
Industry executives and experts share their predictions for 2019. Read them in this 11th annual VMblog.com series exclusive.
Contributed by John Kane, Distinguished Scientist, Machine Learning at Cogito
The Deployment and Demystification of AI
In 2019, society will push the providers of AI solutions to be
more transparent about the intentions of their technology and how they discover
insights and process data. The general buzz of implementing AI for
the sake of it will wear off and companies will demand that AI clearly
demonstrate its positive impact on their business. As people awaken to AI not
as something designed to replace, but rather augment their natural
abilities, we will see AI be more widely adopted in the workplace. Find out more
about what's to come in the AI field from John Kane, Distinguished Scientist,
Machine Learning at Cogito.
A Change in Terminology
In 2019, as an industry, we need to move away from the
over-reliance on catch-all, ambiguous terms like "AI," and develop terms the
provide more insight and clarification into the type of technology being
developed and deployed. With an increase in general sophistication in machine
learning, new terminology is beginning to arise to refer to more specific and
nuanced technology areas. This has already begun with terms like Augmented Intelligence to
refer to machine learning-based technologies which are specifically developed
to help and enhance human activity rather than replace it. The industry as a
whole needs to help classify and categorize AI, so that there is a clear
understanding of what machine learning is and how organizations can glean the
insights they're receiving to help drive results.
Machine Learning Models Will Become More Context-Aware
In 2019, context awareness will become even more important
for machine learning. Machine learning models will need to know more about
context (e.g., what device is being used, what prior information is known about
the user) and adapt accordingly. There will be increased interest in
understanding the emotional and health state of users of these technologies,
along with the need for contextually appropriate responses in the year ahead.
Democratization of Machine Learning
Despite "democratization of machine learning" solutions and
increased research attention on automated machine learning, for instance
Google's recent AdaNet autoML system, there will continue to be a
growth in demand for machine learning and deep learning specialists in the year
ahead. For modeling problems which are considered to be "solved" these automated
modeling systems will receive increased adoption, but for new modeling problems
or for areas which are still very much "unsolved," there will be an increasing
need for skilled scientists and engineers. In 2019, the industry (and academic
institutions) should focus on developing resources such as courses on ML and AI
and continue to invest in addressing the current skills gap issue and keep up
with the ever-increasing labor demand.
Discrimination and Bias in ML/AI
Businesses and society in general are becoming increasingly
concerned about discrimination and bias introduced by machine learning and AI.
This is with good reason, as there have been several cases reported including
credit rating algorithms treating people from certain demographics unfairly and
image processing models incorrectly classifying the color of people's skin. As
a result, there is an onus on technology companies to implement processes which
mitigate bias in their AI systems in 2019.
As we head into 2019, companies should adopt and implement a
model development protocol, which follows the FAIR framework. This involves using data collection
and machine learning techniques to reduce the effect of bias in models. If
companies fail to implement this, they will receive negative responses from the
public and their target customers and have ineffective applications.
##
About the Author
John
Kane is Cogito's Distinguished Scientist in the area of machine learning and
has over a decade of expertise in speech science and technology. At Cogito he
leads the research and development of machine learning algorithms to enable
real-time processing of audio, speech and other behavioral signals. John is an
active member of the speech research community, contributing as a reviewer for
leading journals and conferences in the space and as a maintainer of open
source speech processing tools.