WO2014130499A1

WO2014130499A1 - Personalized agent pool for interaction with multiple automated providers

Info

Publication number: WO2014130499A1
Application number: PCT/US2014/017024
Authority: WO
Inventors: Edward Fredkin
Original assignee: Edward Fredkin
Priority date: 2013-02-21
Filing date: 2014-02-19
Publication date: 2014-08-28

Abstract

An automated apparatus, for enabling interaction with an automated provider, includes a database of personal information, an interaction encyclopedia, and an automated personal agent, sometimes assisted by a human personal agent, that interacts with an automated provider based on procedures stored in that interaction encyclopedia.

Description

PERSONALIZED AGENT POOL FOR INTERACTION WITH

MULTIPLE AUTOMATED PROVIDERS

CROSS REFERENCE TO RELATED APPLICATIONS

This application claims priority to U.S. Provisional Application No. 61/767,450, filed February 21, 2013, the contents of which are hereby incorporated by reference in their entirety.

FIELD OF DISCLOSURE

This disclosure relates to commerce and communications, electronic commerce, Internet based commerce and in

particular, to systems for simplifying electronic commerce, automated commerce, or ordinary commerce and communication.

BACKGROUND

A typical user interacts with many web sites and other automated providers of goods and services. Each such

automated provider has its own interface with its own

workflow. As a result, a user must engage in a great deal of tedious work such as entering various kinds of data and learning to navigate through the workflow of a particular automated provider. The time wasted in doing this is not to be underestimated. In many cases, time spent navigating through an automated provider' s workflow to obtain what one wants can be an appreciable fraction of an hour.

Once one has identified what one wants, it is of

course time to pay. For any business, payment should be as simple as possible. It makes little sense to place

impediments that might block the flow of currency into the coffers of a business. Yet, in too many cases, a consumer must enter complex sequences of numbers in order to part with his money. An appreciable number of transactions fail at this point. Shopping carts are left abandoned for no other reason that that payment is simply too complicated.

SUMMARY

The invention is based on the recognition that

workflows associated with providers, whether automated or otherwise, and preferences and personal data associated with principals can all stored in a memory accessible to a machine and that such machines can provide an interface between different providers, whether automated or otherwise, and a particular principal.

In one aspect, the invention features an apparatus for enabling interaction with an automated provider. Such an apparatus includes a database of personal information, an interaction encyclopedia, and an automated personal agent that interacts with an automated provider based on an entry in that interaction encyclopedia.

In some embodiments, the agent is configured to

interact with an automated provider having a web-based site. In others, the agent is configured to interact with an automated provider having an interface via voice and/or a telephone keypad interface. The agent can be configured to work with any interface that a human being can work with.

In some embodiments, the agent is configured to

interact with a provider having a web-based site. In others, the agent is configured to interact with an automated

provider having an interface via voice and/or a telephone keypad interface. In general, an agent can be configured to work with any interface that a human being can work with. In many embodiments, an agent successfully interacts telephonically with a human being employed by a provider, either on its own via speech recognition and speech

synthesis or through the assistance of a human being employed to act as an agent.

Embodiments also include those in which the

interaction encyclopedia includes a first entry

corresponding to a first automated provider and a second entry corresponding to a second automated provider, and those in which the interaction encyclopedia includes a first entry containing instructions for executing workflow in a first automated provider, and a second entry

containing instructions for executing workflow in a second automated provider. In yet other embodiments, the personal agent is configured to detect an inability to complete a workflow associated with an automated provider.

In other embodiments the personal agent is configured to be programmed to improve an extent to which it interacts with the automated provider. This process is assisted by the fact that an abstracted history of all such

interactions between the personal agent and automated providers can be retained in order to allow for the kind of analysis that allows for both automated learning about the characteristics of the various automated providers, and for the automated detection of changes in the characteristics of the various automated providers. Among these are

embodiments in which the personal agent is configured to solicit human assistance, from experienced, trained specialists, while improving its ability to interact with the automated provider.

Other embodiments include those in which a personal agent can automatically sign up and join airline reward clubs on behalf of the principal in order to later

automatically take advantage of travel reward points. The personal agent can automatically shop for items based on price, brand, and delivery details and can order such items for the principal while taking into consideration its knowledge of the desires and characteristics of the

principal .

