What is artificial intelligence?

CC is actively researching and developing breakthrough techniques in human-machine understanding and interaction. Our vision is to shape a new age of empathetic AI technology where machines ‘get us’ at a psychological level. They’ll infer our internal emotions, attention, personality and health to help us make useful decisions. It’s taking us rapidly towards the realm of hybrid human-machine workforces.

Our work in human-machine understanding is founded on new designs based on human centricity and explores the intersection of AI and psychology. By detecting, perceiving and interpreting changes in the human state, AI can support its human teammates by reducing overload, distraction and fatigue. This allows machines and humans to seamlessly adapt collectively to optimise team performance and minimise risk under changing operational conditions.

The positive business and societal implications are extraordinary.

Innovation that has great positive impact is also at risk of exploitation – and AI is no exception. The challenge now is to build trusted partnerships with AI-enabled machines and protect them from misuse. It is time to confront concerns over ethics, safety and regulation, worries about AI’s impact on jobs, as well as increased inequality and extreme fears of AI getting out of control.

CC is working to facilitate responsible, trustworthy AI through our AI assurance framework. It is designed to help our clients that their AI systems operate as intended and do not suffer from unintended consequences. Based on the foundations of explainable AI, transparency and security, the framework is designed to build trust in systems, dynamically test them, and identify gaps in how they work for remediation.

Read: AI assurance – protecting next-gen business innovation.

This is where AI applications are deployed on smart, connected devices – or endpoints – at the network edge. Generation-after-next AI enabled ecosystems will be increasingly dynamic and responsive. Pervasive intelligence will be distributed closer to data sources and AIoT (the artificial intelligence of things) to infer in real time, protect privacy, improve fraud detection and security while reducing latency and cost.

The big challenge is to enable responsiveness to short-term signal variations while simultaneously evolving functionality in response to long-term contextual variations. Current edge AI systems infer trends locally, but the AI models are trained at large, high-compute, high-power data centres before being deployed. The result is a response that is, by design, not current and probably increasingly inaccurate as trends evolve.

At CC, we are actively exploring new technologies that apply deep learning algorithms at the edge. This transformative method will fundamentally redefine our approaches to communications, manufacturing, emergency services and healthcare – enabling adaptive responses based on learning local trends and performing relevant interventions. This contrasts to the current ‘one-size-fits all’ reality that is trained on an increasingly inaccurate past.

Read: The next frontier is to bring AI – and learning – to the edge.

CC researchers and technologists are working to apply the benefits of deep learning, a subset of machine learning that emulates human performance on specific tasks at scale. Deep learning works by harnessing the power of multi-layered neural networks to learn insights from large quantities of data, classify them, predict outcomes, and even create new content (generative AI). An artificial neural network is a model inspired by how the brain works.

The basic unit of the network is a node that receives multiple inputs and generates an output. To configure the parameters that describe the operation of the network, it must learn the salient characteristics (label data) of data that represents the use case of interest. This learning is achieved through training, which is broadly categorised into supervised learning, unsupervised machine learning and reinforcement learning.

Training is typically data hungry and compute intensive, and data labelling is often manual and time consuming, so hybrid approaches like semi supervised learning aim to reduce the training effort by combining the strengths of different learning methods.