The current consensus is unanimous: a technical revolution is already underway in professional services. The exciting innovations afforded by AI offer breathtaking potential. Dramatic change is inevitable. Firms slow to adopt and adapt may simply get left behind.
However, the initial euphoria, certainly in the legal and tax space, around the potential of ChatGPT and other generative AI tools is increasingly being tempered with a sobering wariness about their evident downsides.
A recent survey*, by the Thomson Reuters Institute, of more than 440 respondents at large and mid-size law firms in the UK, US and Canada found that 80% of partners ”had concerns” about the use of generative AI for legal work. Their apprehension focused principally on questions of authenticity and security, specifically how privacy and client confidentiality issues will be addressed.
These models have a prodigious but indiscriminate appetite for information, and therein lies the problem. Although improving with each new incarnation, their susceptibility to flawed, misleading or obsolete results, that often appear plausible, remains. The fact that they’re typically trained on unprotected sources does indeed prompt serious misgivings regarding data protection and copyright violations. Indeed there are already publicised cases of sensitive and proprietary information being compromised. It appears that, without prohibitively time- and resource-consuming supervision, there exists an inherent risk of security and liability repercussions. On this basis, anxious reservations regarding what regulatory censure may be imposed in the future would appear entirely legitimate.
Any attempt to keep generative AI tools updated with the constant changes in legislation that characterize the legal and tax fields will inevitably be costly and onerous. Equally, the lack of adequate industry pre-trained data appropriate for informing and fine-tuning impedes them from being competent at complex and nuanced classification. Essentially, as Large Language Models don’t yet understand the meaning of the words they’re parsing, they’re effectively trying to predict results based on historical patterns.
A further limitation of inductive models is their tendency to struggle with the demonstration of clear processes and reasoning. In short, they’re unconvincing at justifying their appraisals. As Mark D. McDonald of consultants Gartner puts it, the unmonitored use of generative AI, particularly in areas such as accountancy, creates murky compliance issues: “Referencing an algorithm as the rationale for tax decisions is not an excuse that auditors will accept”.
Generative AI products boast great proficiency at information discovery and retrieval and undeniable superiority as creative tools. But for analytical and deterministic tasks they remain very much a work in progress, prone to unpredictable, erratic and dubious results. AORA’s cognitive AI-based automation software, on the other hand, provides a proven and reassuringly reliable alternative. Its innate precision derives from explicit pre-training on specific data. This delivers consistent accuracy at a granular level. The rigour and discipline of its strict programmed rules ensure authoritative, dependable outcomes.
Being pre-trained on a large body of open public information, then fine-tuned and comprehensively tested on AORA’s IP, privacy and security risks are eliminated. Likewise, there’s no exposure to copyright or confidentiality infringement as client data is not used for training the system. In this respect, AORA’s superiority over generative methods is emphatic.
This same virtue means it’s easy to efficiently update rules-based platforms with amendments. Revisions are neatly and faithfully integrated into the programmed logic in good time for the implementation of new regulations. In areas of rapidly changing and exacting governance, such as corporate taxation, this facility for modification and replication is massively advantageous.
In stark contrast to the chatbots’ auditing woes, AORA offers “explainability” and accountability: its draft opinions are reasoned coherently from rules to conclusions, providing a precise report. These results are categorically supported by appropriate references, in a format acceptable to the relevant authority, equipping the practitioner with a wealth of definitive evaluation with which to advise the client.
A vast amount of time and money is being ploughed into the continued development of generative AI products, largely devoted to gathering enough useful data for testing and training. Meanwhile, knowledge representation as AORA executes it, for example in tax determination, will in many cases yield practical business solutions significantly more quickly and at a fraction of the cost. Most importantly, these solutions are targeted, accurate and robustly reliable.
Innovative tax and legal automation of this calibre yields substantial benefits for advisory firms in terms of augmented workflow. It drastically reduces the expert hours previously devoted to mundane and laborious tasks, liberating valuable time for strategic analysis and client-facing service. This doesn’t just mean a noticeable increase in efficiency and therefore reduction in costs. It also improves accuracy due to the immediate identification of inconsistencies and errors. Also, as well as being a rich learning resource, it enables insightful and streamlined collaboration between colleagues, and continuity in knowledge sharing for employee succession planning.
No less an authority than Conrad Young, Advisory Board Chair at Oxford University’s Internet Institute, remarks: “Tax authorities around the world are transforming fiscal systems through digital technologies. AORA is at the forefront of the market response, developing solutions to codify some of the most complex and often arcane tax rules, in user-friendly applications that augment the professional work needed to apply them. In part automation, in part expert system, the AORA platform is successfully harnessing the latest technologies to dramatically improve the way tax work is done.”
As technical innovation moves on at pace, those slow to adapt and adopt are in danger of being left behind. But in the rush to automate, the importance of choosing the right tool is paramount.
*ChatGPT & Generative AI within Law Firms, Thomson Reuters Institute, March 2023