In other embodiments, the personal agent communicates via texting. Additional embodiments include those in which an agent sets up appointments, manages the principal's calendar, and sets up conference calls or videoconferences .

In general, the personal agent is programmable to perform almost any and every task that could have also have been done by an experienced human personal assistant.

In some embodiments, a principal arranges for his agent to perform or attempt to perform a task that is new to the system of personal agents. In such cases, one or more trained humans may assist the agent. Other embodiments are arranged so that upon determining that human personal assistants are successfully handling a class of actions, those actions are stored and analyzed. Then, based on various considerations such as the frequency or the value of those actions, those actions are scheduled to be

incorporated into the body of actions that can be mostly handled successfully by personal agents. Also among the embodiments are those in which the personal agent is configured to detect improper interaction with the automated provider, and those in which the

personal agent is configured to attempt to predict, based on past experience, a procedure for interacting with the automated provider.

In many embodiments, whenever the principal is dealing with a personal agent, every aspect of the software that supports the work and interaction activity of the personal is programmable in such a way as to minimize one or both of the total time spent by the principal and time the

principal spends waiting for the agent to respond to his instructions .

In another aspect, the invention includes a system designed and implemented to respond in a meaningful and useful way to any and every input from the principal in less than 100 milliseconds (1/10 of a second) . Thus every step in the process of communicating to the personal agent will minimize both the total interaction time and the reaction delays in responding to every input from the principal .

This aspect of the operation of a system according to the invention can most often be accomplished by having all of the software and data that is involved in receiving and understanding a request that the principal is likely to make to the agent reside locally, to the extent it is possible and practical to do so, in the device that the principal uses to make the request, examples including but not limited to the principal's computer, cellphone, tablet, and iPad ™.Any clarification needed in order to understand the exact intention of the principal takes place in this local device.

In some embodiments, if the local device is

concurrently in a state in which immediate digital

communication is available, that communication is then part of an initial phase of a principal-agent interaction so long as the principal is still not subject to any delays greater than some predefined threshold value, such as 100 milliseconds. The principal's request is then communicated by the fastest means practical after being encoded into a space efficient form, by such communication media including, but not limited to LAN, a cell-phone network, or a similar medium for rapid digital communication.

In many embodiments, the system guarantees its rapid response time by prohibiting the implementation, during the software development process, of any method that cannot be made to deliver the specified performance. Thus, the

specified performance becomes a mandatory requirement of the programming.

What can be proven to be impossible to achieve, from a timing perspective, and which would, if implemented, result in delays that are in excess of the specifications will not be implemented, as part of the system. Thus the principal suffers no delays in specifying what the agent is to do.

The principal is then immediately freed of any other

interaction with the agent until, at a later time the agent reports back to the principal on either the completion of an assigned task, on progress towards completion, or on various remaining questions such as those related to various alternatives that the agent decides that the principal might wish to consider based on prior experiences between the principal and the agent. Other embodiments include those in which the personal agent is configured to solicit assistance from support staff in completing a workflow associated with an automated provider .

The apparatus claimed herein is understood to not be "software per se" and is understood to be a tangible device having mass and that consumes energy during operation thereof. To the extent that the apparatus can be viewed as having embodiments that can be viewed as intangible

"software per se," those embodiments are expressly excluded from the claims.

In another embodiment, the invention features a computer-implemented method that is tied to a particular tangible and physical machine for enabling interaction with an automated provider. Such a method includes providing a database of personal information, an interaction

encyclopedia, and a personal agent that interacts with an automated provider based on an entry in the interaction encyclopedia .

In some practices, providing a personal agent includes configuring the agent to interact with an automated

provider having a web site. In other practices, providing the personal agent

includes configuring the agent to interact with an

automated provider having a voice interface.

Also included within the scope of the invention are practices in which providing the interaction encyclopedia includes providing a first entry therein corresponding to a first automated provider and a second entry therein

corresponding to a second automated provider, practices in which providing the interaction encyclopedia includes

providing a first entry therein that contains instructions for executing workflow in a first automated provider, and providing a second entry therein containing instructions for executing workflow in a second automated provider, practices in which providing the personal agent includes configuring the agent to detect an inability to complete a workflow associated with an automated provider, and

practices in which providing the personal agent includes configuring the agent to solicit aid from support staff in completing a workflow associated with an automated provider. In additional practices of the invention, providing the personal agent includes configuring the agent to

improve its ability to interact with the automated provider.

