,Haga clic para modificar el estilo de ttulo del patrn,Haga clic para modificar el estilo de texto del patrn,Segundo nivel,Tercer nivel,Cuarto nivel,Quinto nivel,*,*,Haga clic para modificar el estilo de ttulo del patrn,Haga clic para modificar el estilo de texto del patrn,Segundo nivel,Tercer nivel,Cuarto nivel,Quinto nivel,*,*,Haga clic para modificar el estilo de ttulo del patrn,Haga clic para modificar el estilo de texto del patrn,Segundo nivel,Tercer nivel,Cuarto nivel,Quinto nivel,*,*,Haga clic para modificar el estilo de ttulo del patrn,Haga clic para modificar el estilo de texto del patrn,Segundo nivel,Tercer nivel,Cuarto nivel,Quinto nivel,*,*,Haga clic para modificar el estilo de ttulo del patrn,Haga clic para modificar el estilo de texto del patrn,Segundo nivel,Tercer nivel,Cuarto nivel,Quinto nivel,*,*,Haga clic para modificar el estilo de ttulo del patrn,Haga clic para modificar el estilo de texto del patrn,Segundo nivel,Tercer nivel,Cuarto nivel,Quinto nivel,*,*,December 8-11,2021,AGI 2021 Oxford,United Kingdom,Departament de Sistemes Informtics i Computaci,Universitat Politcnica de Valncia,Spain,Javier Insa Cabrera,Jos Hernndez Orallo,Jos Luis Benacloch Ayuso,On measuring social intelligence:experiments on competition and cooperation,Outline,Introduction,Universal tests and social intelligence,Evaluation:,Competition and Cooperation,Discussion,Introduction,3,/15,Project:,anYnt,(Anytime Universal Intelligence),/,Any,kind of system(biological,non-biological,human).,Any,system now or in the future.,Any,moment in its development(child,adult).,Any,degree of intelligence.,Any,speed.,Evaluation,can be stopped at,any,time.,Can we construct a,universal,intelligence test?,Introduction,4,/15,Intelligence as a cognitive ability,:,General Intelligence:Capacity to perform well in any kind of environment.,Social Intelligence:Ability to perform well in an environment interacting with other agents.,Objectives,:,Create a,universal test,to,evaluate general intelligence,.,The test must be defined in a formal,way.,5,/15,Universal Intelligence(Legg and Hutter 2007).,=performance over a universal distribution of environments.,An intelligence test can be seen as a definition of intelligence.,Turing Test enhanced with compression(Dowe and Hajek 1997),Intelligence tests based on Kolmogorov Complexity(Hernandez-Orallo 1998),But a definition of intelligence does not ensure an intelligence test.,Anytime Intelligence Test(Hernandez-Orallo and Dowe 2021).,An environment class (Hernandez-Orallo 2021)(AGI-2021).,r,i,o,i,a,i,Universal tests and social intelligence,6,/15,Anytime Intelligence Test(Hernandez-Orallo and Dowe 2021).,An interactive setting following(Legg and Hutter 2007)which addresses:,Issues about the difficulty of environments.,The definition of discriminative environments.,Finite samples and(practical)finite interactions.,Time(speed)of agents and environments.,Reward aggregation,convergence issues.,Anytime and adaptive application.,Universal tests and social intelligence,7,/15,An environment class (Hernandez-Orallo 2021)(AGI-2021).,Spaces are defined as fully connected graphs.,Actions are the arrows in the graphs.,Observations are the contents of each edge/cell in the graph.,Agents can perform actions inside the space.,Rewards:Two special agents Good()and Evil(),which are responsible for the rewards.,Universal tests and social intelligence,8,/15,Evaluate general intelligence of different systems.,Experiments concluded that the test prototype is not universal(Insa-Cabrera et al.2021)(AGI-2021).,Environments rarely contain social behaviour.Environment distributions should be reconsidered:Darwin-Wallace distribution(Hernandez-Orallo et al.2021)(AGI-2021).,Goal:modify the setting to include some social behaviour.,Test whether social behaviour better discriminates between humans and machines.,Introduce simple agents in the environments.,Examine the impact of competitive and cooperative scenarios on the performance of the agents.,Universal tests and social intelligence,9,/15,Reinforcement Learning algorithms:,Q-learning,SARSA,QV-learning,Simple,algorithm,Random,Evaluation:Competition and Cooperation,10,/15,Competition:,All the agents compete for,rewards,.,Evaluation:Competition and Cooperation,11,/15,Cooperation:,The agents receive the average of obtained rewards.,Evaluation:Competition and Cooperation,12,/15,Teams:,Two teams(Qlearning,vs SARSA,)compete for rewards.,Competition and cooperation.,Evaluation:Competition and Cooperation,13,/15,Complexity(Lempel-Ziv approximation):,The complexity of the environment(solely)barely affects the results.,The complexity of other agents makes the environment more difficult.,Evaluation:Competition and Cooperation,14,/15,The inclusion of other agents(even random)make other agents perform worse.,RL algorithms increase their cost matrix.,Algorithms should learn to deal with noise.,Complexity increases with the inclusion of social behaviour.,The complexity of the environment is more