Ai-powered coach

Abstract

Problem: A conventional virtual coach, which is based on predefined rules and logic, proves unfit to adapt to the individual needs and preferences of coachees. Different coachees tend to have different goals, challenges, learning styles, personalities, and feedback preferences. Hence, a rule-based virtual coach is unable to tailor its coaching approach and content to suit each user's unique situation and expectations, which may result in frustration or boredom. Ensuring data protection compliance is a significant challenge in coaching and some coachees may be hesitant to share certain information with another person. Solution: Method of coaching a user using natural language, characterized in that a client, served by a backend, receives audio input from the user, the backend processes the audio input, generates a response appropriate to the input by means of a generative, preferably autoregressive NLP model, and synthesizes an audio stream expressing the response in the natural language, and the client plays the audio stream back to the user.

Description

DescriptionAl-powered coachTechnical Field

[0001] The invention relates to the process and products as per the first part of the independent claims.Background Art

[0002] Coaching is a form of development in which an experienced person, called a coach, supports a learner or client - hereinafter referred to as the coachee - in achieving a specific personal or professional goal by providing training and guidance. More specifically, business coaching - sometimes labelled executive coaching, corporate coaching, or leadership coaching - is a type of human resource development for executives, members of management, teams, and leadership. Here, the coach provides positive support, feedback, and advice to improve the coachee’s personal effectiveness in the business setting and help him or her advance towards specific professional goals pertaining to, for instance, career transition, interpersonal and professional communication, performance management, organizational effectiveness, developing executive presence, enhancing strategic thinking, dealing effectively with conflict, and building an effective team within an organization.

[0003] PTL1 , which is hereby incorporated by reference in its entirety, discloses a state-of-the-art online coaching platform.

[0004] NPL1 provides an overview of current developments in digital coaching companions and assistants, especially conversational agents (“chatbots”).Summary of invention

[0005] The invention is set out in the appended set of claims.Technical Problem

[0006] A conventional virtual coach, which is based on predefined rules and logic, proves unfit to adapt to the individual needs and preferences of coachees. Different coachees typically have different goals, challenges, learning styles, personalities, and feedback preferences. Hence, a rule-based virtual coach is unable to tailor its coaching approach and content to suit each user's unique situation and expectations, which may result in frustration or boredom. Further, a conventional virtual coach does not use science-based methodology to optimize the effectivity of coachings for a certain coachee.Solution to Problem

[0007] The problem is solved as per the second part of the independent claims. Advantageous effect of invention

[0008] By employing a generative natural language processing (NLP) model, the proposed method enables a virtual or “digital” coach to engage in dynamic and personalized conversations with users. Contrary to a rule-based agent, such coach may remember the user's progress and feedback over time and adjust accordingly. It can further switch between different modes of interaction, such as coaching, mentoring, teaching, or conversing, encouraging users to actively participate in the coaching process by asking questions, giving feedback, and sharing information and stories. An advanced embodiment may help users translate their insights into actions by providing them with concrete steps, resources, and tools to achieve their goals and follow up with them to monitor their progress and provide support.

[0009] Other than in presence or real online coaching, a coach as per the invention may be accessed immediately anytime and anywhere from different client devices, allowing coachees to choose between voice or text interaction depending on their situation and preference and share their thoughts and feelings without any fear of delays, judgment or bias.Further, a conversation transcript can be stored at the coachee’s preference.

[0010] While there is reason to believe that coaching is and will remain fundamentally human, requiring years of professional training and experience and a large degree of empathy and responsibility to achieve success, such artificial intelligence (Al) could nonetheless supplement, assist, support, and enhance the overall coaching experience for both coachees and coaches. If firmly grounded in behavioral science and in line with strict privacy regulations and coaching ethics, an embodiment of the invention may thus be instrumental in testing the limits of technology, exploring how Al can complement the coaching process, and help democratize coaching for people at all career levels worldwide. A real coach may be able to review virtual coaching transcripts with a coachee to provide further feedback, context, and integrate virtual coaching into a hybrid coaching experience.

[0011] Variants of the invention integrate several advanced technologies, including an autoregressive large language model (LLM) exhibiting the generative pre-trained transformer (GPT) architecture, speech recognition, dynamic context building, and more. These technologies work together seamlessly in real-time to deliver a powerful, engaging, and effective experience.

[0012] Variants of the invention assess the effectiveness of different coaching methodologies for a given coachee and apply several advanced sciencebased coaching methodologies to ensure ideal coaching success depending on the coachee as well as coaching goals and preferences.

