General purpose architecture for intelligent computer-aided training

What is claimed is:

1. A computerized intelligent training system adaptable for use in the training of a trainee having a current skill level in the performance of at least one of a plurality of training tasks within a specific task environment, where the performance of each task comprises performance of certain procedural steps called actions such that each task has at least one desired action called a correct action and training is accomplished using the system by having the trainee perform simulated task by performing simulated actions on a computer, the computer system comprising:

(a) user interface means for simulating the task environment for which the trainee is being trained and for enabling interaction between the trainee and the system;

(b) domain expert means for performing a simulated task and achieving the correct actions for the task;

(c) training session manager means comprising,

(1) error detection means for error detection, an error being a failure of match between an action taken by a trainee, called a trainee action, with a correct action, whereby error detection is made in a hierarchical manner with relatively higher level errors, including a highest level error, distinguished from relatively lower level errors, and p2 (2) error handling means for providing information concerning actions and errors, said information concerning errors being based on the highest level error detected;

(d) training scenario generator means for designing increasingly complex training tasks based on the current skill level of the trainee and on any weaknesses or deficiencies that the trainee has exhibited in previous trainee actions, where such weaknesses or deficiencies are based on comparing the trainee actions with the correct actions from said domain expert;

(e) trainee model means for accepting from said training session manager information concerning correct actions and errors made as a result of trainee actions and compiling a complete record of the correct actions taken and errors by the trainee and, at the conclusion of each training session, creating a training summary of such correct actions and errors as well as the time taken to complete the session and the type of assistance provided by the system to the trainee; and

(f) blackboard means providing an intermodule interface for communicating between said user interface means, said domain expert means, said training session manager means, said training scenario generator means, and said trainee model means; said blackboard means also providing an intermodule interface for transferring control of the training system from one to another of said domain expert means, said training scenario generator means, said error detection means and said error handling means by use of rules contained within each of said domain expert means, said training scenario generator means, said error detection means and said error handling means.

2. The system as defined in claim 1 wherein said domain expert is capable of performing the task to be trained by using rules effecting correct actions and includes rules identifying typical errors.

3. The system as defined in claim 2 wherein said domain expert comprises production rules.

4. The training system as defined in claim 1 wherein said domain expert generates a plurality of correct actions any of which could lead to accomplishment of a particular training task so that any trainee action which could lead to accomplishment of the particular training task is a correct trainee action.

5. The system as defined in claim 4 wherein said blackboard means accepts facts from said training scenario generator to establish the context of a training scenario.

6. The system as defined in claim 4 further comprising a database containing a range of typical procedural steps describing the training context and problems of graded difficulty from which new training scenarios are built.

7. The training system as defined in claim 1 wherein said training session manager compares the actions by said domain expert and by the trainee for evaluating such trainee actions and provides guidance to the trainee which is appropriate to acquired skill level of the trainee.

8. The system as defined in claim 1 wherein said training scenario generator uses information concerning correct actions and errors made as a result of trainee actions in said trainee model for creating a unique scenario for the trainee whenever a new session begins.

9. The system as defined in claim 1 wherein said training scenario generator examines said trainee model and said database in order to create a unique scenario for the trainee whenever a new training session begins.

10. The system as defined in claim 1 wherein said training scenario generator builds training scenarios of greater difficulty as the trainee demonstrates the acquisition of greater skills.

11. The system as defined in claim 1 wherein said domain expert provides a plurality of error texts that allows said error detection means in said training session manager to write appropriate error messages to the trainee through said user interface.

12. The system as defined in claim 1 wherein said trainee model accepts from said training session manager information comprising trainee history including previous correct actions and errors made as a result of trainee actions, where such history may comprise summaries of such information, records the trainee actions, updates trainee history, and provides such information to said training scenario generator to produce new training scenarios.

13. The system as defined in claim 1 wherein said domain expert includes rules identifying typical errors made by a trainee and trainee error messages relating to said typical errors, whereby, upon performance of a trainee action which includes an error, said training session manager selects from said hierarchy of errors said highest level error and provides an error message for the trainee which is appropriate for the trainee's current skill level.

14. The system of claim 1 wherein said user interface means, said domain expert means, said training session manager means, said training scenario generator means, and said trainee model means all contain control rules in the form of production rules; whereby, control of the system may be transferred from one to another of said user interface means, said domain expert means, said training session manager means, and said training scenario generator means.

15. The system of claim 14, wherein said control rules may cause facts to be posted to the blackboard means by any of said user interface means, domain expert means, training session manager means, training scenario generator means, and said trainee model means.

16. The system of claim 15 wherein said facts may be read by any of said user interface means, domain expert means, training session manager means, training scenario generator means, and trainee model means.

17. The system of claim 16, whereby control of the system is transferred to the one of said user interface means, domain expert means, training session manager means, training scenario generator means, and trainee model means which can fire one of its said control rules from said facts posted to said blackboard means.

