understand that. Ontologies make this
knowledge explicit and give the computer rules
for working with and inferring more information
with that knowledge.”
While creating the next generation of machine-
learning algorithms that would enable AI systems
to have “common sense,” Amanda runs into the
issue of avoiding stereotypes, especially when
it comes to the concept of identity.
“How do you convey common
sense to a computer without
generating stereotypes? You
have to do this in a way that
overcomes potential bias
rather than confirming
existing biases.”
In most cases, machine-learning output is only
as good as its input. According to Lawrence
Hall, Ph.D., distinguished university professor
at University of South Florida, it’s the human
element that brings bias to datasets. To achieve
unbiased output, extreme care must be taken
by human users while selecting training data
and developing the algorithm. As the people
classifying and labeling data are trained to
better recognize their own unconscious biases,
machines will be able to pull from cleaner, more
unbiased datasets.
Lawrence’s colleague at the UCF agrees.
Gita Reese Sukthankar, Ph.D., professor and
director of the Intelligent Agents Lab at UCF,
forecasts an industry wide shift toward more
advanced machine learning that reduces
the need for human input, thus reducing the
influence of human biases.
The workforce will undoubtedly experience a
shake up as businesses continue to adopt and
advance AI, but The Corridor’s researchers and
industry leaders would agree the effects won’t
all be negative.
“My personal feeling is that it’s going to kill
some jobs, but it’s going to create new jobs,”
explained Lawrence.
He predicts many new jobs created by AI will
involve curating and fine-tuning data to maximize
the accuracy of machine-learning systems.
As AI-enabled technology becomes more
integrated into our daily routines, human input
and guidance will still be critical, but perhaps
won’t be needed forever. Rather than recording
and analyzing data manually, for example,
humans might someday learn to program
systems to do this work for them.
The processes that enable us to ask smart
personal assistants like Siri and Alexa about
the weather, deposit a check and mark email
as spam – processes that enable us to work
smarter, not harder – are continuously learning,
improving and advancing without signs of
slowing down.
“We’re just going to have to wait and see what’s
next for AI,” Amanda said. “It all depends on
organizational forces and on people’s creativity
– and how those two things interact never
ceases to surprise me.”
There is plenty of speculation as to what’s next
for this burgeoning discipline, but one thing
remains constant: The Corridor’s researchers
and entrepreneurs will be at the forefront as the
future unfolds.
While much of today’s AI runs on “supervised”
machine-learning algorithms, this kind of machine
learning is “unsupervised.” Whereas supervised
machine learning relies on data labeled by
humans, unsupervised machine learning needs
no labeling assistance – essentially, it’s “smart”
enough to analyze data without any guidelines
or variables. While advances in AI technology
trend toward unsupervised machine learning,
most researchers would agree consumers
should temper their expectations, since this
likely won’t become the norm for another five
to 10 years.
f l o r i d a . H I G H . T E C H . c o m | 59