Part 4: Demystifying AI  

This post is part of the AI Apprenticeship series:

• Part 1: AI Apprenticeship 2024 @ DiploFoundation
• Part 2: Getting introduced to the invisible apprentice – AI
• Part 2.5: AI reinforcement learning vs human governance
• Part 3: Crafting AI – Building chatbots
• Part 4: Demystifying AI  
• Part 5: Is AI really that simple?
• Part 6: What string theory reveals about AI chat models
• Part 7: ‘Interpretability: From human language to DroidSpeak’
• Part 8: ‘Maths doesn’t hallucinate: Harnessing AI for governance and diplomacy’

By Dr Anita Lamprecht, supported by DiploAI and Gemini

In Week 4, we centred on comprehending the key aim of Diplo’s AI Apprenticeship online course: demystifying AI. For this purpose, we looked under the bonnet by building our own chatbots, exploring complex concepts and the functioning of AI systems, and, very importantly, reflecting on our own experiences with AI during the course.

We specifically took a closer look at retrieval-augmented generation (RAG) and vector databases. This is not my first course on AI, and usually, when it comes to explaining RAG and vector databases, it quickly becomes very technical and therefore challenging. Not at Diplo. It was truly inspirational to learn these concepts through their relationship with us, the humans in the system. The keyword is ‘cognitive proximity’, and the key statement that goes along with it is ‘knowledge is what defines us as individuals’.

Basically, we are looking at a very interesting combination of answering the ‘why’ and the ‘how’ behind human-centric AI, or humAInism, as we call it at Diplo. To prepare this blog post, I had to dive deeper into the knowledge base of Diplo and utilise our chatbot.

Let’s take a look at the meaning of cognitive proximity for AI.

Understanding cognitive proximity

Cognitive proximity refers to the logical and emotional closeness between individuals, which facilitates the exchange and preservation of tacit knowledge, i.e. the understanding we gain through experience that is hard to put into words (like knowing how to ride a bike or read a room). It represents the expertise and intuition we develop over time in a specific domain. This concept is crucial in the context of AI, as it highlights the need to maintain a human-centred approach in technological advancements.

AI systems, especially those using RAG and vector databases, benefit greatly from our tacit knowledge. This knowledge is embedded in the data we provide, the way we structure it, and through our prompts and feedback. This exchange, however, is not one-sided. By offering new perspectives and insights, AI can also influence and shape our own tacit understanding, leading to a dynamic interplay of knowledge and intuition.

At its core, cognitive proximity functions on three levels of cognition and communication: rationality (logos), values (ethos), and emotions (pathos). This triadic framework ensures that AI integration into organisations and businesses remains balanced and practical, creating an environment where human intuition and machine efficiency coexist harmoniously.

How to build a human-centric AI future

The integration of cognitive proximity, RAG, and vector databases represents a significant step towards a more human-centric AI future. By prioritising the logical and emotional closeness between humans and machines, we can ensure that, besides being efficient, AI systems need to be aligned with human values and needs.

Basically, a human-centric AI means that the output it generates is close to our own thinking. This is what we call cognitive proximity. RAG is key to achieving this, as it allows AI to tap into a vast network of information, similar to how our brains connect related ideas and pull (or retrieve) the prompted information. This network is stored in a vector database, which enables AI to quickly find and use the most relevant information to answer questions accurately and with context. Think of the vector database as a space filled with connected information, comparable to our human brain.

By customising this system with our own data (the dataset), priorities, and guidelines (the system prompts), we can shape AI to be an extension of our own intelligence, working in harmony with human values and goals.

Roles in the triad

Understanding the roles of humans and AI in this triadic framework of rationality (logos), values (ethos), and emotions (pathos) involves a complex interplay of cognitive abilities and ethical considerations. 

• Rationality (logos): Humans are great at reasoning and critical thinking, while AI excels at processing vast amounts of data to uncover hidden patterns. Together, they can process more complex information and make more informed decisions.
• Values (ethos): Humans are responsible for setting ethical guidelines, while AI reflects the values programmed into it. It’s important to ensure AI aligns with our ethical principles.
• Emotion (pathos): AI can analyse patterns in human language to understand and respond to emotions, but it doesn’t actually feel them. Think of it as AI possessing emotional knowledge, but lacking empathy.

Learning by doing

Writing about this complex interplay itself is demanding. For this reason, I used AI to help me compose this part of the text. I consulted the knowledge embedded in our own AI system, DiploAI. As this question is very philosophical, I chose DiploAI’s philosopher agent and then aligned the text with my own knowledge and writing style. Afterall, part of the AI Apprenticeship is to acquire the skills not just of building our own chatbots, but to learn how to best incorporate them into our everyday lives. 

The advantage of using our own system is that it is built upon the expert knowledge of more than 20 years of our organisation. In other words, it comprises the cognitive proximity and intuition of all of our colleagues, embedded in articles, blogs, and also our internal annotation system. Thanks to its very diligent design, the sources of the output are transparent through the provision of links to the actual source. This reduces the risk of unwanted random output (known as ‘hallucinations’) and allows me to use our chatbot as a reliable assistant for my research and publications in the spirit of my organisation, Diplo.

To finetune the tone of this specific paragraph, I used Gemini. Why? Because at the moment it is the easiest tool for this purpose. My convenience however comes at a price. The price is that I am giving away parts of my knowledge to Gemini, instead of keeping it within a system that is mine.  

My human takeaway

By customising AI systems with our own data and values, we can shape AI to be an extension of our own intelligence, working in harmony with human needs and goals. This is the essence of cognitive proximity – bridging the gap between human and artificial intelligence, allowing them to work together seamlessly, even without invasive neuro-technologies.

As we’ve seen, exploring the ‘why’ and the ‘how’ behind human-centric AI unlocks a deeper understanding of its potential and challenges, especially when it comes to the role of human intuition and the potential for AI to influence our tacit knowledge. This dynamic interplay between humans and AI raises important questions about the future of knowledge, identity, and collaboration.

Specifically, we need to consider how AI might reshape our understanding of ourselves, given its potential to access and process our tacit knowledge, especially aspects we may not be consciously aware of. After all, our knowledge is what defines us as individuals, organisations, and nations.

The AI Apprenticeship online course is part of the Diplo AI Campus programme.