18. The system of claim 14 wherein said control rules include at least one standard match rule which changes the step, called the context, of the training task.

19. The system of claim 14 wherein said control rules include at least one input conversion rule which converts the trainee's action into a format required by the system.

20. The system of claim 14 wherein said control rules include at least one clean-up rule which deletes facts from the blackboard that are no longer needed by the system.

21. The system of claim 14 wherein said control rules include at least one backup control rule which reconstructs a previous context.

22. The system of claim 14 wherein said control rules include at least one iteration rule which reconstructs a previous sequence of action contexts.

23. The system of claim 14 wherein said control rules include at least one special-match rule which under appropriate conditions omits certain steps in a training task.

24. The system of claim 14 wherein said control rules include at least one outside-request rule which responds to trainee's request for help or information.

25. The system of claim 14 wherein said control rules include at least one context-switching rule which changes control of the system from one of said user interface means, domain expert means, training session manager means, training scenario generator means, and trainee model means to a different one of said user interface means, domain expert means, training session manager means, training scenario generator means, and trainee model means.

26. A computerized intelligent training system adaptable for use in training persons, called trainees, in the performance of training tasks in a specific task environment, the computer system comprising a plurality of modules, each module comprising a set of production rules, said rules of the various modules acting in concert to implement the system, wherein said modules communicate with each other via an intermodule interface by means of messages sent in accordance with a message passing protocol, wherein all of said modules may write facts to said intermodule interface and all of said modules may read facts from said intermodule interface, and wherein each said message, to comply with said message passing protocol, contains facts indicating which of said modules sent the message and which of said modules is to receive the message.

27. The system of claim 26 wherein control of said system at any given time is performed by control rules contained within a first of said modules and whereby said control rules transfer control of the system to any second of said modules by writing messages to said intermodule interface in accordance with said message passing protocol.

28. The system of claim 27 wherein control of said system is transferred from said first to said second module, said second module being the module having a rule which can be fired by facts written to said intermodule interface by said first module.

29. The computer implemented method of training a trainee having a skill level to perform actions necessary to accomplish the training steps of a training task in a specific task environment, the trainee performing trainee actions at various times in the training, the various actions being represented by rules, the method using a computer having modules comprising sets of rules and including an intermodule interface to which facts may be written by modules and from which facts may be read by other modules, comprising the steps;

(a) writing to the intermodule interface predetermined correct, optional, and typical error actions;

(b) waiting for a trainee action;

(c) comparing the trainee action with the predetermined correct, optional and typical error actions;

(d) continuing to the next training step if the trainee action matches the predetermined correct action or a predetermined optional action;

(e) determining a specific error if trainee action matches a predetermined typical error action;

(f) reporting an error message to trainee appropriate for the trainee's skill level;

(g) recording the specific error for use in both trainee and system evaluation;

(h) performing each step by use of one or more of said modules; and

(i) transferring control of the process from a first to a second of said modules by use of rules contained in any said first module, whereby said transferring of control is accomplished by writing facts to said intermodule interface by said first module and reading therefrom by any said second module.

Description Submit all comments and votes

FIELD OF THE INVENTION

The present invention relates generally to the use of artificial intelligence for providing computer-aided training. Specifically, the present invention relates to a general purpose architecture adaptable for use in the training of personnel in the performance of complicated tasks to produce the desired results with a minimum expenditure of energy, time, and/or resources.

BACKGROUND OF THE INVENTION

Applications of Artificial Intelligence in Training and Tutoring

A number of academic and industrial researchers have used artificial intelligence in an effort to teach a variety of subjects including geometry, computer programming languages, medical diagnosis and electronic trouble shooting. The earliest published reports which suggested the application of artificial intelligence concepts to teaching tasks appeared in the early 1970's. The article entitled "AI in CAI: An artificial Intelligence Approach to CAI" by J. R. Carbonell in the IEEE Transactions on Machine Systems, Vol. 11, No. 4, p. 190 (1970) and the article entitled "Towards Intelligent Teaching Systems" by J. R. Hartley and D. H. Sleeman in the International Journal of Machine Studies, Vol. 5, p. 215, (1973) are of specific interest. Hartley and Sleeman proposed an architecture for an intelligent tutoring system. However, since such proposal, no agreement has been reached among researchers on a general architecture for intelligent tutoring systems.

