Artificial Intelligence (AI) is a branch of computer science and mathematics that enables computers to analyse data, identify patterns, make decisions, and learn from experience.
Unlike traditional software, which follows fixed instructions, AI systems can improve their performance over time through a process known as machine learning (ML).
A simple example:
Imagine you show a computer a thousand pictures of 500-krona and 100-krona banknotes.
At first, the computer guesses wrong, but each time you tell it the correct answer, it adjusts itself a little.
After many attempts, it learns to recognise the difference on its own, even with new pictures it has never seen before. That’s exactly how AI learns – through practice and feedback, just like we humans do.
The basic principle of AI
AI aims to replicate aspects of human cognition.
Instead of using hard-coded rules, AI relies on artificial neural networks – mathematical structures made up of thousands or even millions of interconnected “nodes” (neurons).
Each node receives input, applies a weight (importance value), processes the data, and passes the result to the next layer of nodes.
During training, these weights are continuously adjusted so that the model’s predictions become more accurate.
This process, called backpropagation, allows the model to learn from errors and refine itself.
Training an AI system requires:
Large datasets (training data) – examples from which the model learns patterns.
Optimisation algorithms – mathematical procedures that minimise prediction errors.
Computational power (GPU/TPU) – high-performance processors capable of running billions of calculations in parallel.
The result is a system that can recognise patterns, make predictions, and even generate new content based on what it has learned.
Different types of AI
AI is an umbrella term covering several distinct approaches, each used for different purposes in finance, industry, and everyday life:
Rule-based AI
– Follows predefined logical rules (“if X, then Y”).
– Example: early credit-scoring systems or basic customer-service bots.Machine Learning (ML)
– Learns relationships from data to predict outcomes or classify information.
– Example: fraud detection, credit-risk assessment.Deep Learning
– Uses multi-layer neural networks capable of identifying very complex patterns.
– Example: speech recognition, image analysis, natural-language understanding.Generative AI (GenAI)
– Can create new content such as text, code, images, or speech.
– Powered by Large Language Models (LLMs) like OpenAI’s GPT models, Google Gemini, Anthropic Claude, and Meta LLaMA.
LLMs are trained on enormous text datasets to learn the statistical likelihood of words occurring together.
When you ask a question, the model generates a response word by word, based on probability – but enhanced by internal contextual representations known as embeddings.
How an AI model is trained – step by step
Data collection
Vast quantities of raw data (text, numbers, images, or transactions) are gathered.Pre-processing
The data is cleaned, normalised, and anonymised to protect privacy.Training
The model learns relationships between variables by adjusting its internal parameters (weights).Validation and testing
The model is evaluated on unseen data to ensure it generalises correctly and does not overfit to training data.Fine-tuning
For large language models, fine-tuning is used to align the model with specific tasks or domains – for example, Nordiska’s assistant Casey is fine-tuned on Nordiska’s product and account information, not on customer data.Deployment
Once validated, the model is implemented in production under strict monitoring and governance.
How AI is used in banking and finance
Within financial services, AI is widely used to:
Improve customer experience through chatbots and personalised digital service.
Detect fraud and suspicious activity (AML/KYC monitoring).
Forecast financial trends using predictive models.
Automate back-office processes like document handling and risk reporting.
Enhance cybersecurity by recognising abnormal network patterns in real time.
Security, ethics, and regulation
All AI systems at Nordiska are operated within strict regulatory and ethical frameworks:
GDPR (General Data Protection Regulation)
Swedish Financial Supervisory Authority (Finansinspektionen): guidelines on IT security and risk management are followed.
EU AI Act (upcoming): Nordiska already aligns with its key principles of transparency, human control, and accountability.
Internal Ethical Standards: Nordiska applies the principles of explainability, fairness, and responsibility in all AI-related work.
Example – AI in practice
Imagine an AI system designed to detect suspicious transactions:
It is trained on historical data to learn what “normal” activity looks like.
It then analyses new transactions and compares them to learned patterns.
When it finds a deviation beyond an acceptable threshold, it flags it for manual review.
The same principle applies to language models. The system recognises linguistic patterns and determines the most probable word or response. AI is also utilised to identify and respond to common customer enquiries via our digital assistant, Casey, delivering faster, more accessible, and personalised service while maintaining the highest standards of security and data protection.
In short:
AI works by learning patterns in data using neural networks and algorithms.
Nordiska uses AI to enhance security, efficiency, and customer service – never to replace human decision-making.
All AI use is governed by GDPR, Finansinspektionen’s regulations, and the forthcoming EU AI Act.
Nordiska monitors the development of global AI models (OpenAI, Google, Anthropic, Meta, and others) and adopts only secure, compliant solutions.