Computational approaches for AI systems

When Charles Darwin was deciding whether to propose to his cousin Emma Wedgwood, the great naturalist grabbed a pencil and a few sheets of paper. He proceeded to jot down arguments for and against the nuptials under the headings “Marry” and “Not marry”. Finally, weighing the two lists against each other, Darwin concluded that a wife would probably provide better companionship than a dog, and wrote, at the bottom of the page, “Marry — Marry — Marry”, followed by the mathematical sign-off “QED”.

Stylized representation of technology blueprint by Reto Scheiwiller from Pixabay
Image by Reto Scheiwiller from Pixabay

Darwin designed his tongue-in-cheek cost-benefit analysis to help himself make a choice. In that respect, it was an algorithm, as are recipes, business processes and just about any other instructions that we use in our daily lives either to solve problems or to complete tasks. Nowadays, algorithms are programmed into devices to automate jobs that past generations had to do by hand. We call it artificial intelligence and it has moved into the mainstream.

Rapid advances in software and hardware

All this has been made possible by recent improvements in software and hardware, which have boosted computational performance, data storage capabilities, and network bandwidth. AI technologies are driving the digital transformation of industry and society by satisfying demands for more intelligent services and analytics. 

AI expert systems with built-in trustworthiness measures are helping healthcare professionals to make better decisions for patients, improving quality and efficiency in smart factories, and quite literally driving autonomous vehicles. The use of AI in the smart manufacturing sector is driving higher efficiencies by making it possible to collect and analyze production data in real-time. This allows for the deployment of predictive maintenance, as well as reducing equipment failure, increasing reliability, and improving asset performance. There are countless more examples of how AI is being deployed to boost safety, efficiency and to create new business models.

New ISO/IEC Technical Report

Applications such as these rely on a variety of sophisticated algorithms. A new ISO/IEC Technical Report, ISO/IEC TR 24372, aims to help users understand algorithms by categorizing them according to the purpose of the AI system. It is an approach that will be familiar to anyone who has read Stuart J Russell and Peter Norvig’s seminal university textbook on AI. It is a pragmatic way of structuring the TR that has the benefit of making it easier to browse and find relevant information.

The TR provides an overview of the state of the art of computational approaches for AI systems. It describes not only the main computational characteristics of AI systems, but also the main algorithms and methodology used in AI systems, referencing use cases contained in ISO/IEC TR 24030

“Computational technologies are at the heart of AI systems,” said Wael William Diab, who chairs the IEC and ISO joint technical committee (SC 42) that develops international standards for artificial intelligence. “Looking at AI applications and the systems powering them from a computational perspective enhances SC 42’s holistic ecosystem approach by complementing the other perspectives such as foundational, data, trustworthiness and governance.”

ISO/IEC AI workshop

The holistic approach that SC 42 adopts considers the entire AI ecosystem. It not only addresses technology capability but also non-technical requirements, such as business, regulatory and policy, as well as application domain needs, and ethical and societal concerns. The international standards developed by SC 42 address the ever-evolving requirements and challenges of an array of technologies, across diverse industries and applications. SC 42 draws on the expertise of a very broad and diverse set of stakeholders, together with IEC and ISO domain experts. New SC 42 publications in 2022 will cover areas such as AI data, trustworthiness, and ethics, as well as governance implications of AI, foundational standards and use cases. In 2022, SC 42 also expects to launch an ISO/IEC AI workshop, which will run bi-annually.

The workshop will target a broad, multi-stakeholder community interested in AI and the work of the IEC and ISO. In addition to promoting and proving information about the committee’s work, it will scout emerging trends, identify requirements, offer insights and provide opportunities that may help future standardization work. Additional aims include attracting new stakeholders and providing a platform for discussion. You can find out more about the work of SC 42 here.

As for Darwin, Emma wrote to her aunt to say that she had been pleasantly surprised when he proposed because he had never shown any romantic inclination in their fireside chats. They had a happy marriage by all accounts and went on to have 10 children. Darwin’s rudimentary algorithm  — notwithstanding the rather conspicuous Victorian era sexism in his lists — can therefore only be judged a success. Quod erat demonstrandum.