The AI landscape is undergoing a seismic shift, with GenAI poised to disrupt a range of industries. Discriminative AI and traditional machine learning models have long been instrumental across forward-thinking financial organizations, helping them to be more efficient and effective. But the allure of GenAI’s transformative potential is undeniable. As this clash of old and new unfolds, we need to ask: Should capital markets prioritize proven methods or embrace the promise of GenAI?
Why capital markets should stick with discriminative AI and traditional AI
GenAI might grab headlines, but it’s important to not overlook the enduring power of traditional methods, such as linear regression, decision trees, clustering, LSTMs (Long Short-Term Memory) and support vector machines. These are not new to capital markets, which enthusiastically embraced such methods. And there are good reasons to stick with them over GenAI:
- Explainability: Traditional methods are more explainable and interpretable than GenAI. This makes them a good fit for financial markets, which will constantly be considering regulatory compliance, risk management, and mitigating bias.
- Reduced costs: Traditional methods tend to be more cost-effective, in part due to being less computationally expensive than GenAI. Running a simple neural network will be cheaper than using a large language model.
- Maturity: Traditional methods are proven, robust, reliable, and really good at specific tasks. Traditional ML is great at well-defined objectives when you’ve a plethora of labelled structured data from a trustworthy source.
There’s a good chance your organization is well aware of these benefits and already using discriminative AI and traditional ML to improve risk management and loss mitigation, boost efficiency, simplify ops, and more besides. Even if not, these methods remain excellent for classic use cases like classification, categorization, and prediction.
For example, if you have a time series prediction task, where you have stock market or manufacturing data and want to predict what will happen next, neural networks offer a good way to do that now. Or if you need categorization for email spam – a classic ML use case – there’s no reason to switch to GenAI.
So if your model works for a specific task, keep it. Don’t move to GenAI for the sake of using the latest tech. In fact, maybe you can even forget all about GenAI, right? Well, not so fast…
Why capital markets need to start exploring GenAI
So here’s the twist. Even if you’re the most conservative financial organization around and feel traditional methods have served you well, you can’t stand still. It’s possible you’re already feeling the pressure to use GenAI – and rightly so.
You might argue that GenAI is often not a good fit for the industry. While that in many cases may be true today, it won’t be forever. Rapid advances are pushing the boundaries of what’s possible, which means that even if your traditional methods are working perfectly, you need an understanding of what GenAI is capable of.
Consider the time series prediction task I mentioned earlier. That remains an excellent use case for traditional methods. But what if I told you GenAI is shifting to the point it can do this just as well? And that we may soon find GenAI hasn’t just caught up, but blazed past?
Then there are areas in which traditional methods are already outperformed by GenAI, such as finding patterns in natural language and unstructured data. GenAI is more flexible, dynamic, and personalized in how it creates content, in ways traditional methods don’t have the capability to do at all. This means there’s huge potential for disruption and acceleration in a range of capital markets use cases.
Bank to the future: a hybrid model
So, where should you head next? Traditional models? GenAI? The best advice right now might be both. Or at least, don’t discount anything, while working with what’s best today and preparing yourself for what’s to come. With a hybrid approach, where each technique is strategically employed, you have the key to navigating a rapidly evolving AI landscape.
This means continuing to recognize the benefits of traditional methods and where they have their place, using them to their fullest. Simultaneously, explore the possibilities of GenAI in risk-free environments. Then you can consider how to use the two methods together, to leverage their unique and different strengths.
For example, you might extract information from unstructured data with GenAI, where it’s really strong, and turn that into structured data to feed your statistical model. In fact, we’re increasingly exploring such hybrid approaches at KX, because the best solutions may involve not making a choice between traditional and GenAI, but using them together.
Even that might be too much for some organizations – and I get the unease. It feels safer to stick with what you know, and costs less to run discriminative AI over GenAI. But there’s another cost to consider: the opportunity cost if you don’t experiment and invest in GenAI, which could lead to you missing out to rivals that were just that little bit more eager to try something new.
For more information on what it takes to build a successful AI program, read our AI factory 101 series. Discover why KDB.AI is a crucial part of the AI factory. Learn more at kdb.ai/learning-hub .
