From robot to intelligence applied to finance

From robot to intelligence applied to finance

Mon 05 Oct 2020

Robotic objects are part of our life: we interact with them daily and we deliver them our most personal data. But not all robots are identical and do not raise the same questions. Overview with simple words of what robots really are and their future contributions to the Finance Function.

A concept that summons a strong and evolving image

When we think of robots we tend to think of them based on literary or cinematic depictions (Frankenstein, for example, created over 200 years ago!). Typically the robots we imagine are mechanical beings that have been given some form of intelligence to help them perform tasks that can be labour intensive. Nonetheless, development has been significant. Only ten years ago, robots were used for ‘4D’ work, i.e. work that was ‘Dangerous, Dull, Dirty and Dumb’ (1). Presently, with the advances made in artificial intelligence, robotics must address four new needs, the ‘4Es’: ‘Everyday, E-health, Education, Entertainment.’

Today, the robot plays more of a facilitator role, although some people fear that ‘super-intelligent’ robots could seize power, putting themselves in direct competition with other humans. Let’s start by identifying the difference between an automated machine and a robot with artificial intelligence.

What is the difference between a ticket machine and Sophia, the first humanoid robot, who holds Saudi nationality as of October 2017?

An automated machine is controlled by computer programmes, all of its actions are predetermined in a logical sequence. The robot’s actions are based on a more complex set of factors, for example, its sensors enable environmental interaction. The robot is characterised by the following components:

  • Sensors — enabling it to be aware of its surroundings: camera, temperature, brightness, moisture, accelerometer, etc.
  • An IT system with electric circuits, and microprocessors for carrying out the robot’s tasks.
  • Mechanical systems creating motor actions, sometimes with tactile components for taking hold of objects.
  • Batteries for autonomy and energy.

The humanoid robot also possesses intelligence known as artificial intelligence (AI). AI enables actions through advanced choices, which can alter the environment. Without the physical shell, the robot is reduced to an IT application corresponding to an action system or decision-making system containing rules for initialisation parameters.

To summarise, AI goes beyond an expert system (traditional IT development) in which a series of static rules determine the robot’s actions. AI has this ability because it relies on a method of learning known as ‘machine learning’, which creates dynamic rules depending on the goal and/or the environment.

Recently, AlphaGo by DeepMind, the most highly publicised example, learned to play Go by analysing past matches, and went on to beat the champion of this strategic game. The new version, AlphaGo Zero, learned directly by playing matches via simulation without human assistance, and it outperformed its big brother without playing against a single opponent.

How can machine learning be useful in more abstract areas such as finance?

Currently, there are two technologies that can be used. Using semantic analysis, data scientists can, for example, develop tools to create reconciliations between contracts in real time (on a literary basis) and invoices from a service provider. This tool is useful when the process involves a large volume of invoices and contracts. With a ‘machine learning’ approach, finance departments can easily identify drivers of important financial developments by analysing the links between correlations proposed by AI, in order to identify root causes.

Machine learning has the ability to base its decisions on all available data. In the world of finance, data is structured. Today, there are many upstream data systems that are not being used to their full potential. If we add exogenous data to this financial environment, such as data relating to human behaviour, the possibilities become endless.

Let’s take a loan risk analysis, for example. At the client request stage, AI requests the amount of the loan and asks all the usual questions. It immediately compares the responses with the data from the digital life of the consumer that is accessible online, and then calculates the cost of the loan depending on the amount of risk. First of all, the methods for calculating risk would have been previously determined by data from the applicant’s previous loans, based on data of cases of default/non-default and any other available and relevant data, regardless of how personal. Secondly, AI continues to learn independently with the new cases it encounters.

Machine learning is not perfect since data is to the model what an egg is to a chicken – which comes first?

All of these examples draw on dynamic rules. The rules in question are determined by data science and machine learning applied in these areas. Data associated with algorithms produces a predictive and/or prescriptive model and this model will determine the machine’s actions. These elements are regularly updated (and will soon be updated in real time!) to ensure there is consistency between the situation and the clients.

What is the essential part of this process? The data obtained by sensors, of course! However, there is a link between the quality of the data and the quality of the model. In this context, the data must be accessible, legal to use, specific and statistically correct. Otherwise, AI recommendations would be mediocre in quality and could call the systematic model into question. Remember Microsoft’s chatbot ‘Tay’ who displayed ‘inappropriate’ behaviour after its model was based on social media conversations?

To sum up, the model must be confronted and validated by input data regularly, and this data must, at a minimum, be formed by a representative sample.

Robots are starting to revolutionise the way we work – and finance is no exception. Contact us to learn more about the advantages they can bring to the way you work. 

(1) G. A . Bekey, Robotics : State of the Art and Future Challenges, Imperial College Press, 2008
(2) Laurence de Villiers ” Des Robots et des Hommes: mythes, fantasmes et réalité “, Plon, mars 2017