As a visitor to this travelling exhibition, we'd like your help.

The "Science meets Parliaments/Science meets Regions" exhibition provides Knowledge4Policy with a unique opportunity to test an innovative way of delivering scientific knowledge to policymakers: an AI system, or 'bot', able to answer users' questions via a chat interface. 

We have therefore quickly put together a 'pilot bot' so you can test the idea and give us your feedback. 

How to Test our chatbot

It will only take you a few minutes to test the bot.

This pilot was developed for Facebook's Messenger chat platform, so you'll need a Messenger account. If you do have a Messenger account, you'll need to log into it before you test-drive our bot:

Test K4PBot 

Once on Messenger, click on 'Get Started' and the bot will guide you through the process of discovering Knowledge4Policy content matching your interests.

Please log off from Messenger before leaving this kiosk.

Don't forget your Feedback!

Remember, K4P Bot is primitive - there is no AI, for example, and sometimes the underlying platform simply doesn't work, so you'll need to Restart it (see image, below). These and other known issues are listed below.

could you imagine using a chatbot to get 'quick answers' like this? What sort of content should it focus on?

Further development could make it more powerful, but before we invest we need your perspective: Could you imagine using a chatbot to get 'quick answers' like this? What sort of content should it focus on? Should we bother with AI?  

K4P Bot therefore regularly asks you for Feedback. You can also give Feedback at any time by clicking the 'hamburger menu' to the left of the 'Type a message' text entry, and choosing 'More > Give Your Feedback':

Why a Chatbot?

Knowledge4Policy focuses on bridging the gap between policymakers  and scientists (see About Knowledge4Policy). While our launch phase focused on integrating several databases of knowledge into a single platform, our audience research shows we need more content and services to help policymakers use that scientific knowledge easily.

In particular, audience research showed, an 'Ask an Expert' feature would be popular with policymakers:

a common first reaction is not to ask What is known about a pressing issue, but Who knows about it. To seek the Scientist, in other words, rather than the Science
- Knowledge4Policy: what do policymakers want?

Clearly we will need systems and processes for handling such requests, such as Online Community Managers (OCMs) responsible for "fielding questions from policymakers: where possible answering them, and where necessary bringing together the right scientists to answer it".

A chatbot could theoretically support those Online Community Managers: answering some questions automatically, and passing better-informed questions onto the OCM.

Known issues, Next Steps

As we launch the pilot bot in early February, we already know of several ways it could be improved:

• Sometimes it just doesn't work! While we built this bot on a well-known platform, it sometimes simply doesn't bring back knowledge from the K4P database when it should. We're investigating. If it happens during your test, please Restart via the hamburger menu (see image, above);
• Guess the tag: the user must currently enter a keyword exactly corresponding to one of the ~1500 tags in K4P's thesaurus. Users, however, are not psychic - the bot should be able to work with any text;
• One word tags: while some of the terms in K4P's thesaurus are multi-word, the bot doesn't process them well;
• Refining the search using Related Tags: when a user wants to add a second 'Related Tag' to refine their search results (similar to using 'faceted search' on the K4P website), the bot does not always offer the 10 most used Related Tags, nor allow users to get the next 10 Related Tags, until they find the one they want.
• Presenting and Refining the search by Content Type: currently the bot returns knowledge ranging from videos to datasets. It should at least tell the user what sort of content each piece of knowledge is. It would probably be useful to allow the user to choose the content types they want. And it should possibly 'favour' some content types over others. 

If we receive sufficient positive feedback, we will continue developing the bot and report our findings both here and via our Medium Publication.