49992
Mind AI
A natural language reasoning engine based on a new paradigm
The Republic of Korea
Market: Artificial Intelligence
Stage of the project: Idea or something is already done
Date of last change: 23.10.2020
The Republic of Korea
Market: Artificial Intelligence
Stage of the project: Idea or something is already done
Date of last change: 23.10.2020
Idea
Mind AI is an artificial intelligence engine that converts natural language into novel data structures to perform human-like reasoning. The technology behind Mind AI is a basic unit of reasoning called the Canonical, which can model anything that can be put into natural language within its structure. We call Canonicals units of reasoning because they embody the three types of logical reasoning that humans utilize to solve problems and answer questions: deduction, induction, and abduction.
Current Status
● Mind Expression is a conversational reasoning AI infrastructure, initially targeted for customer support in the telecommunication domain (internet, mobile, cable TV). The product will be released by 2nd quarter of 2021.
● True Mobile, a subsidiary of CP Group, has over 38M subscribers and is the leading telecommunications company in Thailand. They had been experiencing difficulty in expanding and providing coverage to their user base with their customer service chatbot, Mari (based on Google Dialogflow). Through multiple POC models and investigation, Mind AI proved its technology to be more advanced and agile, and we began our commercial partnership in August. Together we are working to release the Mari assistant on Mind Expression by 2nd quarter of 2021.
● True Mobile, a subsidiary of CP Group, has over 38M subscribers and is the leading telecommunications company in Thailand. They had been experiencing difficulty in expanding and providing coverage to their user base with their customer service chatbot, Mari (based on Google Dialogflow). Through multiple POC models and investigation, Mind AI proved its technology to be more advanced and agile, and we began our commercial partnership in August. Together we are working to release the Mari assistant on Mind Expression by 2nd quarter of 2021.
Market
● Current Endeavors
- True Mobile (38 Million Subscribers) within the CP group in Thailand has signed a commercial contract, with confidence in the technology of Mind AI. Currently, the project is to completely migrate the Google Dialogflow-based chatbot Mari to Mind Expression, starting with a kickoff in August, and aiming to end the project in April next year.
- South Korean Edu-Tech company ST Unitas (Acquired Princeton Review in 2017) and Mind AI are conceiving of a completely new form of Conversational AI Tutor that will bring innovation to the field of education.
- CP All within the CP Group in Thailand is discussing the development of products using the artificial intelligence technology of Mind AI in a Companion Robot for elderly care (in the health tech field) and at 7-Eleven (in the food tech field).
- Smart, under Daimler, Germany, is working on a project to migrate their currently operating chatbot to one based on Mind Expression.
● Future Directions
Conversational AI Infrastructure in the following domains:
- Telecommunication
- Retail and eCommerce
- Travel and Hospitality
- Banking, Financial Services and Insurance (BFSI)
- Media and Entertainment
- Education
- True Mobile (38 Million Subscribers) within the CP group in Thailand has signed a commercial contract, with confidence in the technology of Mind AI. Currently, the project is to completely migrate the Google Dialogflow-based chatbot Mari to Mind Expression, starting with a kickoff in August, and aiming to end the project in April next year.
- South Korean Edu-Tech company ST Unitas (Acquired Princeton Review in 2017) and Mind AI are conceiving of a completely new form of Conversational AI Tutor that will bring innovation to the field of education.
- CP All within the CP Group in Thailand is discussing the development of products using the artificial intelligence technology of Mind AI in a Companion Robot for elderly care (in the health tech field) and at 7-Eleven (in the food tech field).
- Smart, under Daimler, Germany, is working on a project to migrate their currently operating chatbot to one based on Mind Expression.
● Future Directions
Conversational AI Infrastructure in the following domains:
- Telecommunication
- Retail and eCommerce
- Travel and Hospitality
- Banking, Financial Services and Insurance (BFSI)
- Media and Entertainment
- Education
Problem or Opportunity
● Transparency
- Google Dialogflow and other competitor products have a configuration that responds to one intent. For example, let's say the meaning of the intent is “Mobile Usage”. In order to classify this intent, generally, more than 10 to 100 similar expressions are needed. The reason being, when trying similar phrases that had the same obvious meaning to humans, the machine did not pick up on that meaning. And when creating intents, without transparency, when some unexpected phrase does not work, there is no way to find out exactly why it didn’t work.
