lessons learned creating a virtual personality - botscamp - ondrisek

Download Lessons learned creating  a virtual personality - Botscamp - Ondrisek

If you can't read please download the document

Upload: barbara-ondrisek

Post on 16-Apr-2017

305 views

Category:

Internet


0 download

TRANSCRIPT

Lessons learned creating
a virtual personality

Hi guys!

My name is Dr. Barbara Ondrisek. Today Im going to give a talk about personalities of chatbots.

Background

Ive been working as a freelancer and consultant for more than 15 years for different mostly big companies. I have a strong Backend dev background and in the last couple of years Ive been working as a Senior Backend developer with a disposedness to full stack and mobile. I also love to play around with other technologies and as Facebook announced on their F8 conference mid of April to open up their Messenger platform to bots I was eager to try their API!So I created one of the very first chatbots on Facebook and definitely Austrias first Facebook Messenger and Skype chat bot: Mica, the Hipster Cat Bot.

Chatbots

Over the past few months a lot of useful chatbots and virtual assistants were released . Their functionality and complexity gets better and better day by day. Now you can book a flight, order food via chatbot or get advice by virtual shopping assistants just to mention a few!

Mica, the Hipster Cat Bot

For instance one of my chatbots is Mica, the Hipster Cat Bot, a chatbot that helps you discover the best venues. While developing it, I figured out that in contrast to an app or a website, creating a chatbot is far more difficult than expected in terms of user experience (UX). The graphical user interface (UI) of chatbots is still very limited and restricted by the messenger platform in comparison to colorful dashboards, headers or menus all available for web or app. So designing a chatbot means to integrate all the functionality in a new type of user interface: the conversation.

Talk to me!

Concentrating on the dialog, especially the text you send to the customer, could be an advantage since you dont distract people with a fancy user interface. On the other hand you are not able to seduce and hook people with a colorful or decent UI.

The whole functionality of a service is built in a conversational matter, that naturally reveals itself during a dialog, going over the steps through a decision tree. This way most of the features are hidden inside of the chatbot. Some platforms offer buttons, menus or other UI elements that enhance the interaction, but within a conversation normally its hard to get an overview of all the features a chatbot can offer. Some chatbots remind of the old text based adventure games like Zork: You have to talk to the chatbot to get to know it and a possible path it can lead to.

Tell me more about your father!

Historically speaking the very first chatbot was developed at the MIT AI Lab by the computer scientist Joseph Weizenbaum in the mid1960s. This bot, ELIZA, did not really had a sparkling personality, but was able to simulate a Rogerian psychiatrist. With this project Weizenbaum wanted to find out, how natural language can be used in human-computer-interfaces. ELIZA was programmed to analyze the input of keywords and to respond with a number of canned phrases containing therapeutical language. So, when someone said Oh, last night I had a dream about my father, ELIZA responded with Tell me more about your father! since she scanned the text input for keywords and rephrased typically a new question to keep the person talking.

Another early example for chatbots is Parry, a bot mimicking a person with paranoid schizophrenia, in the early 1970s. Later in 1972, ELIZA and Parry were thrown together for a bot-to-bot conversation, which Parry quickly steered down a rabbit hole of corrupt horse racing gambling.

Add value add character

One of my most favorite chatbot characters is Poncho, the weather cat. He is a cute and funny cartoon cat cute, that loves pizza. The purpose of this character is to inform and entertain the users. You can even play games with him!The personality of a chatbot is the key to the commonness using the service. It really is the secret sauce! Otherwise you could use another interface of a service like a website, an app, a hotline or a widget. But why do people use Poncho? Because they can actually talk to him and be entertained.Our usage statistics of Mica show clearly that from the total 550k incoming requests only 120k resulted in the number one use case (discover venues); thats approximately 20% of all the requests. What do the other 80% want? They want to talk and chitchat! Depending on the messenger platform (we launched on Facebook, Skype, Telegram, Kik, Beta on WeChat, Line, and Viber) they want to small talk, ask whether shes a cat or a bot, request help, amuse themselves with cat pictures or exchange meows with the cat. The most used phrases are greetings with Hi as the top used word.

It's the Tone That Makes The Music

Decide on what sort of tone and approach you want to use to communicate with your customers. Mica for instance is based on my cat, my real cat, at home. Mica is friendly, always hungry, sometimes a bit intrusive. She also likes hip and trendy coffee shops, she likes to share funny cat pictures and doesn't like water. Although the hipster mustache implies a male character, Mica is a female cat.However. at the beginning Mica was chosen (as a joke) as the official Head of Communication to be addressed for PR requests with another project of mine: LIKE A HIPSTER. When Facebook announced to open their messenger platform to bot, I wanted to translate the LIKE A HIPSTER service to a chatbot version. I also decided to use a cat as the character for a dialog, because I think every single cat owner would love to actually talk to their cat! Also, people tend to be disturbed when they cannot distinguish whether they are chatting with a person or a computer program. Thus, I chose Mica as the perfect dialog partner.With Mica I learned a lot about user experience-design. Most people just try out which combinations or commands a chatbot understands and where its capabilities end. For instance, there is still no common standard established for chatbots, although most understand hi, start and help.

Emotional reactions to
conversations to bots

People, especially teenagers or seniors, tend to text with bots more. Studies show that seniors tend to chat with Siri when they are lonely; the same happens with bots that are capable of conversation.Another interesting aspect is that people react emotionally to bots. They love them and tell the bot this, or they hate them and start using foul language. Based on this data, you could create personality profiles of bot users. For instance depending on the type of language, grade of used slang or number of grammar and spelling mistakes one could assume the age of a user.

What could possibly go wrong?

Microsoft just recently released Zo, a teenager girl bot on Kik, that uses machine learning and it seams they kind of did not learn from Tay, their former twitter chatbot, which turned out to become a holocaust denier.Users try out whats possible and tend to text with bots like no one is listening. This was also shown very early, because when Weizenbaum was building ELIZA, he realized that one tester felt ashamed when he entered the room, saying, Sorry, but Im currently talking to ELIZA! Could you please leave!Some people start to swear just for fun or send unrelated questions such as Where is Barbara?. Since Mica also has quite a narrow aim (to find hip venues worldwide), people are usually quickly satisfied when trying a chat.People also get upset easily when a bot does not meet their expectation. In order not to frustrate the users, Mica sends a funny cat picture to entertain if she doesnt find venues in the vicinity or does not understand the request. Mica is polite and responds accordingly with basic conversation skills.

Key Learnings