Most experts define artificial intelligence
as technology with the capability of thinking of itself and making decisions
based on its own ideas. It’s going to be years before this becomes a reality,
if it actually does…as some experts argue. More philosophically, why would we
want computers not to require guidance and input from time to time where the situation
is new or uncertain? Particularly when dealing directly with customers.
What we should be using and aiming towards,
instead of Artificial Intelligence then, is Augmented Intelligence. That is,
man plus machine rather than man versus machine. The definition of this is
inherently vague but essentially it is where software supports human decision-making
and actions, and when it carries out repetitive or known tasks but defers to a
human for more complex or unique ones. Unless one is familiar with the
state-of-the-art of the technology, it is easy to believe the hype. The reality
though is that even the most sophisticated AI applications ironically require
armies of data scientists to develop and maintain them. For many, the Holy
Grail in Augmented Intelligence is an application that is trained and guided by
a non-data scientist, in particular so that the front-line personnel are not
directly doing all the tasks, but they are guiding the bots which help them.
How can businesses develop machine learning
models which automate processes not just today, but reliably ongoing? How can
they get continually rich insight from models when the data are changing around
them? Is the irony that the data scientist cannot bring a model to life, but
actually needs to constantly be the puppet-master: You can’t just build a
training set of data and then automate, you need to keep feeding training data
to keep the models up to date and prevent model degradation. Now the business
problem is starting to emerge: You wanted to use AI to automate a process based
on historic data so that you could free up the human resource that is currently
being deployed in the process. But if all those humans disappear and you let
the model loose and operationalise it, then ironically the source of fresh
training data also disappears. The computer is left to mark its own homework.
If the model is to be kept up to date, then you need to keep some or all of
those humans around, and it isn’t at all calculable how many you need. This is
the bane of a data science team, i.e. having to curate models and constantly be
involved in refreshing and validating, when there is plenty of new data science
opportunities to focus on to generate value. Also the curation process itself
is almost always sub-optimal as there is no way the data science team would be
able to second-guess if new signals have appeared without validation and
involvement from users. Further, if the business process itself needs to change
due to demands from the business, then the whole model could be left redundant
with no relevant training data for the new model.
However, the latest in technology being
used to analyse customer data uses machine learning to be able to classify interactions
accurately to guide processes and generate early warnings of issues and trends.
A perfect example of Augmented Intelligence. The first trick is that it
‘understands’ when it is uncertain about something, and it invites a human for
assistance (sometimes this is referred to as ‘human-in-the-loop’). The human
doesn’t need to be a data scientist, they only need to understand the domain to
impart their judgement and knowledge to the machine and once it’s done, it’s
done forever. The second trick is that it does this in an optimal way, asking
for the minimal amount of input from a human to maximise the performance, and
therefore the accuracy, of the machine learning. In this way, it can process
more and more tasks to an acceptable threshold accuracy, and hand off to a
human seamlessly when it is below this.
Analytics is one such company which has cracked this. It is a spin-out from The University of Warwick and it
has developed software called PrediCX which uses
machine learning to learn from various customer interactions to be able to classify
them accurately to guide processes automatically.
Let’s take chatbots for example, they need
to be curated to constantly improve and learn to new and changing signals from
customer intents. There is no reliable feedback loop, even if the customer ticks
“helpful” or “please can I speak to a human”, there is a lot of things that
could go badly wrong to use this for training and maintaining. Further, it is
critical to understand and classify the topics being talked about across all
channels, to encourage and facilitate the right channels for the right topics,
i.e. self-service/chatbots for FAQs and for complex queries to be quickly
routed to a human on the ‘phone, with chat and other semi-synchronous channels
perhaps somewhere in the middle. The latest Augmented Intelligence facilitates
an ongoing virtuous circle of harmonised classification across all and any
channel, to break down silos, improve internal processes, save costs and most
importantly optimise customer satisfaction.
In conclusion, AI is here and very much
here to stay. However, AI is Augmented, not Artificial Intelligence and for the
foreseeable future, if not forever, blends the best of machines with the best
of humans to make the perfect customer experience.