business
‘‘
TALKING
////////////////////////////////////////////////////////////////////
3. Train your AI model
Once you’ve figured out the problem
and located the information to solve it,
the next step is to work on developing
your AI model. The first stage is to
provide training data that contains the
correct answers.
For example, creating a model that
recognises different vehicles requires
images of hatchbacks that are labelled
‘hatchback’, images of tractors that
are labelled ‘tractor’ and so on. Given
accurate training data, Machine Learning
algorithms will find the patterns that can
predict the correct answers. Applying
those patterns on raw unlabelled data is
where the AI model shows its value.
Proving the conceptual model can be
done with a relatively small set of data,
however, to support robust decision
making, comprehensive data sources
will need to be used.
The larger the dataset, the fewer
anomalies you’ll get. Use too little data
and you’ll have an answer based on
what that set of data is telling you, but
not reflective of a true business pattern.
4. Choose affordable compute power to
develop your AI capabilities
The volume of data you need to trawl
through when training your AI model
requires a powerful amount of compute,
but once the model is trained to do its
job, running it against real world data is
not nearly as processor intensive.
If you’ve invested lots of money in the
kind of high-performance infrastructure
needed to train your Machine Learning
models, the chances are you’ll be left
with a large chunk of it unused when
that first phase is over.
An alternative approach is to use a cloud-
based AI platform to do all the heavy
lifting from the outset in gathering and
analysing data, avoiding a considerable
capex investment. Designed specifically
for this type of data-intensive workload,
these specialised cloud services offer ‘pay
as you use’ access to some of the most
powerful servers commercially available.
40
INTELLIGENTCIO
It’s important to work with a partner
that can find you the right cloud
environment for your AI model.
Cost, performance and cybersecurity
concerns need to be finely balanced.
For example, once the model’s
up and running, it can be hosted
either on-premise or on a private
or public cloud or a multi-cloud
combination depending on what’s
best for the application.
Real life scenarios of AI and
cognitive computing
AI isn’t limited to any particular field or
function and can boost the capabilities of
any sector and sphere in the world. The
public spotlight often shines on high-
profile examples such as analysing digital
images for early cancer diagnosis, natural
language processing in Siri and Alexa or
the predictive capabilities of a self-drive
Tesla, but Machine Learning can transform
almost any market and often in seemingly
mundane ways.
In retail, an AI-based personal sales
assistant can prevent shopping trolleys
being abandoned before a purchase is
made by delivering real-time product
targeting. Modelling millions of users every
day can predict a shopper’s intent. It’s
then possible to match a brand’s product
offering to an individual’s preference and
increase conversions.
The route to lean innovation
There is now so much that the wider
enterprise world can do with AI that just
wasn’t possible five years ago. Plenty of
pre-bundled software now exists from the
main ‘cognitive in the cloud’ players such
as IBM, Microsoft and AWS. For example,
an application to recommend podcasts
can be stitched together with modules that
transcribe audio to text, search text for
keywords, index podcasts based on keyword
hits and display the results in a colour-
coded dashboard.
Proof of concept facilities are also being
made available by the main players that
allow you to undertake a ‘trial’ with a much
smaller dataset. This means that you can
provide a business case to the board and
demonstrate the business functionality of
your idea without investing large sums of
money upfront.
These services are now becoming more
widely available, demonstrating the
many possibilities that AI can bring to
an organisation. Imagine being able to
sift quickly through your business data to
recognise patterns and discern inconsistencies
so that you can then make predictions about
your business. Well, that time is here. Now is
the time to uncover the insights in your data
that will give your company its perpetual edge
over competitors. n
Image processing can reduce waste in ready-
meal manufacturing by sorting potatoes
by size, so chips are made from the longest
ones, hash browns from the medium sized
and mash from what’s left.
University students’ progress can be tracked
and improved by analysing assignments
and recommending personal study where
knowledge gaps are identified.
Booking flights, hotels and rental cars is
simplified when a chatbot can answer
traveller queries. The process of sifting
through vast quantities of data and
spotting correlations and inconsistencies
can be used to make predictions and
identified solve complex problems for
any business. Often the inspiration for a
successful AI project comes from exploring
examples in other companies.
Mark Vargo, Chief Technology Officer, CSI
www.intelligentcio.com