Generative vs. Deterministic Artificial Intelligence in Compliance Workflows

0
445

Written by Sumeet Singh, Founder & CEO of LighthouseAI.

Meta: Discover the use cases for generative and reactive artificial intelligence and why some forms of AI may be better suited for business compliance solutions than others.


In Fall of 2022, ChatGPT took the world by storm, captivating people with the power of Generative AI technology and piquing a particular interest in the seemingly sentient capabilities of this technology.

Indeed, this powerful tool can potentially disrupt entire industries while rendering others obsolete.

Consequently, people’s reactions to the arrival of ChatGPT have been diverse: some are intrigued by its potential; others find it remarkable, and still more feel a sense of amazement. On the other hand, some people – perhaps more than not – fear the repercussions on how this technology will affect their livelihoods.

Generative AI has been demonstrated to show incredible value for various tasks such as text generation, image processing, and much more.

However, despite the advancements and proliferation of generative AI technology, there are still some limitations in its use cases. Specifically, some tasks require complete certainty that is not attainable with a limited memory model, possibly due to the complexity of the problem or lack of data available, and thus are not suitable for generative AI.

Today, we’ll explore an overview of AI development, discover the limitations of certain types of AI in various industries, and propose a solution for applying AI in compliance use cases.

Types of Artificial Intelligence: Limited Memory AI & More

There are four significant evolutions of AI, and each category leverages different technologies and mechanisms to achieve unique goals.

These 4 categories of AI are:

  • Reactive (e.g., Deterministic AI: expert systems)
  • Limited memory (e.g., Generative AI: ChatGPT, Midjourney)
  • Theory of mind (i.e., understand human feelings and emotions)
  • Self-aware (e.g., Terminator, The Matrix)

Generative AI is a type of “limited memory AI,” which means that it is trained on a set of data and makes its predictions by using statistical analysis upon said corpus of data. ChatGPT-3, for example, is trained on all publicly available data collections on the internet through September 2021 (which, granted, doesn’t seem so “limited”!).

The advances of this type of AI are creating the most buzz, spanning text-based AI, like ChatGPT, but also image-based solutions like Midjourney. Moreover, for a good reason – ChatGPT’s capabilities are incredibly impressive, scoring in the top 80% of the bar exam1 and achieving 97% accuracy in diagnosing cancer treatment!

Limited Memory AI Limitations

The possibilities are virtually endless when it comes to Limited Memory AI applications. For example, applying statistical AI models to expansive datasets can be a powerful tool in detecting fraudulent banking activity and meeting legal requirements in real time.

However, it is equally vital to understand that human data scientists are still needed for interpreting the information, validating results, and making decisions based on these findings or indications.

While impressive, implementing AI models presents a unique challenge for companies, as they must rely on more than just accuracy that is perfect for sensitive use cases, such as compliance and patient safety. It can also “hallucinate” AI applications, as it can be manipulated by misinformation and training bias.

AI in Compliance & Beyond

Regulatory Compliance is a unique field that involves applying laws and regulations to industry- and location-specific business processes. Given the immense complexity of each business’s operations, compliance activities are also highly nuanced. Therefore, they demand a compliance officer to hold various considerations in suspension while taking action towards full compliance.

While generative AI is incredibly powerful, it is inherently inadequate to fundamentally disrupt Regulatory Compliance because more than perfect accuracy, as described above, is needed. Generative AI can therefore only augment – not replace – human compliance efforts. As a result, Generative AI is relegated to creating greater efficiencies (reducing costs), but not actively mitigating risk (protecting revenue).

Enter: Reactive AI for Compliance Workflows

For AI to truly disrupt compliance, expert systems offer a potential solution. Expert Systems are a type of Reactive AI. According to Java T Point, “An expert system is a computer program designed to solve complex problems and provide decision-making ability like a human expert.” They go on to explain that an expert system “performs this by extracting knowledge from its knowledge base using the reasoning and inference rules according to the user queries.”2

While generative AI requires extensive datasets and consistent human intervention and oversight to ensure accuracy, expert systems are a form of proven AI that can handle unique situations through thorough human training. This specific training enables the system to make inferences — a hallmark of AI.

Unlike generative AI, expert systems’ knowledge bases can be readily applied to businesses without extensive training or intervention.

Regulatory Change Management (RCM) is a discipline that seeks to keep companies up-to-date on their compliance requirements. Because of the number of jurisdictions that companies have activity in, it can be a significant challenge for compliance departments. In addition, proper RCM involves a great deal of complex work, including computing changes, applying them to a company, and taking action.

However, an expert system can automate this process from discovery to action. This is where Deterministic AI shines in contrast to Generative AI, which may need to be better suited for this type of task.

Of course, limited memory AI and reactive AI have their unique strengths and use cases. For example, limited memory AI is ideal for tasks that require analysis of large datasets or complex patterns, while reactive AI is better suited for tasks that involve real-time decision making with immediate feedback – that may also require 100% accuracy and perfection in results.

While both types of AI can be potent, they should be utilized according to the specific needs of a given problem. However, when used in tandem, these two types of AI offer a comprehensive approach to tackling problems from multiple angles and taking full advantage of their respective advantages.


Sumeet Singh is the Founder & CEO of LighthouseAI. With over a decade of experience in the life sciences industry, Sumeet has a blend of strategic planning and “boots on the ground” operational expertise. Sumeet is a thought leader within the life sciences industry, having been invited to educate state regulatory agencies and published in trade publications, including Pharmaceutical Commerce, and has spoken at events including American Conference Institute Controlled Substances Summit, IQPC Pharmaceutical Traceability Forum, the Fourth Annual PBOA Meeting & Conference and Chain76.

Sources:

  1. ABA Journal. (2023). Latest version of ChatGPT aces bar exam with score nearing 90th percentile.
  2. Java T Point. (). Expert Systems in Artificial Intelligence.