In the early 1930's, Lev Vygotsky formulated the Zone of Proximal Development. In short, he theorised that a child could learn how to execute a range of tasks effectively, within a guidance framework, but not yet fully independent.
His work led to what we today call "scaffolding". It is the framework we create as parents, teachers and employers to allow people to learn, experiment and ultimately become independent of outside support. "That's a proper grounded laaitjie (young person)", the older person at the family BBQ would say.
The importance of grounding in a generative AI solution has a lot of analogy to Vygotsky's work. This week the First Digital AI Research team put this to the test. Using the First Technology web site as data, and CVs from some of our account managers, we configured a chatbot to respond to a client's questions.
We compared two versions side by side. The ungrounded chatbot answered in a generic, surface-level way, with no real anchor to who we are or what we do. The grounded chatbot, given a clear grounding statement and our own data, responded with contextual follow-up questions and a defined call to action, pointing the client to the right account manager by location and industry.
Not only does a well architected grounding statement give the system a narrative, but it is crucial to anchor your AI system in clear, responsible principles, gives it a defined call to action, and ensures it delivers a consistent, trustworthy, and human-centered experience.
