From automation of complex processes to analysis of subtle patterns to aid planning, cognitive technology can be a powerful business tool. However, the pace of innovation has been accompanied by concerns over the risks that new and emerging technologies pose, creating a demand for ways to understand, mitigate and control these risks.

Cognitive technologies, or 'thinking' technologies, fall within a broad category that includes algorithms, robotic process automation, machine learning, natural language processing and natural language generation, reaching into the realm of artificial intelligence (AI). Whether or not these technologies are truly thinking or intelligent is a question of philosophy. For our purposes, these are systems that appear to think or act intelligently, in an analogous way to how humans act, regardless of the ultimate nature of the processes that lead to those actions.

Cognitive technologies offer powerful incentives for companies. They can automate tasks from the routine (robotic process automation) to the complex and abstract (machine learning and AI). They can detect subtle patterns in data and make predictions about what might be coming down the line. Cognitive technology is bringing automation to business processes previously thought un-automatable, such as reviewing contracts, classifying images or detecting inappropriate content.

One area attracting great interest from researchers and businesses alike is machine learning, which uses a variety of techniques to create optimised programs to solve a wide range of problems and tasks. The strength of machine learning is in its ability to learn from experience, rather than having to be explicitly taught the rules by a human expert. This can not only increase the efficiency and ease of creating cognitive technology, but also enables the tackling of open-ended problems for which writing rules might be impossible, such as image classification.

As the technology becomes increasingly accessible and affordable, more companies are experimenting with it. However, this has also drawn concerns over the potential risks, not only for the companies using these technologies, but to wider stakeholders and society. Such concerns include, for example, fears of bias, misuse and even wasted effort.

Regulation of the area is naturally slow to adapt, and guidance and practice may often lag behind fast-paced technology changes.

Key terms

This section provides a brief overview of the key terms if you’re less familiar with them, before we look at the risks of implementation and ways to overcome them, including through design principles, controls and assurance.               

Artificial intelligence (AI)	Technology that has been programmed to replicate human behaviour such as participating in natural-seeming dialogue, making decisions, understanding complexities within content and substituting for people in tasks. It can be used around the clock and in processing large volumes of information quickly.
Algorithm	A series of instructions to carry out a task. Algorithms can be created either explicitly by humans or other computer processes, or developed through trial-and-error processes such as machine learning.
Cognitive capture solution	A program that applies rules to a set of data taken from images of text and uses natural language processing (NLP) and machine learning to replicate human ‘reading’ by recognising the context of a document, such as an HR form.
Machine learning	The ability of a computational device to learn from large volumes of training data and improve upon a given task without having been explicitly programmed to do so.
Neural networks	A system of artificial neurons whose performance is inspired by the biological networks of the brain and which recognises that information can be categorised according to specifications.
Natural language generation	Natural language generation (NLG) refers to systems that generate human-seeming speech or written language.
Natural language processing	Natural language processing (NLP) is an analysis of speech patterns and written language by computer to gain meaningful information.
Optical character recognition	Reading handwritten or typed text electronically to create machine-encoded text that can be used in a different format (for example, text captured from a photograph).
Robotic process automation	A solution to replicate a business process, designed to do a task that would otherwise be performed manually. It is programmed to follow “if this, then that” instructions. It runs with other programs rather than replacing them.

For a fuller exploration of the impact of these technologies on the profession, see our earlier report Artificial intelligence and the future of accountancy at www.icaew.com/ai.

Next, we discuss the key risk areas that have arisen from our research into cognitive technology.