COMMENT
Editor: Andrew Allcock MIET
andrew.allcock@markallengroup.com
Art editor: Neil Young
Sales director: Joe Opitz
joe.opitz@markallengroup.com
M: 07967 169098
Sales manager: Beverly Jepson
beverly.jepson@markallengroup.com
M: 07860 715085
Publishing assistant: Lyn Evola
lyn.evola@markallengroup.com
Circulation manager: Chris Jones
circulation@markallengroup.com
Production: Chloe Jeakins
chloe.jeakins@markallengroup.com
Publisher: Jon Benson
jon.benson@markallengroup.com
Annual subscription: £118 (UK) and
£173 (Overseas).
Machinery (incorporating
Sheet Metal Industries)
Machinery is published on the first Friday of
every month by MA Business, Hawley Mill,
Hawley Road, Dartford, Kent DA2 7TJ
T: 01322 221144 F: 01322 221188
E: machinery@markallengroup.com
W: www.machinery.co.uk
If you change jobs or your company moves to a new
location, please contact circulation@markallengroup.
com to continue receiving your free copy of Machinery.
© 2019. All rights reserved. No part of Machinery
may be reproduced or transmitted in any form, by
any means, electronic or mechanical, including
photocopying, recording or any information storage
or retrieval system,
without permission in writing from
the publisher.
Please read our privacy policy, by visiting http://
privacypolicy.markallengroup.com. This will explain how
we process, use & safeguard your data
The views expressed do not necessarily represent
those of the editor or Machinery. Advertisements in the
journal do not imply endorsement of the products or
services advertised.
Print – ISSN 1753-0482
Online – ISSN 2049-3312
Printed by: Pensord Press, in the UK
Andrew Allcock, Editor
AI and data bias
Artificial Intelligence (AI) is a fast-developing technology and one
that Machinery is coming across more and more. Our EMO review
feature last issue (p14) and this issue’s cover feature (p10) both
reference it.
A press release that fell through the letter box recently drew
attention to the issue of data bias in AI. The release said that the
Royal Academy of Engineering (RAE) employed an AI algorithm trained on online image
search results for ‘engineer’ to generate images of what it learned a typical engineer
looked like – a white man wearing a hard hat.
It is the perpetuation of the stereotype that engineers are white, male and wear
hard hats, with its impact on recruitment from a broader pool for a wider range of
engineering roles, that is the issue for the RAE. Some 12% of the engineering
workforce in the UK is female, and 9% is drawn from black, Asian and minority ethnic
backgrounds, while engineering takes in many non-hard-hat-wearing roles. Leading
brands across the UK have signed a pledge and contributed to a new image library to
“address the misrepresentation of engineers and engineering online”, to try and
change the online face of engineering – www.flickr.com/thisisengineering.
An industrial example of data bias, and associated incomplete data, concerns a
cement plant. Says Saranyan Vigraham, vice president of engineering at AI software
firm Petuum: “Ernesto Miguel, 47, is a plant operator in a leading cement company.
He has spent the last three decades working in the same cement plant. He knows
each and every machine in his cement plant intimately. From the sound they make,
he can tell what can be wrong.”
The company brought in an AI firm to develop a system that would do Ernesto’s
job. The AI engineers working on this problem built a model of the plant equipment,
looking at the time series data of how the equipment behaved over the last two
years. Long story short: the AI system wasn’t up to the job. Says Vigraham:
“Engineers falsely think that they can model the real world by looking at the data…
Identifying the right correlations among the different data dimensions is a problem
that engineers are yet to solve elegantly as a community.” Data does not equal
knowledge, he observes.
Interestingly, with its IoT FIELD system that connects shopfloor equipment (p10),
FANUC is collecting both machine- and human-generated data, aiming to capture
some of that knowledge and provide a larger and more comprehensive data set.
The general point here is that AI requires data for all parameters linked to the thing
to be managed for accurate correlations to be made. Expect to hear more about data
bias and incomplete data sets as AI weaves its way into ever more areas. ■
www.machinery.co.uk
www.markallengroup.com
www.machinery.co.uk @MachineryTweets December 2019 7
/www.machinery.co.uk
/www.machinery.co.uk
/www.machinery.co.uk
/privacypolicy.markallengroup.com
/thisisengineering
/www.markallengroup.com
link
link
link
link
link
link
link
link
link