Also included are practices in which providing the agent includes configuring the agent to solicit human

assistance when necessary to interact with the automated provider, those in which providing the personal agent

includes configuring the agent to detect improper

interaction with the automated provider, and those in which providing the personal agent includes configuring the agent to attempt to predict, based on past experience, a

procedure for interacting with the automated provider.

In another aspect, the invention includes a non- transitory and tangible computer-readable medium, the medium having encoded thereon software for enabling

interaction with an automated provider, the method

including providing a database of personal information, an interaction encyclopedia, and a personal agent that

interacts with an automated provider based on an entry in the interaction encyclopedia. The computer-readable medium recited in the claims is a tangible and non-transitory type of the form that is compliant with 35 USC 101 as of the time of the filing of this application. Those forms of computer-readable media that are not compliant with 35 USC 101 are specifically disclaimed.

Other embodiments include those in which the personal agent is configured to solicit assistance from support staff in completing a workflow associated with an automated provider . BRIEF DESCRIPTION OF THE FIGURES

These and other features of the invention will be apparent from the following detailed description and the accompanying FIG. 1, which shows a diagram of a system implementing the invention.

DETAILED DESCRIPTION

Referring to FIG. 1, a personal agent pool 10 features multiple personal agents 12A, 12B, 12C...12Z, each of which is associated with a corresponding principal database 14A, 14B, 14C...14Z. A principal database 14A includes information that would be useful in carrying out, on behalf of a principal 16A, transactions at web sites or voice interface systems associated with stores, commercial establishments, medical facilities, entertainment facilities, government agencies, postal services, and any other system that provides a structured workflow that can be used to provide goods or services. Collectively, these will be referred to as "automated providers 18A-Z" that are in communication with the agent pool 10 via a communication network 20.

Information stored in a principal database 14A can include identifying information, data identifying available forms of payment, data identifying any affinity or loyalty accounts associated with various vendors, data identifying discounts available, clothing and shoe sizes, and data identifying principal preferences, for example whether a principal 16A prefers aisle seats, whether a principal 16A has an aversion to being on the 13^th floor, whether a principal 16A has any food allergies, and so on. The Brand preferences of the principal 16A along with various other sorts of personal preferences, such as temporal preferences involving day of the week, holiday, weekend, time of day, etc. can also be taken into account.

A personal agent 12A is in communication with a particular principal database 14A. As a result, a personal agent 12A has access to considerable useful information about that principal 16A.

A personal agent 12A is also in communication with an interface to a current interface encyclopedia 22. The interface encyclopedia 22 includes entries 24A-24Z for each of a very large number of automated providers 18A-18Z.

A useful feature of an automated provider 18A is that there generally exists a step-by-step procedure that can be followed to accomplish some task. When this procedure is repeatable, it is possible to program the personal agent

12A to follow the interaction with the automated provider

14A. This programming is generally carried out by a human programmer. However, having been done once for a particular personal agent 12A, the resulting program can be used by all other personal agents 12B-12Z.

For particularly simple cases, where the procedure followed by the automated provider 18A is simple enough, a personal agent 14A can learn autonomously and modify its own instructions to follow the interaction with the

automated provider 14A.

The current procedures, simple step by step or more complicated, to accomplish a particular task at a

particular automated provider 18A need to be implemented only once. The details of the implementation, whether

carried out by a programmer who re-programs the personal agent 14A or through autonomous learning by the personal agent 14A, are then stored in the interface encyclopedia 22. Once stored therein, that provider information is available to any personal agent 12A-12Z. In particular, the interface encyclopedia 22 includes entries 24A-24Z that, for a

particular task to be carried out at a particular automated provider 18A, will provide a program of instructions to be executed by a personal agent 12A for carrying out that task. The personal agent 12A can thus function as an

interface between a principal 16A and any automated

provider 18A for which an entry 24A exists in the interface encyclopedia 22. In this way, a principal 16A is no longer burdened with the ordeal of having to learn or remember the vagaries of interaction with a particular automated

provider 18A. Instead, the principal 16A can simply

instruct his personal agent 12A to perform a particular task. Moreover, because the personal agent 12A has access to a principal's personal database 14A, the personal agent 12A can accomplish tasks with minimal intervention by the principal, and in a way that complies with the principal's preferences .