● Intent Classification without a Subject & Disambiguation
- Competitors' products cannot figure out which subject intent should be activated when the user inputs “I have a problem”, for example. You can duplicate the additional logic to handle this disambiguation for each intent it could resolve to, but that adds another problem with reconfiguring this additional logic every time you add a new intent.
● Context Hopping
- When competing products are talking about TV problems, for example, if we then talk about Internet problems, they lose the context of the existing TV problems if we try to talk about TV problems again. This is because only one intent is active at a time.
● Adaptive Learning
- Competitors' products experience great difficulties when new content is added to the user's troubleshooting manual or when the flow needs to be changed. This is because it is necessary to propagate all the effects of the new content on the existing flow, and also check whether that content is properly classified.
- Google Dialogflow and other competitor products have a configuration that responds to one intent. For example, let's say the meaning of the intent is “Mobile Usage”. In order to classify this intent, generally, more than 10 to 100 similar expressions are needed. The reason being, when trying similar phrases that had the same obvious meaning to humans, the machine did not pick up on that meaning. And when creating intents, without transparency, when some unexpected phrase does not work, there is no way to find out exactly why it didn’t work.
● Intent Classification without a Subject & Disambiguation
- Competitors' products cannot figure out which subject intent should be activated when the user inputs “I have a problem”, for example. You can duplicate the additional logic to handle this disambiguation for each intent it could resolve to, but that adds another problem with reconfiguring this additional logic every time you add a new intent.
● Context Hopping
- When competing products are talking about TV problems, for example, if we then talk about Internet problems, they lose the context of the existing TV problems if we try to talk about TV problems again. This is because only one intent is active at a time.
● Adaptive Learning
- Competitors' products experience great difficulties when new content is added to the user's troubleshooting manual or when the flow needs to be changed. This is because it is necessary to propagate all the effects of the new content on the existing flow, and also check whether that content is properly classified.
Solution (product or service)
● Mind Expression is a conversational reasoning AI infrastructure, initially targeted for customer support in the telecommunication domain (internet, mobile, cable TV).
● There will be neither decision tree nor explicit process flows required when designing a conversational AI with Mind Expression, so the conversational AI designer needs only to concentrate on the actual customer support content rather than building a complex decision tree. This is due to our cognitive model, which deals fundamentally with meaning. Hence, an end-user will feel more like he or she is conversing with a real human agent.
● If the platform does not understand some words or phrases, the reasoning engine will infer what the meaning might be. This is to be confirmed by the end-user or conversational AI designer. The confirmed knowledge will then immediately be usable because of our adaptive learning capability. This allows us to continually learn new knowledge. Later, after review, the knowledge can become part of the platform’s ontology, and that can be used across the domain, so the reasoning engine will become smarter.
● Initially, Mind Expression will allow for a simple, curated troubleshooting manual (a bite-size manual) when bootstrapping the platform. We will be able to handle more and more complex definitions as the platform advances in its models of comprehension.
● The entire process, either when the developer designs a conversational AI, or when the conversational AI converses with end-users, is completely transparent. Therefore, it is quick and easy to be able to trace what has been happening, especially if and when it goes wrong, and even when it is functioning correctly.
● Further, the troubleshooting capability in the telecommunication domain will expand to other domains that need such conversation models with end-users. Also, it can be adapted to handle booking/reservation services such as hotel, airline, restaurant, etc. We can evolve both horizontally and vertically; our platform is backed by a natural language reasoning engine that acts as a mind. All we will need is domain knowledge in the form of ontologies.
● There will be neither decision tree nor explicit process flows required when designing a conversational AI with Mind Expression, so the conversational AI designer needs only to concentrate on the actual customer support content rather than building a complex decision tree. This is due to our cognitive model, which deals fundamentally with meaning. Hence, an end-user will feel more like he or she is conversing with a real human agent.
● If the platform does not understand some words or phrases, the reasoning engine will infer what the meaning might be. This is to be confirmed by the end-user or conversational AI designer. The confirmed knowledge will then immediately be usable because of our adaptive learning capability. This allows us to continually learn new knowledge. Later, after review, the knowledge can become part of the platform’s ontology, and that can be used across the domain, so the reasoning engine will become smarter.
● Initially, Mind Expression will allow for a simple, curated troubleshooting manual (a bite-size manual) when bootstrapping the platform. We will be able to handle more and more complex definitions as the platform advances in its models of comprehension.
