How are chatbots born?
After Facebook announcement of the Messenger API for bots and Microsoft racist Chatbot Tay incident I got curious. How hard is it to create a smart chat bot today, how do you do it and how far away are we from achieving the Turinger test?
I created my first chatbots about 25 years ago inspired by Racter at the time. It was designed based on long lists of questions and conditions with the purpose of taking the user through a particular journey. E.g. the conversation looked something like this.
User: Good morning
Bot: Hi! What is your name?
User: Ana
Bot: Nice to meet you Ana. Where are you from?
User: Madrid
Bot: What is Madrid like?
By asking questions and assuming a response it's relatively easy to make the bot seem smart even if it's not.
So how have bots improved in the past 25 years and how do you design one now? Firstly, some of the principles still used today are actually 20 years old. The big change is more recently
Some of the top bots available today include:
- ALICE - from the mid 90s and inspired the movie Her
- Cleverbot - created in the late 90s and learns from human input
- MitSuku - award winning bot launched around 2013 based on AIML
- Rose - won the Loebner Prize four of the last six years
- Tay by Microsoft - artificial intelligence bot launched on Twitter in 2016
There are 100s of other open source and commercial bots you can download, customise and compile including MegaHAL, ELIZA and Cleverscript. We are also aware that Google, Facebook and IBM have ongoing projects in this space.
How smart are the chatbots today?
A couple of the chatbots listed here have passed the Turinger test in one way or another. However, opinions vary about the validity of the Turinger test as a method to indicate progress for human interaction by computers. Some recent predictions forecast that we are about 10-15 years away from chatbots that we will not be able to distinguish from humans.
So how do you create your own chatbot?
After testing various available platforms we choose Pandorabot which is based on AIML? (Artificial Intelligence Markup Language). Pandorabot is originally based on the work on ALICE and the programming language LISP.
1. Sign up for an account on Pandorabot playground
2. ?Download the Pandorabot version of ALICE called Rosie from Github
3. ?Customise it to your needs as much as you want
4. Upload to Pandorabot, publish and try it out
5. Register for a Pandora developer account to access the API and launch your bot
To make the chat bot publicly avai?lable requires a bit more work to create a chat window or integrate with e.g. Facebook Messenger. As a starting point Pandorabot provides libraries and tools for this.
As an example of the results you can try the DMI bot launched with about 4 hours of work. You can ask anything but as DMI is a mobility solution provider the answers will be steered towards this. E.g. ask what's the cost of a mobile app?
What's the meaning of life?
So will your chatbot be able to tell you the meaning of life? Not yet according to recent progress by Google's chat research.
As usual please comment/reply below with your thoughts and questions.
About: Magnus Jern is a serial entrepreneur, mobile envangelist, published writer and speaker at conferences around the world. He helps big and small companies succeed in mobile currently serving as President of DMI International (acquired Golden Gekko in 2013). Get in touch with him through Linkedin.
P.S. DMI is hiring Project Mangers, Solution Architects, Data Scientists, Business Developers and other roles in Barcelona and London. Visit the career section.
Tech Blogger at Techolog
8 年Bot tech I feel is still leagues behind human level interactions. But existing model of integrating chatbots to one platform, like in case of Slack and its bots like AceBot, Statsbot, HumbleBot are on boarding more and more users when it comes to using bots. This will create demands from users for more useful effective bots, which in turn will result in user-oriented innovations - An iterative process that will benefit both users and bot makers alike.
Performance Recovery in a Post-Lean & Six Sigma World | Manufacturing Intelligence | Decision Science | Complex Businesses & Operations
8 年@ Magnus Jern, Chatbots were created in the late 60's by Douglas Engelbart and became commercial about three decades later. We used to call them teleprinters...I used them when entering the job market to make real-time flight bookings and different other tasks. Interestingly enough, it took another 30 years before chatbots evolved from hardware + phone/telex setups to digital interfaces...
Entrepreneur on a Mission to reduce human impact on the planet
8 年Pandorabots, Inc.: Thanks for the support!
Senior Director at DMI (Digital Management, Inc.)
8 年Magnus, Great post. Microsoft threw new born Tay in to noisy Twitter world during an election year. It had no chance in figuring out meaning of life. Could have started in meditation blogs!!