AI is not a panacea and there are many scenarios where it cannot be usefully applied. Below, MGTL’s Managing Director Simon Shepherd expands on the wisdom of approaching AI with caution.
Anyone trying to crowbar GenAI into the Due Diligence function is misguided. The tools already exist and though they might be deemed mechanical or ‘procedural’, the heart of the matter lies ultimately in the judgements made by Network and Vendor Managers alike. Asking well-directed questions will remove the need for an over-engineered and expensive solution which, as an aside, may well reinforce the natural bias of GenAI anyway.
Prospects and Clients alike should be deeply suspicious of Vendors jumping on the AI bandwagon as a way of desperately trying to differentiate themselves. There needs to be some level-setting on the language used and the meaning of the terminology; and then an informed conversation about whether the use of AI is sensible and relevant.
To go back to definitions: GenAI (generative artificial intelligence) is artificial intelligence capable of generating text, images, or other media, using generative models. Generative AI models learn the patterns and structure of their input training data and then generate new data that has similar characteristics (according to Wikipedia, 21st October 2023). ‘Generating new data’ is not what due diligence is all about: due diligence is about checking facts and verifying opinions.
This is precisely what is problematic with GenAI in the context of due diligence and close scrutiny of this definition suggests GenAI is at odds with many, if not all, aspects of the due diligence function (which Wikipedia also defines as a comprehensive appraisal of a business, not least when buying a service). The ‘similar characteristics’ is the bias reinforcement mentioned above, when due diligence is actually designed to weed out the (natural) bias in responses received. That is the point about due diligence: at its best, it is designed to reveal inconsistency and inadequacy, not simply to re-affirm what you are being told.
Generative models to conduct due diligence – to generate text – sounds at best like a re-write of what is there already and at the extreme, the authorship of brand-new material. This is what is concerning the authorities at the moment and is the first step on the road to self-awareness for AI, something which is hardly understood at all.
But due diligence is both the scrutiny of alleged facts and statements and the consequent forming of an opinion as to the validity of those statements, whereupon an informed conclusion can be reached. Traditional AI (TradAI) can already be used for delta management (spotting new or different material), scoring and scorecards, ranking or rating and reportage. Even calling this ‘TradAI’ is a stretch, but the functionality exists already, is highly fit-for-purpose, proven and readily available. True GenAI is simply not relevant to due diligence.
To avoid being entirely philosophical about this, a recent study by Harvard Business School for Boston Consulting Group (BCG) released late last year revealed some interesting conclusions. Using GPT4, creative product innovation boosted performance on average by 40% across 750 Consultants, but business problem solving suffered an on average drop in performance of 23%. BCG Consultants had no idea whether the AI was helping them or hurting their performance – they actually mistrusted it when it helped and trusted it too much where the technology was not competent.
Furthermore, the study identified a “creativity trap” – which might actually be good where due diligence is concerned – whereby individual performance improved on average but collective creativity actually dropped. The Study also suggested that there is a danger of “falling asleep at the wheel” i.e. simply accepting what the AI is telling you and not challenging it. If AI was designed as a ‘navigation aid’ as with GPS, users have to be careful not to lose their map-reading skills.
Therein lies the point: conducting a due diligence exercise needs both subject matter expertise and interpretation. Many responses to due diligence questionnaires are nuanced and GenAI is ill-suited for the purpose of interpretation. Doing the heavy-lifting of combing through reams of documentation is also a mis-fire; a handy Word-Editor will do the same and make up for any shortcomings in the answers to those well-directed questions – and at far lower cost.
We would be more than happy to showcase CODUDE’s capabilities in this respect, though it will not be badged as ‘AI’ in any respect.
