19 February 2018
Parth Desai

Parth Desai

Parth Desai - Pelican

7Posts 29,091Views 0Comments

Building Intelligence: Part Two

28 September 2016  |  2832 views  |  0

While there are potentially endless opportunities for the technology in financial services, there is currently a gap in understanding how Artificial Intelligence (AI) works.

As part of a series of blogs looking into what it means for the financial services industry, I have already explored the basics of AI. In this blog, I will look at how we can begin to implement the technology – and how different types of ‘knowledge’ encoded into a machine can achieve different outcomes.

Required knowledge

Essentially, knowledge is what makes computers intelligent. However, there are different types of knowledge that all need to be applied to varying degrees in order to carry out their purpose. Therefore, we need to understand how the categories work and how it can be used in a machine.

Based on the uses of knowledge, there are three types of AI machines. The first, smart machines, are based on common sense and can be seen in applications such as Google’s search functions or Siri’s understanding of language.

Intelligent machines – the second tier – are based primarily around specific domain or industry knowledge, but which also apply common sense or domain understanding when required. The third level, expert machines, are focused around solving a very specific problem, such as the chess playing Deep Blue, or robo advisors in wealth management, using the expert knowledge.

When we talk about AI, we are generally referring to the third level of expert machines. Where the real benefit lies, however, is in that second tier, which combine the best of the first and third tiers while remaining focussed on a specific domain. This can ease the workload burden by automating processes that have previously required human understanding, such as screening sanctions or handling deviations in payments.

Knowledge management

But how do we achieve this? The success of AI will depend on how we create and manage knowledge for machine use. The two main challenges for this are how can data be converted into knowledge and how to ensure the right knowledge is provided to AI machines.

Let us say that a bank wants to automate its payments system using AI. The first thing it needs to do is to ‘extract’ information from the humans already doing that job, looking at what information they are using and how they go about using it in their daily processes – for example, what they block and what they allow in sanctions checks.

The information needs to be relevant and accurate in order to work efficiently. This, along with the practical domain insights from human sources, creates knowledge.

The human touch

Human input will still be needed. These machines, unlike standard computers, have to understand and learn, and the knowledge given to the machines has to be validated in practice. Specialist ‘Knowledge Engineers’ will be responsible for this, educating machines in the same way we teach children and refine processes.

Finally, this knowledge needs to be encoded to be used by an AI machine capable of natural language processing, which can then apply specialist and common sense understanding in order to solve a problem. The bank has now introduced AI into payments.

AI represents the next step for the financial sector to improve efficiencies without compromising results – and arming yourself with the right knowledge will be key to its success.


TagsPaymentsTransaction banking

Comments: (0)

Comment on this story (membership required)

Latest posts from Parth

Edging ever closer to the abyss: why AI will reverse this trend

24 October 2016  |  7983 views  |  0 comments | recomends Recommends 0 TagsPaymentsTransaction banking

Building Intelligence: Part Three

06 October 2016  |  3339 views  |  0 comments | recomends Recommends 0 TagsPaymentsTransaction banking

Building Intelligence: Part Two

28 September 2016  |  2832 views  |  0 comments | recomends Recommends 0 TagsPaymentsTransaction banking

Building intelligence: Part One

20 September 2016  |  3828 views  |  0 comments | recomends Recommends 0 TagsPaymentsTransaction banking

Parth's profile

job title Founder and CEO
location London
member since 2009
Summary profile See full profile »
I founded Pelican in 1996 to develop an artificial intelligence solution that enables Intelligent Payment Management for banks and corporates. To date, Pelican has processed over one billion transacti...

Parth's expertise

Member since 2009
7 posts0 comments
What Parth reads
Parth's blog archive
2017 (1)2016 (4)2009 (2)

Who's commenting on Parth's posts