The opportunities and challenges for AI in 2022
Ian Hall, head of AI & analytics at Capgemini UK, gives his top AI predictions for 2022 and the challenges that need to be overcome
The Covid-19 pandemic created a business environment of uncertainty, with a need for continuous flexibility and innovation. Data and analytics have become more crucial than ever to business strategy and digital transformation, with AI widely used to support hybrid and remote working.
The awareness of its true potential is prompting a spike in investment and reinvestment in AI and machine learning, as businesses learn how to use AI to speed up processes and gain market share over rivals. There are many drivers shaping the evolution of AI over the next 12 months, and the journey is not without its challenges.
Improving patient care using federated learning
The pandemic has radically transformed healthcare; AI-driven solutions have been given a much bigger role and have helped to improve safety, efficiency and outcomes. But these changes happened almost overnight, and the challenge of legacy systems, data silos, poor UX and vendor lock-in needed to be addressed quickly.
Now the focus will need to be on maintaining patient data privacy by applying federated learning (a type of machine learning) with hardware-enhanced security, in order to enable the development of more accurate automated medical diagnosis models using much larger sample sizes. These techniques facilitate data sharing in an anonymised way, ensuring private data never leaves a given device or healthcare system.
Cross-organisation global datasets will deliver significantly improved results: in one case the federated model performed 18 per cent better than the best locally developed model. In research settings, federated learning enables health research participants to remain in control of their data while still contributing to advancing science - two things that used to be mutually exclusive.
The high importance and benefit of such collaborations will drive use of these concepts in 2022, notably Intel Xeon Scalable processors and Nvidia graphics hardware, across both the public and private sector.
Digital twins - the fastest route to optimal design
The concept of data-driven digital twins isn't new, but like so many AI applications it has been accelerated by the pandemic, and it now represents even more value for businesses across a range of sectors - most notably manufacturing, with aerospace and automotive industries really leveraging it to their advantage.
Physical prototypes in manufacturing typically involve significant time and money, but digital twin technology enables multiple adjustments from a computer - making it possible to pinpoint the most optimal design and to run numerous simulations to measure response to varying conditions.
For 2022, there will be a strong focus on the evolution of digital twins - moving from just basic modelling of devices (and lowering prototype costs), to being able to model the capabilities, specific attributes, and properties as a more accurate replication of the physical device. It's also likely that there will be a stronger focus on the use of digital twins to model the relationship between multiple devices, and even the factories that produce them.
Given the huge investments planned in the automotive and (aero)space industries, coupled with the accelerating shift towards high-powered engines fuelled by electricity, there is also room to use digital twin technology to understand, measure and calibrate emissions and impacts before manufacturing even begins.
Explainable artificial intelligence (XAI)
The success of machine learning predictably led to an explosion of AI applications - all with an impressive suite of autonomous abilities to learn, perceive, decide and act. Now, there is a demand for explainability: processes and methods that allow human users to understand, trust and manage the results and output created by the algorithms of their artificially intelligent ‘partners'.
We predict that 2022 will be the year for a focus on more explainable AI models, which also maintain a high level of learning performance/prediction accuracy. These high XAI systems, with better transparency in algorithms and decomposability, will keep humans in the loop because they are able to explain their rationale, strengths and weaknesses as well as future performance - so they can be used to assist decisions and will also have a faster route to production. They also tick the box for the wider public, who have struggled to understand and trust AI both in terms of capabilities and ethics.
Smart factories interoperating at scale
Bringing AI to create smarter factories isn't a new concept, but many businesses have failed to deploy at scale. But now in 2022, more mature businesses have the IT infrastructure and change management in place to apply AI at scale, using IoT devices to share data and achieve interoperability, gaining huge cost optimisation benefits.
Implementing IoT devices enables data to be shared across multiple sites in real-time, enabling any interdependencies to be monitored and autonomously managed.
In a fully connected and integrated plant, all systems can share data across multiple sites in real-time, meaning that more interdependencies can be monitored and autonomously managed. If an issue is identified with a specific raw material, then all the processes that depend upon it can be halted automatically, regardless of location.
Sustainable AI: optimised logistic flows help decarbonisation
With sustainability high on the agenda, soon everything will be seen through a sustainability lens, including AI. One way we might see AI helping organisations to reduce their carbon footprint in the year ahead is through route optimisation and fleet management throughout the supply chain. This will be driven by companies feeling duty bound to take responsibility not just for their direct impact, but also for the emissions of their extended supply chain of partners and vendors.
So, in 2022, expect collaboration across this global spread of stakeholders in order to optimise logistics and make a greater contribution towards decarbonisation. AI will be used to interpret data from electricity or natural gas invoices, and we'll see the rise of central sustainability tracking models, which will form the basis for footprint reduction projects.
Ian Hall is head of AI & analytics at Capgemini UK