● The entire process, either when the developer designs a conversational AI, or when the conversational AI converses with end-users, is completely transparent. Therefore, it is quick and easy to be able to trace what has been happening, especially if and when it goes wrong, and even when it is functioning correctly.
● Further, the troubleshooting capability in the telecommunication domain will expand to other domains that need such conversation models with end-users. Also, it can be adapted to handle booking/reservation services such as hotel, airline, restaurant, etc. We can evolve both horizontally and vertically; our platform is backed by a natural language reasoning engine that acts as a mind. All we will need is domain knowledge in the form of ontologies.
Competitors
- There are no direct competitors that offer a cognitive model in creating conversational AI as infrastructure. There exist other solutions (“chatbots”) that are machine learning based statistical training platforms, among them (with market share): Dialogflow (60.53%), Amazon Lex (19.04%), IBM Watson Assistant (9.08%), and Chatfuel (5.80%).
Advantages or differentiators
● Transparency
- Mind AI understands the user's intentions based on meaning through reasoning, not mathematical probability-based intent classification. Understanding the sentences’ logical meaning, the entire process of reasoning from the user's input to the classification of logical meaning is transparent.
● Intent Classification without a Subject & Disambiguation
- Mind AI solves the problem with abductive reasoning, semantic distance and user verification. This is because the models are composed of sentences of logical meaning and is not based on the intent model. For example, if the composed problems are “TV problem” and “Internet problem”, the logical forms that can have a “problem” are “TV” and “Internet”. If they want “Internet”, Mind AI will understand what “problem” is specifically being solved by verifying with the user.
● Context Hopping
- As it is mentioned in the adaptive learning below, Mind AI is not a model in which conversations are made only according to the flow by creating a flow based on intent. Switching between contexts is made based on the Context (lower left node) within which our Canonical model already exists.
● Adaptive Learning
- The method of Mind AI is that we understand a written manual in a logical form, creating logical structures between these contents, and processes it through reasoning. This is not the way chatbot developers or designers configure flow, but a how a user problem is naturally, logically configured. Enter the written manual. Mind AI's inference engine converts the manual into Canonical structures to grasp the meaning and outlines the logic through inference.
● Natural Language Reasoning
- Many AI companies, including IBM Watson, do natural language processing, but Mind AI is the first in natural language reasoning. This is a logical reasoning method that accurately understands all contexts in a completely different way from the approaches of existing AI companies. They try to understand by inferring the approximate meaning by applying patterns or rules.
● Augmented Topological Network
- When a network is formed by applying the semantically fragile structure of the existing symbolic AI to the Canonical structure of Mind AI, different words and contexts with the same meaning exist on the network with the same semantics.
● Linear, Qualitative Process
- It is not a method of grasping an approximate context through a tagging technique for a specific word, but a continuous, logical reasoning that relates meaning to meaning according to the Canonical network structure.
Human-understandable Logic
- Because the Canonical structure of Mind AI takes a form that humans can understand, the causal relationship between cause and effect can be traced back through the Canonical network structure.
● Don’t Need Tons of Data
- Mind AI accepts education in human language, and constructs an ontology with single data statements, and performs logical reasoning based on that data. Therefore, unlike existing AI, it does not require vast amounts of training.
● Any Language
- Since Mind AI's model grasps meaning based on symbols, it can be applied to all languages used by humans.
● Universal
- Unlike existing AI models that can only be applied to specific domains and problems, Mind AI's model can be applied to all areas that humans can understand.
● New Symbolic Paradigm
- It combines the strengths of both Symbolic AI and Neural Network AI, and no other company has this structure as the means by which to represent artificial intelligence. We are unique in this space.
- Mind AI understands the user's intentions based on meaning through reasoning, not mathematical probability-based intent classification. Understanding the sentences’ logical meaning, the entire process of reasoning from the user's input to the classification of logical meaning is transparent.
● Intent Classification without a Subject & Disambiguation
- Mind AI solves the problem with abductive reasoning, semantic distance and user verification. This is because the models are composed of sentences of logical meaning and is not based on the intent model. For example, if the composed problems are “TV problem” and “Internet problem”, the logical forms that can have a “problem” are “TV” and “Internet”. If they want “Internet”, Mind AI will understand what “problem” is specifically being solved by verifying with the user.
