Advice pioneers in the wild, wild AI world
As all and sundry enthusiastically integrate Chat GPT and other artificial intelligence (AI) tools into their daily home and work lives, the financial services experts at law firms are preparing for a potential deluge of challenges to AI-centric approaches to traditional banking, insurance and investment practices.
From marine and war to chatty bots
After bouncing between old-as-the-hills insurance practises in areas like apportionment, marine insurance and aviation-linked war exclusions, it was left to one of the youngest among the presenters at the 2024 Norton Rose Fulbright Insurance Seminar to bring the audience up to speed on how AI might affect them. Triston Marot, an associate at Norton Rose Fulbright South Africa, and one of the top performers in the 10th running of the popular reality styled series, The Insurance Apprentice, introduced his talk by saying the debate over the roll out of AI tools was heating up.
Almost overnight, the financial services industry has segued from one where AI’s integration into businesses was decades away, to one where solutions are imminent, if not already in play. Why? Because even the AI experts were caught off guard by the pace of recent developments. The unfolding preparedness debacle is illustrated by the shift in AI experts’ best guesses on when machine intelligence would be feasible, over time. In 2016, 50% of them said by 2058, with 7% saying by 2030; in 2019, 50% said by 2060, and only 5% by 2030; today over 50% reckon we will have human level machine intelligence by 2047.
The launch and widespread adoption of Generative Pre-trained Transformer (GPT) models, starting with GPT-3 in June 2020 and followed quite rapidly by Chat GPT3.5, in November 2022; and Chat GTP4 and 4o, both in 2023, has opened the world’s eyes to the potential in the technology. According to Marot, the insurance industry is already benefitting from AI in areas like claims automation and management, among many others. “Lemonade, a US-based insurer was able to settle a genuine claim in less than two seconds using their AI-powered claim automation tools,” he said. The applications of the technology are limitless.
Automation with investigation baked-in
“We have seen motor assessment models that are able to take in photos of an accident, connect with a city’s CCTV system to gather photographic or video evidence; check cost estimates to determine the reserves etc. all without human intervention,” Marot said. Commenting on innovations being trialled domestically, he singled out Billy Bot, a chat bot in use at The Liability Company and the use by Discovery Healthy of a range of AI tools to “advise call centre agents in real time as to the client issues they might be dealing with”.
More importantly, insurance industry stakeholders are seeing a massive improvement in fraud detection and prevention thanks to AI models that can process vast amounts of data to pick up fraud patterns and detect things which might not be obvious to human assessors”. Marot then described how AI tools were pushing into various insurance disciplines, from risk assessment and pricing to regulatory compliance and risk management.
Under the first discipline, he said: “We are starting to see models that can rapidly assess risks on a multitude of different data points, essentially making manual underwriting for certain risks completely irrelevant”. You and your clients can look forward to personalised underwriting and the integration of AI models with Internet of Things (IoT) devices, among countless other innovations. As for regulation, the Financial Sector Conduct Authority (FSCA) is “using AI models and tools to go through the vast quantity of data that you [and other FSPs] report back to them to try and identify compliance risks which might be occurring in real time”.
Over time, AI will revolutionise every aspect of broking, financial advice, risk management and underwriting. But there is a more fundamental concern that arises from the world of science fiction movies such as 2001: A Space Odyssey and the popular Terminator series. The question each of us is wrestling with is whether a future AI-powered HAL 9000, T-1000 or Rev-9 will pose an existential threat to humanity. “No one truly knows the answer to this type of question,” Marot said. “But there are many brilliant scientists at many AI organisations who are working on the issue, specifically the issue of how to align AI tools and models with human interests, policies and governance”.
AI: A cybercriminal’s paradise
You and your clients need to be aware of some of the real AI threats that already exist, most notably in the cybercrime realm. Marot said his firm had recently experienced an attack where bad actors used an AI model to clone a CEOs voice in an attempt to authorise a payment. Disinformation campaigns and AI-powered propaganda were also singled out as a potential threats during a year in which dozens of democracies are holding national elections, including South Africa in May, and the United States in November.
The last part of the presentation focused on data protection in an AI world, and the role that government and financial sector regulators would play in keeping AI in check. “Commissioner Kamlana of the FSCA recently said that while they appreciate the business imperative for the use of new technologies, [these create] a need to address ethical challenges, and intensify efforts to be transparent, fair and responsible,” Marot said. He offered section 71 of the Protection of Personal Information Act (POPIA) as an early legislative attempt to protect customers from machines, through its requirements for automated decision making and knowledge surfacing systems.
The legislation prohibits decisions that are based solely on the automated processing of personal information. “The POPIA does carve back in the ability to use such systems if (1) it is governed by a separate law or code of conduct which sets out appropriate measures to protect the data subject … or (2) if the process deals with the conclusion or execution of a contract,” Marot said. In the latter case, there are two further requirements being that the data subject must get what they want, and if not, that their legitimate interests are protected.
Transforming insurance practices, regulation
So, it is mostly bad news for companies hoping to blame the AI when they get compliance wrong. Automated decisions resulting in legal consequences or significant effects on individuals are generally prohibited unless specific conditions are met. “Data subjects must be provided with the opportunity to make representations about an automated decision made about them,” Marot noted. This underscores the necessity for transparency and human oversight in AI-driven processes, particularly in contexts where decisions impact individuals’ legal or financial standing.
Marot also highlighted some of the technical and legal difficulties in fulfilling POPIA’s transparency requirements, especially with advanced AI models like transformer models or large language models, which often function as ‘black boxes’. To address these challenges, many organisations maintain a human in the decision-making loop to ensure compliance. This raises further legal questions about whether such practices truly align with the spirit of the law. “There are some very strong disagreements, even within Norton Rose Fulbright, as to what that may mean,” Marot said.
Internationally, regulatory approaches to AI vary, with the European Union’s (EU’s) AI Act categorising AI systems into different risk levels and imposing stringent requirements on high risk systems, including detailed record-keeping and human oversight. Marot noted that “the EU decided to make this act extraterritorial” thus affecting AI systems used in or impacting EU consumers regardless of where they are developed. Similarly, the Financial Conduct Authority (FCA) in the United Kingdom favours a principles-based regulatory approach in deploying AI systems in financial services, focusing on fairness and transparency.
Not Dickens, but still the best and worst of times…
Marot concluded that robust governance and accountability in AI deployment was non-negotiable, and that local regulators would likely follow their international peers in developing regulation. His parting words drew from Professor Stephen Hawking’s warning about AI’s dual potential, stating: “The rise of powerful AI will either be the best or worst thing to ever happen to humanity; we do not know yet which”.
Writer’s thoughts:
The observations in this piece re POPIA and automation were interesting; but I doubt the regulators had any idea of the power of emerging AI when they penned it. Do you think current consumer protection legislation is adequate given how fast FSPs are adopting and integrating the technology? Please comment below, interact with us on X at @fanews_online or email us your thoughts [email protected].
Comments
Amen, Brother. Report Abuse
This article is a call to arms, and we must e4quip ourselves ourselves with all intelligence available in order to combat this
We must shake off the false securities we are harnessing and take up armour. Report Abuse