In some embodiments, the software that implements the personal agent 12A makes use of effectively parallel computation techniques in order to better achieve the goal that its principal 16A has provided.

Some automated providers have mechanisms designed to assure that they are interacting with a human being as opposed to some kind of robotic software. An example (known in the art as "CAPTCHA") is a spatially distorted set of letters and/or numbers that only a human can decipher.

Since, in this case, it will be in the interest of the automated provider to not exclude the personal agent 12A, it will be in the interest of the owner of the web site for there to be a code or other mechanism that can be used to identify the personal agent 12A as a legitimate user.

For example, suppose a principal 16A wishes to leave Boston for Manhattan by train the next morning and return the following evening. The principal 16A communicates this to the personal agent 12A. This could be done by simply saying that request during a voice interaction with the personal agent 12A, relying on voice recognition, along with stored information related to the principal 16A to establish the identity and other details of the principal's intentions to the personal agent 12A. The personal agent can, in the case where similar travel has been booked

before, make use of that prior reservation to make various assumptions. The personal agent 12A could reply by voice and or by email of its understanding of the request, such as by elaborating the details of the proposed reservations which include aspects known to the personal agent 12A from prior interactions such as desired class of service, choice of seats, taxis or limos to or from the train stations, etc. This allows the principal 16A to confirm, modify or cancel the proposed arrangements. Or, a standard more formal

interaction could take place between the principal 16A to instruct the agent 12A, most likely by the principal

entering the details into a computer, a tablet, or phone. This is preferably a two-way process, where the personal agent 12A reacts in ways that allow it to verify that it correctly understands the intention of the principal 16A.

Upon receiving the instruction, the personal agent 12A reads the principal's personal database 14A and retrieves the information that the principal's preferred mode of train travel is by the Acela high-speed train. The personal agent 12A also retrieves information from the database 14A, that the principal 16A prefers a morning seat on the west side of the car and an evening seat on the East side of the car. The personal agent 12A then consults the interface encyclopedia 22 and retrieves the entry 24A for the train service provider 18A. Based on that entry 24A, the personal agent 12A begins to interact with the automated provider

18A for train service. After some interaction, the personal agent 12A might know that it should communicate a number of travel options to the principal 16A and awaits confirmation. Upon receiving confirmation, the personal agent 12A

continues the interaction and arranges payment. To do so it uses information previously stored in the principal

database 14A. As a result, the principal 16A is spared the trouble of interacting with automated provider 18A. The personal agent 12A can then automatically print out, on the appropriate printer, whatever printed documents are needed by the principal 16A.

In some cases, the agent 12A may be unsure of what to do. Under these circumstances, the agent 12A can pose

certain questions to the principal 16A to clarify the

principal's intent. Or the agent 12A may request human intervention from a staff 26 of trained personnel.

It should be noted that the agent 12A is not

restricted to interaction with web sites. For example, many automated providers 18A maintain an automated telephonic voice interface having a well-defined structure for

accomplishing tasks. These telephone interfaces can be controlled by voice or telephonic keypad commands. Given improvements in voice synthesis, it would not be

unreasonable to expect a personal agent 12A to interact with such a system via synthesized voice, by simulating the pressing of keys on a keypad, or both. In some cases, the personal agent 12A recognizes that the process is not proceeding as it should, for example by noting that the process differs from a structure such as a conditional tree structure stored in a memory thereof as being representative of a potential dialog with the

telephone interface. When this occurs, the personal agent

12A either attempts to solve the problem itself or seeks the assistance of a human agent, or both. On occasion, when the personal agent 12A realizes that it is faced with an ambiguity that requires clarification from the principal

16A, it queries the principal 16A, by any means appropriate, to clarify the situation.

To avoid repeatedly attempting the same series of interactions or actions, the personal agent 12A preferably includes a loop detector. This is accomplished by always having the personal agent 12A compare what it plans to do with what it has recently done. To the extent these share a characteristic in common, this enables the personal agent

12A to identify the possibility that it is about to enter a loop in which it repeats actions that are either identical to or substantially similar to those it has attempted in the past.

