Enabling adoption of AI through increased trust
Trusting technology is to understand its performance
Enabling adoption of AI through increased trust
The digital transformation has been going on for some time, with technology development accelerating at a rapid pace. The last two years with the pandemic have re-enforced this exponential trend, where we have seen fast adoption of new digital technologies and completely new ways of working.
The opportunities offered by the digital transformation goes well beyond enabling Teams meetings and more efficient work processes.
-Fundamentally, we start to see that part of society increasingly rely on decisions and activities made by machines and models that are now fuelled with live data.
This will enable completely new ways of addressing current challenges, but more importantly, these technologies will be cornerstones in addressing all the important transformations we face, such as the energy ransition, climate change and the transition to more sustainability.
Data-driven, model-driven and AI approaches will play a key part and we will also see new business models emerge to even further the value from these technologies.
The starting point for trusting the technology is understanding its performance.
Programme Director at DNV
Important aspects to consider ensuring trustworthy, explainable and safe AI
Better assurance on the data used for training, validation of representativeness of the scenarios experienced during use, and the assessment of robustness towards uncertainty.
Furthermore, by combining domain knowledge encoded into physical models with the datadriven model we may achieve better explainability and performance. This is particularly important for safety-critical applications of AI.
But the value creation potential and performance of AI are not the only decisive factors in its adoption. Use of AI also introduces new emerging risks and new trust gaps. In addition to the challenges of using conventional modelbased decisions, AI introduces something more fundamental as far as trust is concerned, as we do not fully understand the inner workings of the technology.
It becomes important then to know not only the technical accuracy of the technology, but also how it was developed: by whom, with what intentions, how it was trained, potential biases in the data, etc. This calls for trust beyond the technology itself.
To address some of these issues, we see that AI is increasingly regulated.
The EU is proposing regulations on the use of AI, whereby high-risk use will be strongly regulated, and certain uses even banned. This is necessary to ensure that the deployment of AI can be trusted and that we can benefit fully from scaling the technology.