● Context Hopping
- As it is mentioned in the adaptive learning below, Mind AI is not a model in which conversations are made only according to the flow by creating a flow based on intent. Switching between contexts is made based on the Context (lower left node) within which our Canonical model already exists.
● Adaptive Learning
- The method of Mind AI is that we understand a written manual in a logical form, creating logical structures between these contents, and processes it through reasoning. This is not the way chatbot developers or designers configure flow, but a how a user problem is naturally, logically configured. Enter the written manual. Mind AI's inference engine converts the manual into Canonical structures to grasp the meaning and outlines the logic through inference.
● Natural Language Reasoning
- Many AI companies, including IBM Watson, do natural language processing, but Mind AI is the first in natural language reasoning. This is a logical reasoning method that accurately understands all contexts in a completely different way from the approaches of existing AI companies. They try to understand by inferring the approximate meaning by applying patterns or rules.
● Augmented Topological Network
- When a network is formed by applying the semantically fragile structure of the existing symbolic AI to the Canonical structure of Mind AI, different words and contexts with the same meaning exist on the network with the same semantics.
● Linear, Qualitative Process
- It is not a method of grasping an approximate context through a tagging technique for a specific word, but a continuous, logical reasoning that relates meaning to meaning according to the Canonical network structure.
Human-understandable Logic
- Because the Canonical structure of Mind AI takes a form that humans can understand, the causal relationship between cause and effect can be traced back through the Canonical network structure.
● Don’t Need Tons of Data
- Mind AI accepts education in human language, and constructs an ontology with single data statements, and performs logical reasoning based on that data. Therefore, unlike existing AI, it does not require vast amounts of training.
● Any Language
- Since Mind AI's model grasps meaning based on symbols, it can be applied to all languages used by humans.
● Universal
- Unlike existing AI models that can only be applied to specific domains and problems, Mind AI's model can be applied to all areas that humans can understand.
● New Symbolic Paradigm
- It combines the strengths of both Symbolic AI and Neural Network AI, and no other company has this structure as the means by which to represent artificial intelligence. We are unique in this space.
Finance
- Mind AI is to provide an API (Application Programming Interface) where we charge based on the usage of that API.
- We will charge $0.0004 USD per transaction, which is 1/5 of the pricing model for Google’s Dialogflow.
- As per our revenue projection, when we have 2 telecom partners, the revenue will be above $3M for the first year. And following year, when we have 6 more partners, the revenue will become more than $27M that year. As we get more partners, the revenue will grow exponentially.
- We will charge $0.0004 USD per transaction, which is 1/5 of the pricing model for Google’s Dialogflow.
- As per our revenue projection, when we have 2 telecom partners, the revenue will be above $3M for the first year. And following year, when we have 6 more partners, the revenue will become more than $27M that year. As we get more partners, the revenue will grow exponentially.
Business model
● Future Objectives
- Mind AI aims to become the AI CPU in industries (B2B) where AI is used. Mind AI is to provide an API (Application Programming Interface) to charge based on the usage of that API. This is similar to the method of charging according to data and voice usage in mobile communication companies.
- The service through artificial intelligence that understands human language and makes logical inferences has an immeasurable possible value and can be applied to all conceptual areas. When so applied, the usage of the API has unlimited potential. For example, when applied to a customer support chatbot, all inquiries from customers are answered through the API of Mind AI, growing the usage of the API exponentially as the number of customers who converse increases.
- In the Mind Expression Conversational AI Infrastructure, the billing policy and the form of the service that can be used are different according to the service subscription type. In addition to the API billing, service subscription billing is available.
● Short Term (B2B Business)
- Provide Mind Expression Infrastructure for constructing Conversational AI to Telecommunication companies (T-Mobile, Verizon, AT&T, Vodafone, etc.) in English.
- After the True Mobile project, we will develop Thai and Korean packages for the Mind Expression and Conversational Reasoning AI Engine, and provide Mind Expression Infrastructure to other partners and industries in Thailand and Korea.
- Expand into the education space from kindergarten to Grade 12 level to provide AI tutors with total transparency of operation. Our immediate goal is to solve word problems in mathematics and science and establish Conversational Canonical and Ontology infrastructure for the relevant industry.
● Long Term (B2B & B2C Business)
- Expand the Conversational Canonical model and ontology for common sense, expanding to more industries (B2B) and products that anyone can configure as Conversational AI (B2C).