In some embodiments, the agent 12A includes a loop detector that translates records of an agent's actions or communications into a high level descriptive language. This facilitates detection and avoidance of repetition that might otherwise be masked by minor differences in the

details of two substantially identical actions. The object of the loop detection system is to avoid the possibility of the agent falling into various kinds of repetitive actions that do not efficiently advance the accomplishment of its goals. An example of such a possible loop is as follows: The agent 12A seeks to book the

principal 16A on a flight from Boston to San Francisco. The agent 12A notes that there are two candidate flights: one is on a first carrier and another on a second carrier. The agent 12A notes that although there are seats available on the first carrier, none of those seats are of the type the principal would prefer. The agent 12A then notes that there are no non-stop flights on the second carrier. The agent 12A then returns to the first carrier, and then to the second carrier as it attempts to find a flight with the required combination of characteristics preferred by the principal .

While an agent program could be written that

explicitly avoids many kinds of unnecessary repetitious actions by examining possibilities in some kind of

predetermined sequence in such cases, the complexity of a good and effective agent program requires an explicit introspective mechanism to cope effectively and efficiently with the general problem of avoiding loops, especially more complex loops, either during contemplative activity, during interactive activity with either the principal or various providers or under any other circumstances.

Additionally, given improvements in voice recognition and synthesis, a personal agent 12A can also engage in rudimentary conversation with a live provider 18B. For example, when waiting for technical support, a personal agent 12A can wait in a queue for live assistance, and, upon detecting a live operator, cause a telephone call to be placed to the principal 16A so that the principal 16A can take the agent's place and continue the conversation. While doing so, the agent 12A, based on information culled from the principal database 14A, can read off a serial number or other personal information that would routinely be requested by technical support staff.

The personal agent 12A can thus carry out requests in much the same way that an intelligent personal human personal assistant might. For example, the agent 12A could book a flight and hotel on line, select class and seat locations, and specify any special meals. It could do so not just for one airline and hotel but for every airline and hotel that maintains an automated provider 18C. The personal agent 12A would then present its proposed

transactions to the principal 16A for approval. Upon receiving such approval, the personal agent 12A would proceed to consummate the transaction. An advanced version of the personal agent 12A could accomplish many goals on behalf of the principal, by means of telephonic voice communications with various human beings or automated voice recognition systems. For example, when the principal wishes to have the agent 12A book a room in a small rural inn that has no website and where the phone is answered by a person, the agent 12A could either attempt to arrange the booking by use of speech synthesis and speech recognition or it could request the assistance of a human agent. By listening in on the conversation between the human agent and the innkeeper, the automated personal agent 12A could compare its own contemplated actions and communications with those of a skilled and experienced human agent in order to improve its

capabilities of handling such systems. In such cases, the human agent could, after completion of the interaction, give a useful evaluation as to whether or not the situation was a typical one likely to often occur, or perhaps a bizarre one-time situation where no useful modifications of the programmed agent 12A's behavior are indicated.

Of course, the agent 12A could react to various things from experiences. However what is significant is that when there is an instance where the agent 12A acquires the ability to do something useful, the information acquired could eventually become part of the memory of every acting agent 12A. While this is useful in allowing the entire agent system to grow into a more intelligent and capable system, that process must proceed with caution and

intelligent oversight. Otherwise, "learning" a crucial but inappropriate fact could cripple the whole system. Thus, when something new is potentially "learned" by the system, the use of that information must be put on probation so the use of it is both cautious and closely monitored to

reinforce the conclusion that useful information has been retained. In many such cases, human experts could be brought into the process of reviewing situations where the system concludes that it may be useful to modify its behavior based on various experiences.

In some cases, there may be holes in the knowledge available at the interface encyclopedia 22. Or the workflow at an automated provider 18A may have been changed in a way that makes it difficult for the personal agent 12A to carry out its function. To accommodate these cases, the personal agent 12A has its own technical support line linked to live trained personnel 26 who can assist the personal agent 12A. In doing so, the trained personnel 26 will have a basis for providing information that could be used for updating the interface encyclopedia 22, thus ensuring that the next personal agent 12A that is asked to carry out that task will have the necessary knowledge and capabilities to do so. Websites are constructed using HTML and related

software protocols known in the art. Such sites can be analyzed for recognizable entry fields, such as fields requesting the user to enter their first and last names.

