Working out what we want from AI
It is hard to find something when you don’t know what you’re looking for, writes Mike Mullane
Thomas Edison reportedly tried 10,000 times before he managed to create a filament for electric bulbs, but he dismissed any talk of failure. “I have not failed,” he corrected his detractors. “I’ve just found 10,000 ways that won’t work.”
We seem to enjoy hearing about mistakes more than celebrating success. Although AI technologies have transformed our lives, at home and at work, many recent media reports focus on the failings of smart devices, from disappointing gadgets on show at the CES to malfunctioning hotel bots. Some of the stories are very funny, but all they tell us is that the technology is still in development and that some products are better designed than others.
The Wall Street Journal writes about a guest in a robot-staffed hotel in Japan who was woken every few hours by the in-room assistant asking him to repeat his command. The hotel manager finally realized that heavy snoring by the guest had triggered the robot’s voice recognition system.
For every clanger, though, there are success stories. Even when a machine achieves something as banal as winning board games it can be the harbinger of transformative benefits. For example, a chess-playing programme called AlphaZero, developed by the Alphabet-owned (Google’s parent) AI research company DeepMind, has been making significant advances.
AlphaZero has developed a new style of playing chess which is much closer to human improvisation than traditional computer chess. That is because AlphaZero learns from its past successes and mistakes, rather than calculating millions of possible permutations as it plays. According to Wikipedia, AlphaZero searches 80,000 positions per second in chess, compared to 70 million for the Stockfish chess engine.
AlphaZero uses (deep) neural network technology — sometimes called deep learning — which has resulted over the past decade from notable improvements in machine learning. As computing power has increased, deep neural networks have produced machines capable of performing tasks in a way that would not have been possible using traditional programming techniques.
This has transformed technologies such as computer vision and natural language processing (NLP), which are nowadays being deployed on a massive scale in many different products and services. Manufacturing, healthcare and finance are just some of the sectors that use deep learning to uncover new patterns, make predictions and guide decision making.
“In the area of smart manufacturing, AI can help to streamline efficiency,” says Wael Diab, who is leading international standardization work in this field. “It can help to provide insights in terms of where improvements can happen and more importantly it can provide insights into where a particular organization may want to go in terms of its production planning.”
Sales of industrial robots have doubled in the past five years, according to the International Federation of Robotics. The IFR predicts that in 2021 the annual number of robots supplied to factories around the world will reach about 630,000 units. Industrial robots are satisfying a real need.
In contrast, much of the focus in consumer electronics is still on the novelty value of gadgets. To a large extent this is because we have not quite worked out yet how we intend to use AI-enabled devices in our everyday lives or what we expect of them.
The Korea Joongang Daily reported in October that Koreans not only use their smart speakers for changing the TV channel, but also to discuss their feelings. In people’s homes, a staggering 15% of the things said to smart assistants appeared to be attempts at conversation, including “I’m bored” and “I’m sad”. The newspaper noted a similar pattern in hotel rooms, where more than 18% of the commands were attempts at conversation. The Joongang Daily acquired the data from KT Corporation, the country’s largest telephone company.
In 2017, IEC and ISO became the first international standards development organizations (SDOs) to set up an expert group to carry out standardization activities for artificial intelligence. Subcommittee (SC) 42 is part of the joint technical committee ISO/IEC JTC 1.
SC 42 is working with other JTC 1 subcommittees, such as those addressing the internet of things, IT security, and IT governance, as well as the IEC Systems Committee (SyC) for Smart Cities. SC 42 has set up a working group on foundational standards to provide a framework and a common vocabulary. Several study groups have been set up to examine the computational approaches to and characteristics of AI systems, trustworthiness, use cases and applications and big data.
IEC Standards are playing a key role in the transition to the Fourth Industrial Revolution. IEC TC 65, for instance, carries out important work related to industrial-process measurement, control and automation.
“We’re looking at the different components that go into AI, from the computational side to the ethical side. Having standards allows for a common language and way for the different stakeholders to interact,” explains Diab.
“What that leads to is the ability to innovate on top of widely adopted standards in the market place.”