- As the ontology increases and its contents deepen, the ability of the reasoning engine improves, and ontology acquisition is crowdsourced through incentive programs.
- By delivering API as open source, it can be easily used in the global development community to build an ecosystem that can build new platforms with innovative ideas.
- Mind AI aims to become the AI CPU in industries (B2B) where AI is used. Mind AI is to provide an API (Application Programming Interface) to charge based on the usage of that API. This is similar to the method of charging according to data and voice usage in mobile communication companies.
- The service through artificial intelligence that understands human language and makes logical inferences has an immeasurable possible value and can be applied to all conceptual areas. When so applied, the usage of the API has unlimited potential. For example, when applied to a customer support chatbot, all inquiries from customers are answered through the API of Mind AI, growing the usage of the API exponentially as the number of customers who converse increases.
- In the Mind Expression Conversational AI Infrastructure, the billing policy and the form of the service that can be used are different according to the service subscription type. In addition to the API billing, service subscription billing is available.
● Short Term (B2B Business)
- Provide Mind Expression Infrastructure for constructing Conversational AI to Telecommunication companies (T-Mobile, Verizon, AT&T, Vodafone, etc.) in English.
- After the True Mobile project, we will develop Thai and Korean packages for the Mind Expression and Conversational Reasoning AI Engine, and provide Mind Expression Infrastructure to other partners and industries in Thailand and Korea.
- Expand into the education space from kindergarten to Grade 12 level to provide AI tutors with total transparency of operation. Our immediate goal is to solve word problems in mathematics and science and establish Conversational Canonical and Ontology infrastructure for the relevant industry.
● Long Term (B2B & B2C Business)
- Expand the Conversational Canonical model and ontology for common sense, expanding to more industries (B2B) and products that anyone can configure as Conversational AI (B2C).
- As the ontology increases and its contents deepen, the ability of the reasoning engine improves, and ontology acquisition is crowdsourced through incentive programs.
- By delivering API as open source, it can be easily used in the global development community to build an ecosystem that can build new platforms with innovative ideas.
Money will be spent on
To complete and release our product, a Conversational AI Infrastructure by 2nd quarter of 2021.
Team or Management
Risks
We are a natural language reasoning company, and we feature transparency, cognitive capability, and we deal with actual meaning. Because of our unique implementation, we can provide very powerful AI tools. However, there exist many companies who care nothing about these features. Their only care being something that works well enough. These organizations will not see us as a competitive product, and they will not become our customers. Also, we only deal with natural language reasoning via text, and so we do not deal with text-to-speech, speech-to-text, statistical analysis, or computer vision. Any company that wishes these components from us as well will also not become our customers.
Won the competition and other awards
● '2020 Korea Innovative Company Grand Prize' in the AI category, September 2020,
● '2020 Grand Prize (2020 Brilliant People & Brand Awards)', January 2020
● '2019 Korea Innovative Company Grand Prize' in the AI category, December 2019
● '2019 Korea Artificial Intelligence Award', November 2019
● '2020 Grand Prize (2020 Brilliant People & Brand Awards)', January 2020
● '2019 Korea Innovative Company Grand Prize' in the AI category, December 2019
● '2019 Korea Artificial Intelligence Award', November 2019
Invention/Patent
● DATA PROCESSING METHOD AND DEVICE USING ARTIFICIAL INTELLIGENCE (인공 지능을 이용한 데이터 처리 방법 및 장치) | PCT/KR2018/008419 / 2018.07.01
● DATA PROCESSING METHOD AND DEVICE USING ARTIFICIAL INTELLIGENCE | US-2020-0159753-A1 / 2020.01.23
● DATA PROCESSING METHOD AND DEVICE USING ARTIFICIAL INTELLIGENCE (인공 지능을 이용한 데이터 처리 방법 및 장치) | 특허출원 제20-7005512호 / 2020.02.25
● DATA PROCESSING METHOD AND DEVICE USING ARTIFICIAL INTELLIGENCE | US-2020-0159753-A1 / 2020.01.23
● DATA PROCESSING METHOD AND DEVICE USING ARTIFICIAL INTELLIGENCE (인공 지능을 이용한 데이터 처리 방법 및 장치) | 특허출원 제20-7005512호 / 2020.02.25
Product Video
Presentation
Sign in/Sign up