Demystifying AI: Part 3
The final part of this series looks at what other factors a business needs to consider before going live with an AI project.
In Part 2, we took a look at what tasks a Data Scientist spends their time doing, and also dug a little deeper into Machine Learning algorithms.
However, we cannot talk about AI and not look at a couple of important concepts a business will need to consider before putting any AI project live.
Bias
Fairness has been right at the heart of the conversation amongst the recent advancement of AI. Models must be built carefully to ensure no groups suffer unfair outcomes. Bias can be an aspect to the following areas:
- Training data
- Algorithmic
- Cognitive
It’s crucial that your model does not skew unfairly towards a particular group. A simple example would be offering women a lower credit limit than man, purely because of their gender.
Governance
Ideally, you’ll have a governance framework that gives your organisation full control over what, where and how AI is being deployed. Items to consider would be:
- Sign off/approval process
- Oversight of all projects
- Checks for changes in data
In fact, the EU AI Act came into force on 1 August 2024. The legislation applies a risk-based approach to regulating AI, with different obligations applied according to the perceived level of risk. I strongly recommend you familiarize yourself with the Act.
Model Explainability
Coming up with an AI model is almost the easy bit. But being able to understand and explain why it is producing the results it does, is vital. Your colleagues, and indeed you customers, might want to know, what data was used, what key data points influenced the outcome, and of course, if a colleague leaves or you need to take over a project, having your work document is vital. So, do consider:
- What features are important
- What-If? Analysis
- Model documentation
We’ve covered quite a few themes in this three part series which I hope gives you a better understanding of what AI is and it’s core concepts. If you are keen to learn more, I do hope you consider subscribing to the blog and continue your journey.
In Summary
The underlying data is still vital
Whatever you're trying to build, starting with clean, accurate and well randomised data is a crucial foundation for building any AI project.
It’s not all about GenAI
Generative AI might be the new kid on the block, but, AI goes back over half a century. As an organisation you need to be clear on what your goals are and ensure you select the right tools for the right job.
Governance is non-negotiable
AI projects can yield huge benefits to an organisation. But their power requires that every organisation should have a well established governance framework with the relevant checks and monitoring systems in place.
Comments ()