Context-based AI versus loose generative AI: the difference that matters

Many schools, sports clubs and businesses have been experimenting with generative AI in recent years – mostly in the form of chatbots based on broad, publicly available knowledge. The results are often both impressive and sobering: the answers sound knowledgeable, but have little to do with the organisation’s own learning platform, training programme or customer data.

This is precisely where SteP-met comes in: using context-based AI rather than loosely generative AI.

The problem with generic generative AI

Generative AI models are designed to formulate plausible responses. However, without access to the actual information relating to a school, a club or a company, recommendations are inevitably general in nature.

For example, an AI does not automatically know a pupil’s learning progress, a team’s performance data or a company’s internal product and support information.

This distinction is particularly crucial when working with learning, performance, training or customer data. General knowledge alone does not provide institutional context.

SteP-met therefore adopts an approach in which the AI is not the sole focus. What is crucial is the connection between the available data, appropriate integration and the specific task at hand.

What ‘context-based’ means at SteP-met

Rather than allowing AI to operate loosely on general knowledge, SteP-met links it specifically to existing and authorised information from the relevant application context.

The basic principle follows five steps:

Data source

The starting point is existing information. Depending on the area of application, this may be learning, performance, support or organisational data.

Secure interface

The data source is connected via a suitable integration. Options include, for example, LTI 1.3, LMS integrations, APIs, data providers or customised interfaces.

SteP-met is not tied to a single platform. The possible connections to existing systems are explained in more detail on the internal ‘Integration’ page.

Context Processing

Not all available information needs to be provided to the AI. Instead, the information that is relevant and authorised for the defined use case is selected and processed.

AI Analysis

The AI then works with this specific context. The focus is therefore on context-based evaluation rather than purely general generation.

Specific outcome

Depending on the area of application, the context provided can result in analyses, recommendations, personalised learning support, training challenges or assistance functions.

In a nutshell:

Data & systems → secure integration → context → AI → practical support

One principle, four areas of application

This fundamental principle can be applied to various areas.

In education, learning progress, tasks and learning pathways can serve as a context for individualised learning support, feedback, analyses and recommendations. Existing learning platforms can be integrated via suitable interfaces.

In corporate training, the focus is on existing training data, competency models and learning pathways. These can form the basis for more personalised professional development support.

In sport, data from sports or performance data providers can be incorporated. Based on match, training and performance data, analyses, coaching recommendations or personalised training challenges are possible, for example.

In corporate and customer support, knowledge bases, products, internal processes, as well as support and customer contexts can be utilised to provide employees with more targeted support.

The specific possibilities within these four areas are described in more detail on the internal ‘Solutions’ page.

Do you already have relevant data and systems and would like to find out whether they could be used to create a meaningful AI use case? A clearly defined pilot project allows a specific use case to be tested in a targeted manner, without turning it into an open-ended experiment.

Why context-based AI matters for organisations

At SteP-met, context-based AI does not mean giving an AI indiscriminate access to as much information as possible.

On the contrary: the architecture is designed for controlled data processing. Which information is provided as context depends on the specific use case and the relevant authorisations.

Where possible and appropriate for the function in question, information can be processed in the AI context in a pseudonymised form or without real names. The specific implementation is guided by the data protection, security and governance requirements of the relevant institution or company.

Further information on this approach can be found on the internal ‘Security & IP’ page.

SteP-met also relies on Microsoft Azure as a scalable operating and deployment environment for the operation of its systems.

The crucial question is not what AI knows in general

For schools, sports clubs and businesses, therefore, it is less important how convincingly a general-purpose chatbot can phrase things.

The more relevant question is:

What support can AI provide when it receives the correct, authorised information from our actual context?

This is precisely where the difference lies between a standalone generative application and the SteP-met approach.

Existing data and systems do not become a secondary consideration for AI. They form its foundation.

Context-based AI shifts the focus from general generation to a specific task within the actual context of use.

SteP-met connects existing and authorised information to AI via suitable interfaces. This can result in targeted analyses, recommendations and assistance functions for education, corporate training, sport, as well as business and customer support.

The principle remains straightforward:

Data & systems → secure integration → context → AI → specific support

Would you like to check whether a specific use case is suitable for your institution or organisation?

Enquire about a pilot project with SteP-met. Together, we will examine a clearly defined use case using existing data and systems – with a specified data source, user group, function and duration.