[0013] Variants of the invention implement technical and organizational measures to ensure advanced security, data protection, manage the total environmental impact, and ensure a consistently high quality of the provided coaching in line with ISO9001 , ISO14001 , IS027001 , the GDPR, and other requirements.Brief description of drawings

[0014] Figure 1 shows the architecture of a virtual coach.

[0015] Figure 2 shows the complete architecture of an online coaching platform. Description of embodiments

[0016] An embodiment of the invention takes the form of a conversational Al coach readily available to conduct simple, prompted coaching conversations with guidance, recognize language and context, and understand written communication and idiomatic expressions. Due to its machine learning (ML) abilities, the coach may further be trained to interact by way of complex, structured coaching conversations, profound coaching frameworks, and visual communication such as through facial expressions.

[0017] Figure 1 illustrates an implementation of the method compliant with a REST architectural style in a serverless infrastructure based on Amazon Web Services (AWS). In the present embodiment a backend, implemented in the Python scripting language, is deployed using the Amazon simple storage service (S3) and executed by a serverless event- driven platform such as the Lambda (A) computing platform that is provided as part of the AWS infrastructure. The backend exposes its services to a front-end client through the Amazon application programming interface (API) Gateway, authorizing access through the client by means of user and identity pools maintained in an identity and authentication management tool such as Amazon Cognito.

[0018] As any back-end developer will appreciate, corresponding embodiments, without departing from the scope of the invention, may employ Microsoft Azure, Google Cloud, or other suitable cloud or private infrastructure as a service (laaS) that meets the requirements of the platform with respect to frequency and volume of data.

[0019] The backend server consists of two components hereinafter referred to as the transcription service (ch_mls_transcribe) and chatbot service (ch_mls_chatbot). Upon receipt of any audio input from the user, the former service, which is orchestrated as a container in Amazon ECS, firstmakes use of Facebook Denoiser for noise cancelation. It then transcribes, by means of a fast speech-to-text service such as OpenAI Whisper that is capable of handling different accents and languages, the “cleaned” audio input, passing the user audio transcript to both the client and chatbot service.

[0020] Upon receipt of the transcribed text, the chatbot service requests an OpenAI GPT (for example GPT-4) pre-trained transformer model to generate an appropriate response. In line with the NLP technique known as prompt engineering or prompting, each task of this model is embedded in its respective input. Hence, the chatbot service provides a context manager function to keep track of the user's goals, preferences, emotions, feedback, and past conversations and include a summary of the user's profile and active conversation with the prompt. To maintain the continuity and coherence of the conversation, said context manager updates the prompt after each response from the OpenAI API. The prompt may further be statically parameterized by a description of the coaching style and approach (see below for details), but also instructions on creating responses, including sentence length, tone, topic focus, goals for the conversation, type, and number of questions.

[0021] From the response thus generated, the chatbot service, using the Amazon Polly text-to-speech (TTS) converter, synthesizes an audio stream expressing that response in spoken language. Herein, aspects such as pitch, rate, volume, and emphasis of the speech may be controlled by means of speech synthesis markup language (SSML). For playback to the user, the chatbot service submits the resulting audio response to the client through the transcription service.

[0022] As mentioned previously, the chatbot service, much like a human coach, uses instructional material for coachees within a dense framework designed to foster diversity, equality, and inclusion (DE&I), coaching culture, et cetera. Typically, the coachee would study such material, receive further coaching, and so forth. In an advanced embodiment, said material could be introduced automatically and dynamically during the conversation, which in turn would then be based on the previouslypresented content. For this purpose, the chatbot service generates a summary of each coaching and stores it in a table maintained by a relational database management system (RDMBS) such as PostgreSQL (“Postgres”).

[0023] To this end, coachees’ control and rights over personal data are enhanced through extensive and effective measures that meet the principles of data protection by design and by default. For instance, any personal data is encrypted during transit, may be anonymized or pseudonymized and administrators authenticate via a two factor process and important administrators generally by means of a cryptographic hardware security key as per the FIDO Client to Authenticator Protocol 2 (CTAP2). Moreover, the platform enforces role-based access control and data deletion, anonymization, and retention policies wherein digital coaching sessions are not stored except during their transmission, data processing protocols are deleted within a set number of days following the end of the processing, for example potentially necessary evidence for legal disputes may be retained for 4 years, and contracts stored for 7 years after the end of the contractual relationship. Where required, the transfer of personal data to third countries is avoided or performed in compliance with applicable data protection and other law.

