AI errors are often undetected until the final product is launched, or are sometimes deliberately concealed to mislead or impersonate human work.
At Lanware, we have witnessed a surge in client demand for AI solutions, receiving a wide range of requests for projects, services, or evaluations of tools and applications. As a Microsoft Cloud specialist, in particular, we have seen a lot of interest in Microsoft Copilot and its features, but we are also encountering AI applications in more customised and finance-specific domains.
As AI systems and services become more popular within financial services, it is important to be able to spot and avoid big mistakes that can be generated by this powerful new technology.
Why AI can sometimes fail
Many AI failures result from employees violating the company's Acceptable Use Policy (AUP) or the lack of such a policy for AI. If you want to find out more about an AUP policy and need a starter template for your business, you can download one here: Have you written your acceptable use policy for AI in financial services?
The main thing to remember about AI systems and services is that although they may seem to have some "sense", they are unlikely to have much "common sense". Common sense is defined as "the ability to make good judgements based on the practical experience of life". These judgements are usually learned through interacting with other people and the environment, which AI systems have not done because they have only used data and algorithms to produce an output.
Make sure the results are accurate, especially if you are depending on them.
AI systems are not infallible, and you need to verify the output. A lawyer in the US who was suing an airline for a client in a standard personal injury case, faced possible sanctions because the cases he cited were fabricated. This is because he used ChatGPT, a generative AI platform, to find similar cases, but the platform produced fictitious cases from its training data making the claim invalid.
It's therefore crucial to have a clear vision of the desired outcome, especially when using AI image generators due to their inherent simplicity often lacking the intuitive understanding we take for granted. This can lead to unexpected and inaccurate results which can cause problems. To help circumvent potential issues upon release, it’s essential to have a good grasp of how the final image should appear.
For example, check out these "cricket match" images from an AI generator which showcases inaccuracies such as four stumps instead of three, several balls on the pitch at once, and incorrect pitches.
Now observe the qwerty keyboard layout generated below. At first glance, it appears an accurate image resembling the conventional layout, but on closer inspection, it is a ‘made up’ keyboard and it would be a struggle to use.
Ensure your AI system doesn’t create undesirable biases
AI is now playing a crucial role in many business sectors, notably in HR and recruitment to make processes more efficient. For example, creating an automated system to sift through the skills needed for new positions or as part of a pre-qualification phase is beneficial. However, it's critical to prevent any undue bias so it’s important to align HR policies and train systems to reflect your company's stance against discrimination. Also, try and ensure your supply chain is compliant with your policies.
This interesting article details how Amazon removed a recruitment system from its business as it was favouring male candidates.
Verification is key!
In summary, AI systems and services can offer many benefits for businesses, but they also require careful oversight and verification. Users should not blindly trust the output of AI but rather check for accuracy, relevance, and bias. AI systems may lack the common sense and practical experience that humans have and therefore may produce unexpected or inaccurate results. It is important to have a clear vision of the desired outcome, and to align the AI system with the company's policies and values.