The system can recognize when a site is completely new, or is a known site but has undergone modification since it was last analyzed. In these cases, the system may proceed

automatically, but be alert for error messages or other indications that the information is not being entered and processed correctly. The system may also use Bayesian

probability analysis and/or other techniques known in the art to guess from past experience what information is being requested when encountering a new or modified automated provider, and to generate a confidence value for said

prediction, which can be used in conjunction with the

responses of the website to evaluate the risk that

information is not being entered and accepted correctly. An exemplary method and system for carrying out such so-called "machine learning" is disclosed by Goodman, et al, U.S.

Patent No. 7254569, the contents of which are herein

incorporated by reference. Having described the invention, and a preferred embodiment thereof, what I claim as new, and secured by letters patent is:

Claims

An apparatus for enabling interaction with an automated provider, said apparatus comprising a database of personal information, an interaction encyclopedia, and a personal agent that interacts with an automated provider based on an entry in said interaction encyclopedia.

The apparatus of claim 1, wherein the agent configured to interact with an automated provider having a web site.

The apparatus of claim 1, wherein the personal agent is configured to interact with an automated provider having a voice interface.

The apparatus of claim 1, wherein said

interaction encyclopedia includes a first entry corresponding to a first automated provider and a second entry corresponding to a second automated provider .

The apparatus of claim 1, wherein said

interaction encyclopedia includes a first entry containing instructions for executing workflow in a first automated provider, and a second entry containing instructions for executing workflow in a second automated provider.

The apparatus of claim 1, wherein said personal agent is configured to detect an inability to complete a workflow associated with an automated provider .

The apparatus of claim 1, wherein said personal agent is configured to solicit aid from support staff in completing a workflow associated with an automated provider.

The apparatus of claim 1, wherein said personal agent is configured to improve its ability to interact with said automated provider.

The apparatus of claim 8, wherein said agent is configured to solicit human assistance while improving its ability to interact with said

automated provider.

The apparatus of claim 1, wherein said personal agent is configured to detect improper

interaction with said automated provider.

The apparatus of claim 1, wherein said personal agent is configured to attempt to predict, based on past experience, a procedure for interacting with said automated provider.

A computer-implemented method that is tied to a particular tangible and physical machine for enabling interaction with an automated provider, said method comprising providing a database of personal information, an interaction encyclopedia, and a personal agent that interacts with an

automated provider based on an entry in said interaction encyclopedia.

The method of claim 12, wherein providing a personal agent comprises configuring the agent to interact with an automated provider having a web site .

The method of claim 12, wherein providing the personal agent comprises configuring the agent to interact with an automated provider having a voice interface.

The method of claim 12, wherein providing said interaction encyclopedia includes providing a first entry therein corresponding to a first automated provider and a second entry therein corresponding to a second automated provider.

The method of claim 12, wherein providing said interaction encyclopedia includes providing a first entry therein that contains instructions for executing workflow in a first automated provider, and providing a second entry therein containing instructions for executing workflow in a second automated provider.

The method of claim 12, wherein providing said personal agent comprises configuring said agent to detect an inability to complete a workflow associated with an automated provider.

The method of claim 12, wherein providing said personal agent comprises configuring said agent to solicit aid from support staff in completing a workflow associated with an automated provider.

19. The method of claim 12, wherein providing said personal agent comprises configuring said agent to improve its ability to interact with said automated provider.

20. The method of claim 19, wherein providing said agent comprises configuring, by a human, of said agent to cause said agent to interact with said automated provider.

21. The method of claim 12, wherein providing said personal agent comprises configuring said agent to detect improper interaction with said

automated provider.

22. The method of claim 12, wherein providing said personal agent comprises configuring said agent to attempt to predict, based on past experience, a procedure for interacting with said automated provider .

23. A computer-readable medium of the type that is regarded as statutory within the meaning of 35 USC 101, said medium having encoded thereon software for enabling interaction with an automated provider, said method comprising providing a database of personal information, an interaction encyclopedia, and a personal agent that interacts with an automated provider based on an entry in said interaction encyclopedia.