Examples of intelligent tutoring systems are SOPHIE (Brown, Burton & de Kleer, 1972, "Pedagogical, Natural Language and Knowledge Engineering Techniques in SOPHIE I, II, and III"; D. Sleeman & J. S. Brown (Eds.), Intelligent Tutoring Systems (p. 227). London: Academic Press), PROUST (Johnson & Soloway, April 1985, PROUST, Byte, Vol. 10, No. 4, p. 179) and LISP Tutor (Anderson & Reiser, April 1985, "The LISP Tutor," Byte, Vol. 10, No. 4, p. 159). SOPHIE was one of the first artificial intelligence ("AI") systems that was developed. SOPHIE was developed in response to a U.S. Air Force interest in a computer-based training course in electronic trouble shooting. SOPHIE contains three major components: an electronics expert with a general knowledge of electronic circuits, together with detailed knowledge about a particular type of circuit; a coach which examines student inputs and decides if it is appropriate to stop the student and offer advice; and a trouble shooting expert that uses the electronics expert to determine which possible measurements are most useful in a particular context. Although three versions of SOPHIE were produced, SOPHIE was never viewed as a finished product. One of the major problems associated with the SOPHIE systems was the lack of a user model.

PROUST and the LISP Tutor are two well-known, intelligent teaching systems that have left the laboratory for general application. PROUST, and its related program MICRO-PROUST, is a "debugger" for finding nonsyntactical errors in Pascal programs written by student programers. The developers of PROUST claim that it is capable of finding all of the bugs in at least 70% of the "moderately complex" programming assignments that it examines. PROUST contains an expert Pascal programer that can write "good" programs for the assignments given to students. Bugs are found by matching the assertions of the expert program with that of the student; mismatches are identified as "bugs" in the student program. After finding a bug, PROUST provides an English language description of the bug to the student, enabling the student to correct his or her error. PROUST cannot handle student programs that depart radically from the programming "style" of the expert.

The LISP Tutor is used to teach the introductory LISP course offered at Carnegie-Mellon University. The LISP Tutor system is based on the ACT (Adaptive Control of Thought) theory and consists of four elements: a structured editor which serves as an interface to the system for students, an expert LISP programmer that provides an "ideal" solution to a programming problem, a bug catalog that contains errors made by novice programmers, and a tutoring component that provides both immediate feedback and guidance to the student. Evaluations of the LISP Tutor show that it can achieve results similar to those obtained by human tutors. One of the LISP Tutor's primary features is its enforcement of what its authors regard as a "good" programming style. The "good" programming style feature prevents creative authorship by the student.

The existing systems are "intelligent tutoring or teaching systems." The teaching/tutoring task is distinguished from the training task. The training environment differs in many ways from an academic teaching environment. The differences are important in the design of an architecture for an intelligent training system. For example, assigned tasks are often mission-critical, i.e., the responsibility for lives and property depends on how well a person is trained to perform a task. Typically, people who are being trained already have significant academic and practical experience which is utilized in the task they are being trained to do. Also, trainees make use of a wide variety of training techniques. Different training techniques can range from the study of comprehensive training manuals, to simulations, to actual on-the-job training under the supervision of more experienced, trained personnel. Few tasks which require training must be accomplished by one method or style as exists in typical tutoring. Training a person to perform a task may require that considerable freedom be given the trainee in the exact manner in which the task may be accomplished.

People being trained for complex, mission-critical tasks are usually already highly motivated. Training for such complex tasks imposes on the trainer the responsibility for the accuracy of the training content and the ability of the trainer to correctly evaluate trainee actions. Typical tutoring systems do not provide such flexibility. A training system is intended to aid the trainee in developing skills for which he already has the basic or "theoretical" knowledge. A training system is not intended to impart basic knowledge such as mathematics or physics. Simply stated, a true training system is designed to help a trainee put into practice that which he already intellectually understands. Most importantly, a trainee must be allowed to perform an assigned task by any valid means. To achieve meaningful training, the flexibility to carry out any assigned task by any valid means is essential. Trainees must be able to retain and even hone an independence of thought and develop confidence in their ability to respond to problems, including problems which the trainee has never encountered and which the trainer may have never anticipated.

All phases of industry and government must maintain a large effort in training personnel. New personnel must be trained to perform the task which they are hired to perform, continuing personnel must be trained to upgrade or update their ability to perform assigned tasks and continuing personnel must be trained to perform new tasks. Often a great number of training methodologies are employed, singly or in concert. These methods include training manuals, formal classes, procedural computer programs, simulations, and on-the-job training. The latter method is particularly effective in complex tasks where a great deal of independence is granted to the task performer. Of course, on-the-job training is typically the most expensive and may be the most impractical training method, especially where there are many trainees and few experienced personnel to conduct such training.

Programming Languages for Artificial Intelligence Applications

All programming languages can be thought of as being divided into two primary functioning units: data and process. Data involves whatever means the language provides for representing objects which the programmer uses to manipulate. Typical data items might be variables used in formulas, matrices of numbers used for representing dimensionality, or lists of data groups such as patient records or student grades. Most conventional programming languages have evolved rather elaborate schemes for representing data, for example, integer and floating point representations. Process involves the programmer's directions for manipulation of the data structures. By analogy, if a computer program were like a recipe, data would be the ingredients and process would be the step-by-step cooking instructions.

