God Bot

Staak collaborated with the Financial Times and agency partners to create a thought-provoking campaign featuring an AI 'digital deity' that users could engage with to explore life's profound questions and receive answers on spirituality.

Our task was to infuse the interactions with a human conversational flow while maintaining the gravitas of communicating with an all-knowing 'higher being.'

The responses were designed to be prompt, tailored to user interactions, and infused with humor to challenge and entertain users.

Technical Challenges

Identifying Spirituality.

The chatbot needed to be able identify any question it was asked by the user under the category of ’Spirituality’.

Identifying Other Topics Beyond Spirituality.

Not only did the chatbot need to be able to identify Spirituality, it needed to be able to identify topics within questions that were not about ’Spirituality’ and respond with appropriate and convincing answers to show to the use that it has understood their intention.

Controlling Conversation Flow.

To add to that requirement, the chatbot conversation flow had to follow a certain guided structure and wasn’t allowed to flow freely indefinitely. The bot needed to pose an opening statement to the user, respond to the user’s questions one way if the question was ‘on topic’ and another if it was ‘off topic’. The bot had to always steer users back to asking questions on the topic of spirituality. The conversation flow was allowed to cycle in this way for a set number of rounds before the bot would wrap up the conversation and serve the relevant marketing message back to the user

Finding a chatbot platform flexible enough for our needs

In order to identify the topic of spirituality we first began looking various Natural Language Processing services that would be able to correctly identify topics, sentiment and categorise our input.

We tested various available third party NLP platforms with a sample set of user inputs and their results, however there wasn’t one platform that could give us the responses that we needed in order to make our app work. (Note this was several years prior to ChatGPT being released)

At the same time we were also exploring several available third par chatbot platforms and weighing up if they were a viable solution in order to deliver the project successfully.

The issues we faced with all of the chatbot platforms is that a ‘regular’ chatbot tends to have very specific boundaries which within it operates. For example a train ticket booking chatbot - it’s job is to book tickets on trains. It needs to do this in the shortest number of conversational transactions with the user as possible until it has all the criteria required to fulfil the task at hand. It doesn’t need to be conversational, or to understand any topics beyond the world of train tickets, it doesn’t need to be able to offer spiritual guidance or appear witty, all-knowing or engaging on that level. In summary we were unable to give a bot the broad knowledge that would be required to fulfil the demands of the project.

From here we began to look into the workings of an “Alice bot” and the programming language AIML (Artificial Intelligence Markup Language)

A.L.I.C.E. (Artificial Linguistic Internet Computer Entity), also referred to as Alicebot, or simply Alice, is a natural language processing chatterbot—a program that engages in a conversation with a human by applying some heuristical pattern matching rules to the human's input, and in its online form it also relies on a hidden third person.

We began to prototype out various bots using AIML as a language and the service Pandorabots in order to process it for us. Although we started to get some of the flexibly we required we still lacked certain features in order to make the project succeed as desired. The AIML language by itself was not flexible enough for what we needed, and nor was the Pandobots platform on its own.

From here we developed our own processor that would sit in-between the user and the Pandorbot platform. That way we were able to achieve the level of control we needed.

Our next big hurdle at that point was the ability to be able to design the conversation using some sort of tool-set, ideally a tool-set that could be used by a non technical administrator. There were no solutions commercial available that were a viable option for this. As our chatbot became more unique and custom so did the needs of the conversation creation tool.

The conversation tool actually then became the largest and most challenging part of the project. In order to build a huge knowledge base and understanding for the chatbot we built a backend that allowed for an administrator to be able to program the ability for the bot to understand any possible topic and respond accordingly without any technical knowledge. Another requirement of the CMS is that each section of the knowledge base needed to have approval status project stake holders across production and legal

The result was ‘Gabbabot’. A conversation design tool that utilises pattern recognition to deliver contextual conversation. Using conversational building blocks we can create a deep and contextualised knowledge base which can be exported to AIML (Articifical Intelligence Markup Language) for use in a chatbot.

Gabbabot allowed for conversations to broken down into several sections and sub sections

• Categories
  • Topics
    • Entities
      • Synonyms
• Constructors
  • Question Words
  • Auxiliary Words
  • Subjects
• Supports
  • Opening Statements
  • Broad Questions
  • Specific Questions
  • Worship Questions
  • Linguistic Deflections
  • Terminators
• Threads

To accompany "Gabbabot" we also developed a extensive testing, qa backend that would allow us to generate, review and rate hundreds of thousands of test conversations in development and train the algorithms accordingly.

GodBot was designed to work on multiple levels through desktop, mobile and interactive billboards. The out of home (OOH) GodBot experience had the ability to communicate in real-time using digital billboards, stopping passers-by in their tracks and disrupting the traditional one-way advertising.

We designed ‘hook’ messaging to entice users into a conversation and paired with the use of web sockets, we allowed ‘disciples’ direct access to GodBot’s wisdom. As the user’s point of contact, the mobile experience had to be familiar to the point of being intuitive, so we designed the interface to mirror popular messaging services but with the high-end and hypnotic ‘window dressing’ found elsewhere.

One of our challenges was to create video content for the big screen experiences. These would be placed in high footfall areas of London and New York, tempting passing users into communicating with GodBot via their phones.

Having enticed users into a conversation with GodBot, it was then our mission to extract intellectual or spiritual questions for the Almighty to answer. This was done with meticulous copywriting techniques which would ‘funnel’ an interaction into a desired area. Our custom built AI would then analyse the user’s input for sentiment, tone and context then process this against its internal knowledge base and answer appropriately providing bespoke experiences.