[0024] An advanced security information and event management (SIEM) system ensures security relevant events are automatically monitored and can be acted upon. Further technical and organizational measures, including from IS027008 and AICPA SOC2 such as regular audits and penetration tests are implemented to ensure advanced information security.

[0025] To satisfy data protection requirements, the data processing location can be selected. Advanced technical and organizational measures (such as encryption, anonymization, pseudonymization, tokenization, and storage splitting and limitation) address legal and regulatory risks caused by data transfers from such data processing location to countries where different laws and regulations apply (also called “Third Countries”).

[0026] Consent management in the frontend and backend, for example via a consent management tool such as Usercentrics and related backendservices, ensure coachees are transparently informed and can control the way their data is processed depending on the applicable laws and regulations in particular in relation to risk applicable in the field of artificial intelligence.

[0027] Virtual coaching may be provided by an Al coach in line with conformity and quality assurance requirements of ISO21001 and / or ISO17024.

[0028] Online coaching can reduce environmentally sensitive emissions over offline coaching. Careful measurement, tracking, and selection and consideration of the data processing system further reduces the environmental impact of virtual coaching, for example by intermittently muting the microphone, selecting low emission processing locations such as AWS Ireland, carefully configurating OpenAI API parameters, and avoiding video transmissions.Industrial applicability

[0029] The invention is designed to drive conversations, research, and innovation around Al in the coaching industry.Citation list

[0030] The following documents are cited hereinbefore.Patent literature

[0031] PTL1 : PCT / EP2022 / 053536 (COACHHUB GMBH) 14.02.2022

[0032] NPL1 : MAI, Vanessa, et al. Potenziale und Einsatzmoglichkeiten von digitalen Coaching-Begleitern und Assistenten. Organisationsberatung, Supervision, Coaching. 18.01.2023, vol.30, no.1 , p.45-57.

Claims

ClaimsClaim 1. Method of coaching a user using natural language, characterized in that a client, served by a backend, receives audio input from the user, the backend processes the audio input, generates a response appropriate to the input by means of a generative, preferably autoregressive NLP model, and synthesizes an audio stream expressing the response in the natural language, and the client plays the audio stream back to the user.Claim 2. Method as per Claim 1 , wherein the model is a preferably pre-trained transformer such as OpenAI GPT-4.Claim 3. Method as per Claim 1 or Claim 2, wherein the backend is implemented, preferably using a scripting language such as Python, in a serverless infrastructure such as AWS.Claim 4. Method as per Claim 3, wherein the backend provides a transcription service that, upon receiving the audio input, transcribes the audio input into a text input.Claim 5. Method as per Claim 4, wherein the backend provides a chatbot service that, upon the transcription service transcribing the audio input, prompts the model based on the text input.Claim 6. Method as per Claim 5, wherein the chatbot service is deployed using a storage service such as Amazon S3 and executed by a computing platform, such as AWS Lambda, of the infrastructure.Claim 7. Method as per Claim 5 or Claim 6, wherein the chatbot service is exposed to the client through an application programming interface, for example, by means of Amazon API Gateway.Claim 8. Method as per Claim 7, wherein upon the client accessing the interface, the chatbot authenticates the user, for example, by means of Amazon Cognito.Claim 9. Method as per any of Claim 5 through Claim 8, wherein, upon request, the chatbot service generates a summary of the coaching and stores the summary in a database table, for example, by means of a relational database management system such as PostgreSQL.Claim 10. Method as per any of Claim 4 through Claim 9, wherein, the transcription service transcribes the audio input by means of a speech-to- text service such as OpenAI Whisper.Claim 11. Method as per any of Claim 4 through Claim 10, wherein the transcription service is orchestrated as a container by means of an elastic container service, such as Amazon ECS, of the infrastructure.Claim 12. Method as per any of Claim 3 through Claim 11 , wherein the audio stream is synthesized from the response by means of a text-to- speech service, such as Amazon Polly, of the infrastructure.Claim 13. Method as per any of the preceding claims wherein the virtual coaching is combined with classic online coaching to create hybrid coaching for example by providing suggestions to the coach and / or coachee during or outside of a live coaching session.Claim 14. Method as per any of the preceding claims wherein the virtual coaching is provided according to ISC9001 , ISC14001 , IS027001 , and SOC2.Claim 15. Method as per any of the preceding claims wherein the virtual coaching is provided according to ISC21001 and / or ISO17024.Claim 16. Data processing apparatus comprising means for carrying out a method as per any of the preceding claims.Claim 17. Computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any of Claim 1 through Claim 15.Claim 18. Data carrier signal carrying the computer program of Claim 17.

Images