The TFI framework: A simple socio-technical lens to assess AI initiatives

Last week, a manager reached out to me for advice on thinking through a technology project. They were being promised big efficiency gains from adopting a specific AI-based solution, but they are also committed to responsible AI adoption, and wanted a way of looking at this technology adoption decision beyond simple time or cost saving equations.

I suggested using the TFI framework. I find this deceptively simple framework really helpful when thinking about technology-based projects. Here is a quick overview of what the TFI is, and how it can be used.

The technical, formal and informal (TFI) framework

The TFI framework is based on the work of anthropologist Edward T. Hall. In the book “The Silent Language”, Hall argues that all humans operate on, and understand the world in, three levels: the technical, the formal and the informal. For instance, when it comes to learning:

• The technical level refers to the books, materials and artefacts that contain or give access to knowledge;
• The formal level refers to the rules (e.g., how to perform a certain maths task) and the processes (e.g., feedback or assessment) of teaching;
• The informal level refers to learning-associated behaviours that are modelled and observed, such as noticing how those that do well in the learning environment organise their notes, which actions are admired vs frowned upon by others, and so on.

The three levels vary from context to context, which explains why the same intervention may work in one case but fail in another. Or, in the manager’s case, why a particular AI solution may make sense for a company but not for another.

For instance, learners may have access to the same learning materials (technical level) and processes (formal level), but if they have observed different links between behaviours and academic success, in their respective contexts, they will follow a very different approach. One group may believe, through observation in their specific context, that the key to academic success is memorisation, and that’s what they will focus when preparing for an exam. However, another group which believes, based on observation in a different context, that the key to academic success is argumentation, will follow a very different exam preparation strategy.

Examples of application of the TFI framework

The TFI framework has been widely used to analyse or predict behaviours in relation to technology-based projects. For instance, Hamid Khobzi, Mohammad Sadegh Ramezani and I used it in our paper “Content Creators at a Crossroads with Decentralised Social Media”. But, before us, Aseel Alghafis and Ali Alkhalifah used it to explore children’s online behaviour, in the paper “Children’s Behavior on the Internet: Conceptualizing the Synergy of Privacy and Information Disclosure”. Another example is Lara S. G. Piccolo and Roberto Pereira’s exploration of technology design in the paper “Culture-based artefacts to inform ICT design: foundations and practice”.

So, how can the TFI framework be used to look at AI deployment?

At the technical level, the manager would need to understand the technology that they are thinking of investing on, so that they could assess whether there is indeed a good fit between the AI solution and the business problem.  While Artificial Intelligence (AI), Machine Learning (ML), Large Language Models (LLMs), and Generative AI are all related concepts, they have some differences in terms of functionality. Even within machine learning, supervised, unsupervised and reinforced machine learning work in different ways and, therefore, are best suited for different applications. They would also need to reflect on the datasets that they own or have access to, given that AI’s performance is dependent on access to good data.

At the formal level, the manager would need to reflect on the relevant rules and regulations for their industry, as well as their organisations’ policies and formal procedures. For example, financial services’ organisations have a legal requirement to justify their decisions with regards to customers (e.g., whether to give a loan), and, therefore, they need explainable algorithms and a human in the loop. They also have rules regarding when and how to give investment advice, which could rule out generative AI applications. Alternatively, the organisation might have certain commitments to environmental goals, and they would need to consider very carefully the carbon cost of using different AI products.

As for the informal level, the manager would need to consider the impact of AI on staff’s perceptions and loyalty to the organisation. As discussed in the paper “Power Negotiation on the Tango Dancefloor: The Adoption of AI in B2B Marketing”, they would also need to consider power issues – taking time to develop AI skills internally might be better for the organisation in the long-term, even if it is tempting to use a third-party solution in the short-term. And, last but not least, the manager would also need to understand customers’ attitudes towards the use of AI. Namely, if the use of AI is seen as an inferior good, the manager should not use it on customer facing tasks.

So, that’s the TFI framework in a nutshell. Do you think that the TFI framework can support your organisation’s approach to AI adoption? Feel free to reach out, if would like to know more.