AI benefits are being delayed by lack of skills and data
Image credit: Dreamstime
Despite the clear business value of adopting enterprise artificial intelligence, asset-intensive, process-based industries are significantly behind other sectors.
Artificial intelligence has become a well-recognised and frequently used buzzword. However, it means different things in different situations and as such can be tricky to define. Whilst most people think of AI as a technology in its own right, it’s actually more of a general term used to refer to a number of different technologies that enable systems to act intelligently.
When it comes to business applications, AI can support intelligent functionality by helping a system to sense, understand, perform and learn. A system that has been trained using machine learning or deep learning can assess how to act by analysing data, rather than relying on prescriptive, hard-coded actions. The resulting agility and responsiveness mean that quality, accuracy and overall performance are dramatically improved – and this is what makes the system truly intelligent.
In the current climate and with uncertain times ahead, a lot of enterprises are looking at how they can rapidly adapt and accelerate their digital-transformation strategy. As remote collaboration, operational agility and autonomous production become ever more critical to business continuity, AI is at the forefront of many executives’ minds.
What sets AI apart from other automation technologies is its ability to learn and adapt. In an industrial environment, it can have a significant impact on business performance by dramatically reducing manual labour, quickly identifying patterns in large amounts of data and analysing and extracting features from both structured and unstructured datasets. Most importantly, it can learn from these tasks and improve over time.
There are three principal ways in which machine learning can be deployed: supervised, unsupervised and reinforcement learning. Supervised learning uses pre-organised training data and feedback from humans to learn the relationship of given inputs to a given output. This method is useful if the input data and predicted behaviour type is already classified, but the algorithm needs to be applied to multiple different datasets. Unsupervised learning doesn’t require any pre-defined labels in the data – no output variables need to be pre-identified and the algorithm can analyse input data to find patterns and make classifications.
Reinforcement learning allows the system to learn to perform a task by trial and error. In essence, this technique is based on rewards and punishments, with the overall aim of maximising rewards and minimising punishments in the feedback received for its actions. This approach is particularly useful when there isn’t a lot of training data, it’s difficult to identify the desired outcome and this is the only real way to interact with and learn from the data.
In an increasingly digital world, organisations are looking to AI to revolutionise more than just their technology; it’s redefining business processes as a whole. From pioneering innovation to everyday customer service, AI is transforming the business landscape and defining this paradigm shift is the key to understanding enterprise AI.
The ‘constellation of AI’ paradigm - introduced in Paul R Daugherty and H James Wilson’s book Human + Machine: Reimagining Work in the Age of AI - is one framework that explains the application of AI on an enterprise level.
Using this framework, enterprise AI can be viewed across three levels. The first identifies the ‘why’ and the ‘what’ – the business applications that use data to provide greater value to its stakeholders. The second identifies the suite of AI capabilities that can be leveraged to power these application. Finally, the third looks at the ‘how’ – which machine-learning methods can deliver that capability?
Using this framework, the complexities of AI-based business applications can be simplified and fully assessed to allow enterprises to build an all-inclusive AI programme, analyse and define the business value for each AI initiative, and determine the basic requirements that would drive a successful AI program and justify investment.
While there is clear business value in adopting enterprise AI, asset-intensive, process-based industries are significantly behind other sectors when it comes to implementation. This is largely due to the need for new skills and a lack of quality data. According to Gartner, 56 per cent of enterprise leaders feel they need updated skills to accomplish AI-enabled tasks and 34 per cent say that poor data quality is a key concern. Also, 42 per cent of Gartner respondents said they don’t fully understand the benefits of AI or the implied return on investment (RoI) due to the challenge of quantifying them.
By 2024, RoI will be measured by quantifying AI investments and linking them to specific KPIs, giving the future of enterprise AI a clear direction of travel in terms of measurement and real-world statistics. By establishing a common understanding of AI’s enterprise value and setting out clear guidance for business application, organisations can capitalise on the simple ‘constellation’ framework to implement successful AI projects, now and in the future.
Adi Pendyala is a senior director at Aspen Technology.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.