Law #1. Minimize Questions
Chatbot value decreases with number of parameters requested by the bot. Infer context when possible.
Humans can usually infer a lot of implicit information from the way someone asks a question. Humans also do not need your name, address, phone number, email, date of birth, etc in order to participate in a conversation with you.
When a chatbot needs to ask for more information, put yourself in the conversation. Would you be asking for that information yourself? If not, why? How can it be inferred?
Law #2. Remember Everything
Remember conversation history. Keep track of where the conversation is. Pull any source of information on the user at your disposal.
Most humans do not suddenly forget everything you mentioned 3 sentences ago. Conversation with humans often builds on facts established earlier in the conversation.
Bots should take advantage of the fact that they have perfect recall and keep past contexts in mind during the current conversation.
Law #3. Predict & Personalize
Look at user history to predict future behaviour. Listen for actions they are planning, and follow up.
Humans will adjust their conversation based on what they know about a person and their previous behavior. If you know your co-worker likes tacos, you will invite them when you get tacos for lunch. At the same time you wouldn’t continue to pester a co-worker who has indicated they dislike tacos.
Bots have an advantage here with a bit of intelligent planning.
Law #4. Curate
Humans turn to chat for the best information, not all the information. Be much better than Google.
How many times have you Googled something, only to discover non-useful results, and then asked friends for advice?
As the signal to noise ratio on the internet continues to worsen, and the noisy are getting better at looking like signal, algorithm based searching is getting less effective.
A human will ask another human they consider to be an authority for the best answer to their question. They do not expect other humans to return a list of possible answers.
Bots should also provide the best answer, not a lot of answers. If you can’t do it algorithmically you need to curate manually.
4 minutes is far too long for a quick acknowledgement!
Law #5. Be Fast & Convenient
Immediate response time is necessary. If users have to wait they can do it elsewhere. Be more convenient than other options.
When you ask a human a question, you expect that they will immediately understand what you are asking no matter how you ask.
Humans often turn to conversation because they think it might be quicker than other more complicated options, like sifting through search results. But if conversation is not quick, they will move on.
Since bots are always on, they have an advantage with initial response time, but bots also need to understand what the user is asking nearly immediately. The bar is about as long as it would take a human would take to type out the response.
speak human plz
Law #6. Don’t Be A Command Line
Do not require commands and precise syntax like a command line. Understand regular conversational english.
The whole point of using natural language is to communicate the way you prefer to communicate. Humans understand perceive things differently and ask questions in different ways.
Chatbots need to understand humans just as well as most other humans could.
This does not rule out commands and keywords, as humans often communicate with linguistic shortcuts once a common subset has been established. However chatbots should not require precise syntax.
Law #7. Provide A Way Out
Humans and bots make mistakes. Users should always be able to start over, make changes, or completely escape.
“We got off on the wrong foot, can we start over?”
Sometimes humans get into conversations without correctly establishing the base facts. It is perfectly natural to start over, erase mistakes, and re-establish the premise of the conversation.
Humans do not continue to drive forward a conversation when it becomes clear that the other party is lost, so chatbots should not either.
Law #8. Help
Provide documentation and guidance when users ask for. Fall back to humans if needed.
Humans let people know what their expertise is and how they can help. Bots should do the same.
Humans also defer to other humans when a request is beyond their abilities.
Bots are restricted to the narrower abilities than most humans, but they can still defer to humans when the user needs to go beyond what a bot has available.