagonal movement is a combination of zig-zag steps. There is no collision problem, in the sense that two or more agents can occupy or pass through the same grid-element; each movement reduces the agents' energy by 2points, while all the other actions reduce it by 1 point; eating increases the energy by an amount that has been varied from 10 to 40 points depending on each simulation. The action of choosing a food unit among the set of perceived ones is performed in a slightly different way depending on the pursued goal: when executing the plan for finding food, an agent simply selects the food unit nearest to itself, while in case of giving help an agent selects the food unit that is nearest wrt both itself and the recipient.1. 2. Some preliminary resultsIn order to test the effect of helping behavior on agents, we realized a set of preliminary simulations by varying the type of agent (social vs. lonely) and the food choice function (random vs. nearest). Some selected results,which have been obtained with suitable food energetic values (i.e. the values that increase agents' energy each time they eat some food) are presented in the following figures, where the percentage of alive agents is plotted against time; each point represents the mean value across 30 simulations.20406080100% o f A l i v e A g e n t st 50t 100t 150t 200t 250t 300t 350t 400t 450t 500TimeInteraction Line Plot for % of Alive Agents Effect: Type of Agent * Time Figure 4 - % of Alive Agents as a function of timeIn the first figure, it is evident how the social condition dramatically increases the percentage of surviving agents compared with the non social condition. Furthermore, the high percentage of alive agents in the social condition remains rather stable since the start, whereas the percentage of alive lonely agents impressively decreases quite early and seems to become stable later on.In figure 5, three interesting differences in comparison with figure 4 emerge: the first one concerns the percentage of survived lonely agents, which is much higher and seems to be stabler than in figure 4; the second difference is about the percentage of survived social agents, that decreases wrt figure 4; finally, the strong advantage of social agents over lonely agents which can be observed in figure 4 is not kept in figure 5 but, on the contrary, it is slightly reversed.From these results, there seems to be a very interesting interaction between the type of agent and the food choice function variables. When choosing the nearest food unit, lonely agents find themselves in strong competitionover limited food resources; in other words, agents located near one another probably choose to go toward and try to eat the same food unit, therefore wasting energy and decreasing the probability of their survival because only one agent will succeed in eating the chosen food. When choosing a random food unit, there is less competition over the food resources, so agents do not waste energy by trying to eat the same food units, and consequently the life duration is longer. On the other hand, social agents seem to suffer from the random food choice, while they have a strong advantage over lonely agents when choosing the nearest food. The first result could be explained in the following way: when a social agent committed to help a needy agent chooses a food unit randomly, it wastes an amount of energy which is greater than that wasted when selecting the nearest food unit. So it is more likely to pass from a normal state to a hungry state, hence decreasing the probability that a needy agent be helped and survive. The second result could have two reasons: first of all, the abovementioned advantage given to helping agents by the choice of the nearest food unit; secondly, in case of danger, needy agents do not move until they die or receive some food from another agent, thus decreasing the number of agents competing over the food resources.And since each agent can become a needy one in given circumstances, helping behavior turns into a powerful strategy for increasing the probability of survival of the entire social network.20406080100% o f A l i v e A g e n t st 50t 100t 150t 200t 250t 300t 350t 400t 450t 500TimeInteraction Line Plot for % of Alive Agents Effect: Type of Agent * Time Figure 5 - % of Alive Agents with Random Food ChoiceIt is important to stress that these are preliminary results obtained with particular values of the simulation parameters; nonetheless, they look interesting and encourage us to pursue the goal of exploring our simulated world, which seems to be rather sensible to many variables that can be experimentally investigated.References[MIC94]Miceli, M., Cesta, A., Rizzo, P., Autonomous Help in Distributed Work Environments, to appear inProceedings of ECCE7, European Conference on Cognitive Ergonomics , Bonn, Germany, September 5-8, 1994.[MON90] Montgomery, T. A., Durfee, E. H., Using MICE to Study Intelligent Dynamic Coordination.Proceedings of IEEE Conference on Tools for AI , November 1990.[RIZ94]Rizzo, P., Cesta, A., Miceli, M., Basic Ingredients for Modeling Help-Giving in Multi-Agent Systems, to appear in Working Notes of the 2nd Working Conference on Cooperative Knowledge Based Systems (CKBS94), Keele, UK, June 15-17, 1994.。