The importance of trust
Trust in your product or application is essential in user acceptance. Often, the level of success is directly correlated with the level of trust the user has. If your users don’t trust the technology, then how likely are they to engage with it? And if they do, are they likely to engage as you expect?
Trust is built when users feel they understand the decisions made by applications - i.e. the reasoning behind a decision is upfront and clear. It’s vital that the factors which led to those decisions being made by machine learning (ML), are clearly outlined. The process of tracing and explaining the decision is what we know as Explainable Artificial Intelligence (XAI).
Explainable AI in practice
To put XAI into perspective, imagine a scenario where you’ve landed at a London airport, hired a taxi, and halfway through the journey the taxi driver deviated from the familiar route. You’d likely feel anxious. Shortly after, you’d probably ask the taxi driver why the regular route wasn’t taken. If the taxi driver said the regular route was congested due to construction, and that this is the quickest route, your mind would instantly be put at ease.
Now consider the same scenario with an AI driven taxi deviating from its defined route. You’d certainly want to know the reason the self-driving car made that decision. With no reasoning or explanation, you’re going to be uncomfortable.
As we start to see automated AI solutions such as loan approval or credit limit decisions, XAI comes into its own. It not only provides the reason for the credit decisions, but also the counter measures to ensure the financial situation and loan application is right for the customer. Explainability also helps developers identify any biases or shortfalls, and ensures that the application works as expected.
For AI to be used in real-world situations, the factors used to make the decision need to be shared with the user.
Clear instructions
AI cannot think for itself. It requires instructions from data scientists and engineers in the form of data. Examples include numbers, text, code, images and much more. The larger the data set, the better it will perform. AI will do what it is programmed to do. So, if the machine is given poor data to begin with, it will learn potentially biased or assumed behaviours, leading to unwanted side effects.
Implementing specific techniques that ensure every decision made during an ML process can be traced and explained, provides much greater trust in the application. It also allows both the developer and the end user to understand the decisioning.
Summary
Artificial intelligence is used everywhere. And why wouldn’t it be? It’s proven to provide huge efficiencies and therefore, saves organisations money. But can it be used more ethically, and improve the user experience by gaining their trust first? The path to a true and complete XAI has just begun. However, we must initiate questions around the ethics of artificial intelligence and where its faults lie.
Examining data and scrutinizing the potential for bias before any system is developed, is one of the best approaches to creating trustworthy AI.
AI is here to stay, so it’s our collective responsibility to ensure that while developing and using it, it creates transparent, unbiased, and adaptive solutions.
Sopra Steria’s leading AI solutions
Sopra Steria delivers a comprehensive suite of innovative data-leading solutions enabling organisations to realise significant business value from data. The Data Insight and Intelligence team of experts at Sopra Steria helps organisations across Europe to leverage data effectively to make strategic data-driven decisions.
Sopra Steria’s data-leading solutions