Software Agent has wide variety of application in different fields like robotic teams, distributed control, collaborative decision support systems, data mining and learning systems. Agents possess properties like autonomy, proactiveness, responsivity and adaptivity that is suitable for the learning system use for educational purposes. The agent took the input from the user of the system agent stored and process the input of the learner, the result of the system will be returned to the learners. The physical interaction of the user with the agent is through voice and visual data captured from the environment or traditional textual based. For learning, in general communication and participating is major aspect which is cater by the agent system. The agent system designed should cope with this aspect. This is very important to consider key factors of the agent's environment. The learning system performance will be enhanced if the dynamics of the learning are transformed better. The main problem for the agent is to get feedback of agent and learn from its environment to return coherent result to the learner.
In students are taught through program called design-a-plant.in which students are interact with the program to learn the characteristics of the plant and its behaviour to different environments. Keeping in view of the relation of the physical characteristics of the plant, environmental choices are picked, and student are feedbacked by the program at the same time. At the end all students were given retention test, a problem-solving transfer test, and a program-rating sheet to evaluate. In  effect of learning is discussed in agent based multimedia environments. The author  suggests that agent with multimedia support environment is very helpful for the student learning process. For verification of the result author  carried out some tests. Accordingly four studies are performed by author. In the first study, students are divided into four groups. Each group was taught with different ways. The first group was explained with the help of narrated explanation, the second group was explained with voice, the third group was explained on-screen text with visuals and fourth group was explained with the help of on-screen text only. The students performed better on tests when material was presented as speech. According to the hypothesis relevant visual material may have better impact. The presence and interaction of the agent motivate and increase the student confidence. Those students who listen the agent verbal instructions have the ability to solve the new problems in better way. According to third study meaningful study was promoted by interactive session between the agent and students. According to the fourth study multimedia environment presence with the agent verbal instruction.
2.2 Adaptive Learning 
For the learning agent system it should be dynamic, able to adapt new environment and knows the individual needs and accordingly respond to the learner. In  concept of adaptive learning is describes as the agent will adjust to the new environment and need of the learner. The agent should dynamically adapt the domain and individual behavior. The agent based learning system should be adaptive, student centric, able to learn and dynamic in nature. The adaptive agent will be able to pick one of available learning style to meet the requirements of the learner and crop the objects to best fit in a given scenario. Agent should study the environment and tailored itself to new environment. The adaptive agent support personalized learning strategy and different ability level of learner. Earlier agent system separates the pedagogic foundation to the agent. The current system focused on training, group work and human resource requirements. In group work agent only monitor the group progress and help in communication among group members. In  proposed system learning styles are changed during learning process. The agents are decomposed according to their functions ad processes. The agents are divided into five types. The student agent is responsible to take student information, preference, requirement and it communicate with the learner. This information is then passed to the learner object agent who will subsequently send the relevant learning object to the evaluation agent. The evaluation agent will have access to the record agent. The evaluation of learning objects with student records is performed by the evaluation agent. The evaluation agent checks the suitability and relevancy of the learning object for the students. if the leaning objects are suitable and relevant then this will send to the student agent and record agent is update. The learning objects agent manages the learning objects and fetches relevant information. The modelling agent create model of student to their skill. In  author did not discuss the detailed architecture and implementation details. There is lack of discussion about the efficiency and performance.
2.3. Decoupled Agent Learning 
In agent based design it is very important to consider the information access mechanism, whether an agent is single or it is multi agent environment and how it will communicate with other agents and how much information it can access from others. If an agent can't access the other agent utility function then it will be called as decoupled learning. The current agent system lack very important consideration of capturing domain knowledge of environment. This is also important for the scalability and dynamic adaptability.
2.4 Merging of Learning Theories 
Here five coherent goals of MAL research area. For multi-agent system development these agenda points have clear motivation and success criteria. They can be described as computational( it views learning algorithms as an iterative way to compute properties of the game), descriptive ( how agent learn in context of other agents), normative (determine which set of learning rules are equilibrium with the others), prescriptive -cooperative ( determine how agent should learn but there is desire to decentralize the control of a system operating in a dynamic environment like in a team game) and final agenda is prescriptive-non cooperative ( agents who do may have learning of their own).
In MAS each agent has two mode of operations a) Learning mode b) meta learning mode. In learning mode agent learn from its data source and extract local style that represent the knowledge model of the domain problem. Each agent uses classical learning algorithm like C4.5 to obtain the local theory. when an agent operate in meta learning mode in the system it produces a global theory. Each agent queries the global theory and executes the fusion process on global and local theories. Agent merging its knowledge base into global knowledge base. In  paper agent mines the data from local source and then merges with the global knowledge by using knowledge fusion techniques.
2. 5 Possible and Impossible in learning model 
A multiagent interactive learning system is very complex system, where an agent A communicates with the agent B and learns its behaviour so as agent B. The behaviour of the agent changes constantly, there is need of dynamic adoption of new changes. This feedback loop scenario will become more complex as the number of agent will increase. The behaviour of the agent is depends on its learning ability. The agent rationality in game is used which states as the study of the opponents behaviour. This  discussed about the issues which are possible and impossible in multi-agent learning. There are two types of learning models are generally followed i.e. model based learning and model free learning. In model based learning opponents history is maintained, on the basis of that opponents next strategy is to be predict. This includes also the pattern recognition and forecast the next moves. In model free learning agents are not learning to predict.
2.6. Mediated colearner
The learning agent performs some step in some sequence by applying two functions classifier training and measuring of the resulting qualifier quality. In the proposed framework integration operation may perform on knowledge came from different sources. In computer based learning, student computer effective interaction is key to successful implementation, computer acts as tutor. The intrinsically social nature of agent urge the agent to learn with the human learner this is called as computer mediated learner. Interactive learning is very effective it can process and evaluate the user information and this feedback of social and communicative characteristics beneficial for human learners. This can be effectively implemented through mediated colearner character.
Survey focused on the in the field of educational learning and somewhat game theory. Different aspects are discussed to focus these problems. Many systems are developed for research purposes.