AI Platforms: Staying Compliant in a Fragmented Ecosphere

AI has attracted a vast amount of interest over the past few years. Not only are we now doing things in a fraction of the time it previously took, but we are also doing things that were previously not feasible.

And it is not a revolution for the few, AI is seeing broad use across a number of industries. What is fewer, are the current areas of interest in AI. Gen AI is dominating the business conversation around AI. 

A McKinsey survey on AI showed in 2023, before Gen AI captured widespread attention, AI adoption did not surpass 66% in any region globally. However, in 2024 when the tools became more widely known, the data showed that more than two-thirds of respondents across nearly every region reported their organisations were using AI. The biggest uptick of which is found in professional services.

GenAI chatbots like ChatGPT gained acclaim in 2024 due to their ability to dramatically augment workers’ output by generating marketing copy, summarising large reports; answering general questions and even translating documents.

Therefore, it is no wonder that platforms used for building AI applications are largely being tailored to creating chatbots and virtual assistants, something Gen AI excels in. 

“One of the most common applications of AI platforms is the development of chatbots and virtual assistants,” explains Steve Barrett, VP of EMEA at Datadog. “These tools engage people in natural, conversational interactions, providing support and enhancing user experience across virtually every industry sector.”

Yet in the AI frenzy the world finds itself in, many may be steaming ahead with implementation of these AI platforms to build their Gen AI without considering the hurdles. This may lead to abandoning of the projects, with a Gartner study showing up to 30% of all Gen AI projects may be abandoned as early as 2025.

AI’s growing pains

Implementing AI is costly. You need to consider getting the right staff, finding the right platform, allocating resources and then trialling it across your operations all whilst hoping it does not result in any downtime.

It is for reasons like this that AI platforms face trouble being implemented into business operations.

“Initial investment, operation, and continuous improvement of AI-powered applications can be expensive, and organisations often struggle to measure and realise ROI,” says Steve.

Equally, many companies are currently facing limitations from such platforms, putting a wet blanket on their dreams of what AI can do. 

“They are prone to performance (response latency) and response quality issues, like hallucinations, biases,” he elaborates. 

Whilst technical challenges are often mentioned in this context, the key challenges may actually fall into other areas of external concern. 

Alignment with corporate goals and strategies, introducing proper business and corporate measurement for the use cases at hand, but most importantly, the increasing governance that is coming with the use of AI platforms. How therefore, companies will balance their ambition with their limitations, particularly in the realm of governance, will pose an issue. 

Despite these challenges, organisations maintain bold expectations for AI’s potential. 

“A recent survey by Pega AI, which included over 500 business decision-makers globally, revealed that 74% are either “extremely” or “very” confident that AI can add significant business value within the next 5–10 years,” explains Peter van der Putten, Head of AI Labs at Pegasystems. 

Improperly administered AI

With the EU having implemented its AI Act, and the US pushing ahead with AI policies, it is now a matter of when and not if your AI operations will have to conform.

Therefore, organisations who may have only just begun to get their heads around AI implementation in their operations, now have to work out how to responsibly administer it. 

Luckily, many government policies on AI focus on risk and outcomes rather than attempting to regulate AI systems or platforms broadly. 

For instance, the EU AI Act categorises high-risk AI applications and imposes additional documentation and governance requirements. 

Therefore, some use of AI platforms may require much less attention than others. However, when dealing in industries like banking, if you use an AI system to deliberate on the credit worthiness of an applicant, or the eligibility and pricing in life insurance, then additional documentation and governance is required.

This necessitates a clear understanding of the processes and records of AI usage within such enterprises. However, the nature of AI applications means that multiple platforms may have been used to engineer various different elements of the process that was used to come up with that decision.  

This presents one of the primary difficulties: tracing the machinations that went into the decision. This diffusion can hinder efforts to implement effective oversight to ensure the systems you are running are working in accordance to regulations. 

“The implication for AI systems is that there needs to be some form of centralised registration of where and for what purposes AI used in the enterprise,” explains Peter. “This is virtually impossible to achieve in fragmented AI ecosystems, so it is a further driver towards the use of centralised AI platforms, especially for deployment and operationalisation of AI as opposed to mere development platforms.”

Bringing it all together

Although a fragmented approach may give organisations the ability to pick from the best in class, the problems it creates for sticking to regulations are some that have to be addressed.

Peter explains how a way to aid governance in Gen AI systems would be through the implementation of a RAG system. 

“With RAG, Gen AI chatbots or copilots could be programmed to generate an automated explanation on why, given a set of models, rules and data a customer is not approved for a loan,” explains Peter.

This approach may more easily allow financial institutions to demonstrate to regulators that their AI-driven decisions are based on legitimate, non-discriminatory factors. 

Despite being much more accurate, RAG models are still not 100% free of hallucinations, and so such an approach may only be totally compliant if used for customer feedback and not for explaining AI decision making. 

Yet, using a centralised platform may come out as the simplest way for many organisations wishing to develop, test and deploy AI across their organisations to remain in line with regulations. 

Such platforms offer a more cohesive approach to AI governance, enabling organisations to maintain better oversight of their AI implementations across various departments and functions. 

So while the potential of AI is vast, organisations must navigate the complexities of implementation carefully. By addressing the challenges of cost, performance, alignment with business goals, and regulatory compliance, businesses can harness the transformative power of AI effectively and for the long term. 

To read the full story in the magazine, click HERE.

Make sure you check out the latest edition of AI Magazine and also sign up to our global conference series - Tech & AI LIVE 2024

AI Magazine is a BizClik brand