In recent months we have witnessed an explosion of interest in Artificial Intelligence (AI). This tsunami is being driven by the unveiling of ChatGPT and other generative AI tools that have piqued mass market attention through providing user friendly and game changing applications.
These breakthroughs have shown the bridging of the intelligence void between human and machine, expediting concerns on the ‘singularity’ occurring – the tipping point where AI transcends human intelligence.
This understandably has stirred up a great deal of excitement and trepidation around Artificial Intelligence, orientating around: what is it; opportunities that it can drive; risks/harms it can cause; how it can be managed and what the future holds.
This article will seek to shine a light into the AI abyss, with a specific focus on the author’s expertise in Machine Learning and financial services.
What is Artificial Intelligence? The Oxford English Dictionary’s pin-point accurate description of artificial intelligence is as follows:
“The capacity of computers or other machines to exhibit or simulate intelligent behaviour; the field of study concerned with this.”
The branches of AI are diverse with a multitude of applications, key typologies used in financial services/potential application are outlined below:
AI underpins the 4th Industrial Revolution, following on the back of:
• 1st Industrial Revolution (1760 – 1840) – introduction of steam power, the move of production from home to factories and the transportation revolution.
• 2nd Industrial Revolution (1870 – 1914) – introduction of electricity, assembly line production and steel used at scale.
• 3rd Industrial Revolution (1969 – 2000) – growth in use of computers, introduction of the internet, robotics and automation systems.
Summarising the first 2 Industrial Revolutions: humans created machines that mimicked human muscles and replicated the work of humans’ hands and legs. Machines scaled up hard human labour, artificial power as opposed to intelligence. It transformed the world.
“Without doubt as our industries’ embryonic understanding of these technologies matures, we will begin to see AI woven into the fabric of wide-ranging business decisions and processes”
Summarising the 3rd and 4th Industrial Revolutions: humans created machines that mimicked the human brain and replicated the workings of humans’ eyes and ears. Machines digitally scaled up human intelligence. Could it transform the world?
A new dawn – opportunities unlocked through AI application
The last decade has witnessed unprecedented growth of AI, and early indications that Generative AI such as ChatGPT that creates seemingly new, realistic content such as text, images or audio are here to stay. Testimony to its impact is the fact it took ChatGPT 5 days to generate 1 million users; by comparison Facebook took 10 months and Netflix 41 months! Phenomenal.
I have witnessed first-hand the game-changing impacts of applying Machine Learning in risk management. This is where existing automated process using traditional statistical models have been turbocharged through using superior Machine Learning models:
•
Significant reduction in Credit Losses on new Loan business with maintained approval volumes – a Machine Learning Credit Score was able to improve risk targeting on well- established automated controls that supplement the human based element of the underwriting processes. As a customer owned organisation this was highly desirable as it supports lending in a responsible manner.
• Bolster Money Laundering defences through the identification of more complex acts of criminality – through supplementing our transactional trigger-based rules with outlier detection models we were able to prevent more niche’ elements of criminality such as vulnerable young customers being exploited. Thus enabling our fraud specialists to review alerts not identifiable by industry standard modelling approaches. Without doubt as our industries’ embryonic understanding of these technologies matures, we will begin to see AI woven into the fabric of wide-ranging business decisions and processes.
Pandora’s box – Risks/harms that could be unleashed
There is no ignoring the risks associated with AI. Testimony to this being that AI safety was a priority discussion point during a recent meeting between Joe Biden and Rishi Sunak.
Blind adoption of AI would be very dangerous. As stated, AI capability is in its infancy and to ensure it is used in a safe and sound way there will be a need to have a ‘human in the loop’ to ensure little to no harm in its use.
Very real risks for AI in financial services include unfair and biased customer outcomes, not being compliant with regulations, reputational risk and even destabilising the financial system.
A prime example that such a risk exists with Generative AI is a concerning phenomenon called ‘hallucination’” – this is where the chatbot can give a confident and credible responses that is in fact false information. A very real and unsolved risk, and until resolved could cause significant harm.
Tapping into AI in a responsible manner – risk mitigation
Industries need to work hand in hand with the regulator in ensuring that AI is adopted in a safe and sound way.
Key areas of focus (amongst many!) include:
•
AI ready workforce – investing in AI experts and training programmes to ensure that it is applied in a safe way.
•
Robust Governance – having a robust Model Risk Management framework, data governance process and ensuring full transparency to accountable executives on risks, key assumptions and limitations.
•
Human in the loop – ensuring throughout the process human expertise and knowledge is integrated into quality assurance and checking.
•
Control Framework – monitor and maintain AI performance with robust controls. Have mechanisms known as a ‘kill-switch’ in place or alternative solution if not performing as planned. From a practitioner’s perspective, we de-risked the adoption of AI in the following ways:
•
Walking before we can run – the model was fixed (static) as opposed to self-updating as new data became available (dynamic). This eroded some of the benefit of the model, however significantly de-risked the model use, due to the live model being deeply understood based on the characteristics used and how they were assigning risk.
•
Socialising approach throughout model development process – regular engagement with subject matter experts across the business (compliance, oversight, audit etc.) and senior executives ensured a no surprises approach. This also enabled questions to be asked in transit that could be addressed prior to seeking final approval.
•
Champion-Challenger testing – not putting the model in live on a 100% basis allows a comparison between the incumbent solution and the new AI model. Thus, providing robust validation on original assumptions.
Crystal ball into the future……..
Artificial Intelligence is going to have a profound impact on the world as we know. For this to be successful business leaders and regulators will firstly need to put robust risk frameworks and controls in enable safe adoption.
When this is in place, there will be a huge opportunity to complement/enhance/scale existing as well as introduce new processes