Historically, computers have been typically utilized exclusively for mathematical calculation. However, more recently computers have begun to do reasoning, sometimes called "symbolic reasoning" in the computer science community. The standard upon which artificial intelligence systems are based is that intelligent systems reason about objects in the world, and do so in a rational way. Thus, the data in the artificial intelligence community was the representation of objects in the real world which were sometimes labeled as facts. Process became an inferencing scheme which could be used to manipulate the facts in a formal way. Artificial intelligence developed somewhat like first order logic which has a very precise means for defining axioms (facts or data) and a very orderly way of performing deduction and induction (inference or process).

The resulting languages which are used to implement most expert systems are termed "rule-based" languages. A programmer writes instructions for inferencing in the form called rules. Each rule has a "left-hand-side" used to match facts in the current database of facts and a "right-hand-side" used to perform actions on the facts in the database. Each rule is basically of the form "if you see such-and-such among the facts currently known, then do so-and-so." Sometimes rules are generally called "if-then" rules because of this analogy.

The application of any single rule is very simple. The underlying language checks to see if the description of the facts cited on the left-hand-side of the rule match any of the facts currently in the system. If so, then the actions described on the right-hand-side of the rule are carried out. This process continues until no more rules can be matched to the facts in the database. When no rules can be matched, the program ends.

Generally, facts are described as representations of data objects about which the system is going to reason. Facts are typically described in terms of a "relation" which is meant to describe a relationship between some object and one or more of its attributes. In general, facts are an assertion of a more general relation form. Rules can be used to retract or delete facts from the database. Further, rules can be used to assert new facts to the database. Thus, whenever a rule is executed or "fired," it may change the contents of the facts in the database. Any rule which has all of the patterns of its left-hand-side matched is placed on an agenda of rules which can potentially be fired. However, since the execution of any single rule may change the database, only one of the rules on the agenda is fired at once. When the particular rule is fired, the database of facts must be updated and the matching process is restarted. The cycle is repeated until the process of matching all of the rules does not produce any rule which has a fully satisfied left-hand-side.

Features of the Invention

Of primary concern in the present invention is to provide a general purpose architecture suitable for intelligent computer-aided training which can be readily adapted for use in numerous training disciplines.

It is, therefore, a feature of the present invention to provide an intelligent computer-aided training system which utilizes a general purpose architecture for adaptation to training in different fields.

A feature of the present invention is to provide an intelligent computer-aided training system which utilizes a plurality of expert systems which communicate via a common "blackboard" arrangement.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general purpose architecture which provides a user interface which is sufficiently similar to the actual task performed so that training skills are easily transferred from the training environment to the task environment.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general architecture which provides a domain expert system which is capable of performing the task to be trained by using rules describing the correct methods of performing the task and rules identifying typical errors.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general architecture adaptable to teach different tasks having an expert training scenario generator for designing increasingly complex training exercises based upon the current skill level of the trainee.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general architecture which has an expert training session manager for comparing the assertions made by the domain expert and by the trainee for identifying both correct and incorrect trainee assertions and for determining how to respond to incorrect trainee actions.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general architecture including a trainee model which contains a history of the individual trainee's interactions with the system together with summary evaluative data which can be accessed by both the trainee and an evaluator.

Yet another feature of the present invention is to provide an intelligent computer-aided training system having a general architecture whereby trainees can carry out an assigned task by any valid means.

Yet another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture for use in a wide variety of training tasks and environments which require modification of only one, or possibly two, of the components when changing tasks.

Still another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture whereby each altered component is designed to make the modifications necessary to produce an intelligent computer-aided training system for a specific task rapid and capable of being accomplished by persons skilled in the art of computer programming.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture for use by trainees already possessing the necessary educational background for the task for which the training is initiated.

Yet another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture whereby trainees are permitted great latitude in how they achieve a particular task such that trainees are permitted to follow any valid path to achieve the task, and further optional actions need not be taken, but the omission of optional actions is noted in the system and can be used in the generation of future training scenarios.

Still another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture whereby the modules communicate by means of a common fact base which fact base is termed a blackboard.

Another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture for segregating portions of the system that can be applied to other training environments and tasks.

Yet another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture whereby one module is a user interface designed for a specific environment and which can be used for training in other tasks that are performed in the same environment.

Still another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture whereby all task-specific items are confined to a single module for incorporating domain knowledge of the specific task as well as explanations, error messages and database information from which new training scenarios can be derived.

Yet another feature of the present invention is to provide an intelligent computer-aided training system having a general modular architecture whereby new training scenarios are designed uniquely for each trainee every time the specific trainee interacts with the system.

Yet another feature of the present invention is to utilize time constraints and distractions when the t