Designing a Conversational Intelligent Bot which can cook
5 likes519 views
The document discusses the design and development of a conversational cooking bot named Samantha, emphasizing the importance of intent and trigger keywords for effective natural language processing. It outlines the integration of speech processing, data intelligence, and decision-making systems to enhance the bot's functionalities. The author invites participation in a social experiment to train the bot, offering resources for further engagement.
Designing a Conversational Intelligent Bot which can cook
- 1. >< nextprevious Designing a BOT Kaushik Das twitter.com/theKaushik
- 2. >< nextprevious What is a bot?
- 3. >< nextprevious No we are not discussing that…
- 4. >< nextprevious Conversational Intelegent Bot Lets discuss how to build a simple
- 5. For the sake of an use case lets say we want to create a bot who can cook food for us. Say we have a robot with all functioning mechanical parts and all we need to do now is build a brain for the robot. If you think of cooking its all about formula and set of processes (Ignore Gordon Ramsay for a moment and his cooking instincts). So if you think of it in a simple world all this brain needs to do is understand what i am asking to cook and use the formula to do set of activity to cook the item. USE CASE >< nextprevious Kaushik’s Future Kitchen
- 6. >< nextprevious Introducing Samantha Good cook and loves to chat
- 8. >< nextprevious Decide Purpose ‣ Molecular design approach, think as one bot is good at solving one problem. ‣ Multipurpose bots are basically a cluster of multiple bots COOKING What to cook? For how many people? How to cook? What groceries i need? How do i serve? cook bot should do you need another bot
- 9. >< nextprevious Purpose is your Intent
- 10. >< nextprevious What to cook?
- 11. >< nextprevious Decide Trigger Words ‣ Imagine this as keyword which will decide certain flow ‣ Your bot need to know that keyword Cake means a Food item and not shit ‣ Need training ‣ Exponentially increases estimate for cascading functionalities, cake vs dung cake ‣ Analogy would be think it as a new born baby and you are trying to teach her how the world works
- 12. >< nextprevious Trigger Keywords is your Entities
- 13. >< nextprevious
- 14. >< nextprevious Integration ‣ This is where you will take it to next level of intelligence. ‣ Before we jump to how integration works for our bot; lets discuss some fundamentals for this process.
- 15. >< nextprevious High level view QUERY PROCESS DECIDE INTENT I am hungry? With NLP decide its intent. Do you want some pizza? I can cook something
- 16. >< nextprevious High level view QUERY PROCESS DECIDE INTENT SPEECH PROCESSING DATA INTELLEGENCE
- 17. >< nextprevious So where can we start? SPEECH PROCESSING DATA INTELLEGENCE SPEECH PROCESSING: • Natural language Processing (NLP) • API.AI , FB Deep Text DATA: • Data is knowledge. • IBM Data, Geo Data, Alexa Skill Kit , Wolfram Alpha INTELLEGENCE: • Decision making, Probability. • Rule based AI for simple tasks, Baye’s Theorem of Probability
- 18. >< nextprevious so back to Integration
- 19. >< nextprevious Integrating with NLP & .. ‣ Most new gen AI platforms gives NLP OOB. ‣ Multi lingual NLP is still not up to the mark In future we will have more complex task available just using an on-off switch.
- 20. >< nextprevious Integrating Complex Services ‣ Understand the activity it needs to achieve, multiple end point vs single end point. ‣ Time taken to execute, Synchronous vs asynchronous ‣ WTF Webhooks and timeouts (Sample webhook url below for you to try) Approach 1 APP API.AI CUSTOME AI SERVICE Approach 2 APP CUSTOME AI SERVICE API.AI Git Hub: https://goo.gl/2Xaijb
- 21. >< nextprevious Integration is your Fulfilment
- 22. >< nextprevious At intent level
- 24. >< nextprevious Lets do a social experiment ( Dec 8 to Dec 31, 2017 ) ‣ Lets train the baby cook!!! ‣ Ask her every possible thing. ‣ Every weekend I will re - train her. http://sntech.xyz/bot/cook?talk=who are you? (Sorry if the url is down you can watch the demo here: https://www.youtube.com/ watch?v=V_us1GSz4TI )
- 25. >< nextprevious Source code ‣ Reach out to me at twitter.com/theKaushik if you need the source code to play around.
- 28. >< nextprevious In 1997, Watson precursor, IBM’s Deep Blue beat the reigning chess grand master Garry Karparov in a famous man-vs- machine match. After machines repeated their victories in a few more matches human largely lost interest https://www.youtube.com/v/ NJarxpYyoFI&start=06&end=109
- 29. >< nextprevious But thats not end of the story…. If Kasparov had the same instance access to massive database of all possible moves that Deep Blue had he could have performed better.. If this database tool was fair for an AI, why not for a human? To pursue the idea Kasparov pioneered the concept of Man+Machine matches like mixed martial art where players can use what ever technique they want.
- 30. >< nextprevious SAMANTHA 2.0 THANK YOU http://twitter.com/theKaushik