By
Ben Medlock, SwiftKey
Buzz phrases such as “artificial intelligence,” “machine learning”
and “natural language processing” are becoming increasingly commonplace
within the tech industry. There is a lot of ambiguity around these
phrases, so I’ll explain the substance behind the technologies and why I
believe they’re transforming the way we live, work and play.
When Ben graduated from the University of Cambridge in 2007, I left
with a compelling sense that the technology I’d been working with for
the last five years had the potential to change the world. I’d recently
completed a Ph.D. on the application of a new set of tools and
techniques from the emerging field of Machine Learning (ML) to a range
of tasks involving human languages — a field known as Natural Language
Processing (NLP). If this sounds confusing, I’m not surprised! Many of
the concepts are inherently complex. However, to try and make things
clearer, ML is about building software capable of learning how to
perform tasks that are too complex to be solved via traditional
programming techniques. For example, during my research I built programs
that were able to recognize topics in news text, grade essays, and
filter spam email. When the tasks are language focused, we call it NLP.
This represents a fundamental shift in the way software engineers
build complex systems. Historically, coding has been about distilling
the expert knowledge of the programmer into a series of logical
structures that cause the system to respond in predictable ways. For
instance, accounting systems follow rules, encoded by software
engineers, that automate the process of recording and managing accounts.
However, many of the tasks we come up against in our
information-saturated digital world require a level of sophistication
that can’t be captured in a series of human-engineered logical rules.
For instance, if I’m building a system to translate a sequence of text
from one language into another, there’s no manageable set of rules I can
encode that will solve that problem. However, if I create a framework
that allows the software to learn from examples of previously translated
sequences to make new translations, then the problem can be solved, at
least in principle. In other words, the system distills the expertise it
needs to complete the task from the data upon which it’s trained,
rather than directly from the programmer, whose authorial role has now
fundamentally changed. Evidently this new way of creating complex
systems requires a lot of data, but happily the amount of available
electronic data for training ML systems is growing at an irrepressible
rate.
It may be clear that such systems have potentially profound
philosophical implications for their authors. They cause us to question
commonly held definitions of understanding, intelligence and even free
will. To take a simple example from my own experience, when building an
ML system to grade essays, does it matter that the machine doesn’t
“understand” the content of the essay in the same way a human being
would? If you can demonstrate mathematically that the system is as
reliable as an expert examiner, does it matter that the method by which
it determines grades is based on subtle interactions between thousands
of underlying “features”, without an overseeing sentient mind? What role
does sentience actually play in the tasks most of us carry out on a
daily basis anyway?
Whatever the philosophical implications, software built around these
new technologies is changing our lives, even if we don’t yet know it,
and I believe this transformation heralds good news for us as consumers
and citizens. These new systems will enable our personal devices to
better adapt and anticipate what we need, right down to an individual
level. The days of the generic tech experience are numbered. People will
expect something completely tailored to them, from text-prediction
algorithms that understand the context of what you’re writing to
concierge systems that learn to preempt what you want to find, say or do
next. In 20 years I believe we’ll be surrounded by invisible systems
that mine a wealth of data about every aspect of our lives, constantly
learning, adapting and enhancing our decision making, health and general
wellbeing.
There are downsides, of course. Data privacy and protection must be
taken extremely seriously, and people are understandably wary of
computers that can “think” and learn like humans. If algorithms start
taking on the roles of teachers, personal assistants and others, does
this distance us from each other? I believe we need to wrestle with
these questions honestly and openly, and that the debate will ultimately
lead us to a better understanding of what it means to be human in a
technological world. Academic-sounding ideas like ML and NLP have clear
implications for the tech industry and the way we live that extend far
beyond our universities and research labs.
