Artificial
Intelligence Market worth 16.06 Billion USD by 2022 http://www.marketsandmarkets.com/PressReleases/artificial-intelligence.asp
Since around 1998 I was creating chatbots and building up libraries of AIML Artificial Intelligence Markup Language, XML dialect used in natural language software agents. Example sites: pandorabots.com and alicebot.org
More recently I've been testing out some of the newer solutions like api.ai, they make it very simple to build conversational interfaces using Natural Language.
One of the biggest problems with Machine leaning is the Training System Required to build a large enough data set to handle variations in Conversation. So most of my thoughts have focused on producing Input based learning from large data souces (e.g. books) Or Learning based on Conversation and Articles.
Recently my wife and I had a few friends staying with us and all have kids at differnent ages. It's fasinating to think about how we learn language. We learn based on Input and Experience (I'd love to see a dual language learning app based on aipoly.com). So mapping Input and Output to a Software model would need some kind of Experience Engine.
To produce Output / Results / Reply a system would need to understand the Input based on Experience.
I've also been thinking about a simple architecture / sitemap of those systems
Back to the real world I made a note of some applications from ai.api:
- Small Talk - Make small talk
- Booking - Make reservations at restaurants or hotel …
- Weather - Get weather information
- Wisdom - Ask general knowledge questions
- Flight Schedules - Get flight schedules and statuses
- News - Read news
- Call - Call a number, contact, or venue
- Messages - Compose and read messages
- Email - Compose and read emails
- Media - Control music/video players and radio
- User Name & Agent Name - Change names of users or agents
- Web Search - Perform web searches
- Web Browsing - Open websites
- Apps - Open, close, download, remove, search, …
- Time & Dates - Work with time and date
- Calculator - Perform basic calculations and conversion …
- Calendar - Work with calendar
- Reminders and Notifications - Work with reminders and notifications
- Manage App - Open, close, or update the app
- Maps - Search in maps
- Navigation - Navigate to places
- Points of Interest - Search for venues
- Device Control - Control device’s settings
- Units and Formats - Change default formats for units
- Translate - Translate text to various languages
- Language Selection - Switch between different languages
- Shopping - Search for things to buy
- Notes - Work with notes
- Tasks - Work with tasks
- Social Networks - Work with Twitter and Facebook
- Events - Search for events
- Taxi Search - Request a taxi
- TV Listings - Get TV listings
- Smart Home - Control smart devices
- Sports - Get sports scores, stats, schedules, etc …
- Finance - Get stock prices and market reports
- Authentication - Log in/out or sign up for online service …
- Learning - Train the agent to understand new commands …
- Nicknames - Work with nicknames
- Contacts Search - Search for a contact
- Images - Search for images
- Traffic - Request a traffic information
- Maps Shortcuts - Work with maps shortcuts
Currently my own two faves:
1 Automated conversations with people looking to buy Insurance.
Competition in this space:
- http://www.synergist.io/ negotiation layer
- https://artificiallawyer.com/2017/01/11/legal-tech-pioneer-synergist-reinvents-contract-creation/
- https://mty.ai/ Service to train artificial intelligence engines
- http://www.ibm.com/watson/how-to-build-a-chatbot/ AI Service provider
- https://cloud.google.com/products/machine-learning/ AI Service provider
- https://aws.amazon.com/amazon-ai AI Service provider
2 Semi-automated platform creates written article content.
Competition in this space:
- https://automatedinsights.com/
- http://www.arria.com/
- https://www.narrativescience.com/
- http://www.articoolo.com/
- http://www.emmaai.com/
Building an Experience Engine is key
Experience could be built from Relationship of entities.
Relationship of entities could be done with Content Context. So we need to build up huge datasets of related entities which should be anything and everything e.g. Language, Time , Date, Grammar, Letter, Word, Word, Phrase, Sentence, Paragraph, Question, Answer, Intent, Symbols, Abbreviations, Database of Examples, Fruit, Company, Numbers, Stock market symbols, Currency codes, Currency symbols, ISO codes, Maths equations, Plant names, Animals, Birds, Piano notes, Guitar notes, Voltage, Frequency, Artists, Architects, Websites, Planets, Navigation, Coordinates, Molecules, Particles, Elements, Countries, People names, Boys names, Girls names, Music bands, Music artists, Clouds
For example: Take a short list of Currency symbols (£ GBP, $ USD) for any Real Intelligence there should be a huge number of related of entities to Build an Experience.
For example: USD relates to many entities e.g. Language=English, Time=1.02pm, Date=24 Feb 2017, Grammar=English grammar, Letter=U,S,D,USD, Word=USD, Abbreviations=USD, Currency codes, Currency symbols.
In summary. Talking with Software developers in a past they always had concerns about database storage and CPU processing limits but we're reaching a point now where those limitations will not be a concern. Yes Apple, Google, Amazon, Microsoft are all building conversational agents but I think there'll be room for more. (Note a side note I still don't like the term "AI" cause it's based on Real Intelligence).