5 min read
The case for Traditional AI: swimming against the current
The Generative AI hype
When ChatGPT launched with a bang in November 2022, this mainstreamed the concept of GPT-3 to the masses. And, while everyone has now heard of AI, there’s still quite a comprehension gap about transformer technology aka the “T” in GPT.
Transformer technology has been around for a while (remember the pivotal article from Google “Attention is all you need” in 2017?) but with much larger computational power, and unprecedented volume of training data from the internet, ChatGPT really put AI on the global map with 1 million users in just five days.
This is hardly surprising, given the staggering scale of capability. Generative AI tools like Midjourney, Sora, GitHub Copilot, have transformed content creation, software development, and user interaction. These systems use vast amounts of training data to generate new content (for the previous examples: images, videos, or code) that often mimics human creativity.
Whereas before AI was something exclusively for tech insiders and enthusiasts Generative AI has democratised the field by levelling down the entry barrier so that basically anyone with Google familiarity can now utilise it.
But this comes at a cost.
Generative AI challenges
Generative AI models are computationally intensive and expensive to train and run. They are also, as has been widely reported, prone to hallucination: producing plausible sounding but factually incorrect or nonsensical outputs. They are probabilistic and generative by nature (meaning they may not always produce the same output for the same input) and they can create challenges around trust, explainability, and reproducibility.
Ensuring data privacy is also a concern, particularly when models are trained on sensitive or proprietary information. Furthermore, their deployment raises legal and ethical questions around authorship, intellectual property, and accountability.
The environmental impact is another growing concern, as training and operating large-scale models require substantial computational resources, leading to significant CO₂ emissions and energy consumption.
Is traditional AI still relevant?
Traditional AI methods such as classification, regression, clustering, time-series forecasting, anomaly detection, and classical computer vision remain crucial tools. These technologies have been around for decades, and they continue to underpin critical applications across industries.
For example, GPS route planning often relies on deterministic algorithms and time-series predictions to estimate traffic patterns and travel time. Similarly, Netflix’s recommendation engine incorporates classical collaborative filtering and clustering methods to group users and suggest relevant content. These systems are valued for their interpretability and predictability, which are essential in regulated environments or when transparency is required.
Determinism and interpretability
Unlike Generative AI, traditional AI is often deterministic. This means that given the same input, it will always produce the same output, making it more predictable hence, in many cases, more explainable. Techniques such as decision trees, logistic regression, and support vector machines offer greater transparency, which are invaluable in regulated industries such as finance, healthcare, and legal services.
Traditional models also require less computational power and data compared to their generative counterparts. While Generative AI excels at handling unstructured content such as text or images, it often functions as a black box, making it harder to trust or audit. In contrast, when clarity, accountability, and efficiency are essential (for example, in risk modelling or medical diagnostics) then traditional models remain the go-to.
A greener, cheaper alternative
Another major advantage of traditional AI is its efficiency. Generative AI models require massive amounts of energy for training and deployment. A single GPT-style model can cost millions in compute resources and produce a sizeable carbon footprint. In contrast, traditional models can be trained on a laptop or even in-browser, making them far more sustainable.
For many small to mid-sized enterprises (SMEs) or budget-conscious departments within larger organisations, this cost difference is not trivial. Traditional AI makes machine learning accessible without requiring massive hardware infrastructure.
Forecasting
Classification and regression
Anomaly detection
Clustering
Computer vision
These applications are not only still relevant: their demand is growing as businesses aim to optimise processes and reduce costs.
It’s not either-or: embracing the full AI toolkit
Not all problems are generative in nature. Trying to use a language model to predict sales figures or detect fraudulent transactions is a poor fit, both technically and financially. Generative AI is very powerful in creative and unstructured contexts; traditional AI in data-driven environments where patterns and historical data guide decision-making. Understanding the difference in capabilities is key for selecting the right tool for the task, ensuring efficient use of resources, and delivering reliable, explainable outcomes that align with business goals.
The good news is we don’t need to choose between traditional and generative AI. Both have their place. In fact, they are often more powerful when used together.
An A(I)genetic customer service bot might use generative AI to converse naturally, but rely on traditional AI to route tickets, detect sentiment, or predict escalation risk.
This hybrid approach ensures that organisations get the best of both worlds: the innovation of generative AI and the reliability of traditional models. To make this work, both IT professionals and business leaders need a strong understanding of the full AI toolkit, not just the latest shiny object.
Answering these questions helps ensure that the AI strategy is fit for purpose, not just trendy.
Conclusion: swim with purpose
There’s nothing wrong with riding the wave of generative AI. Millions of us are around the world. It’s a powerful tool that will continue to evolve. But sometimes, the smarter move is to swim against the current, especially when the problem calls for precision, speed, clarity, or control.
Traditional AI may not get the same spotlight, but it’s still doing some of the most important AI work today. It also tends to be more demanding in terms of implementation, requiring skilled data scientists to develop and maintain effective models. However, its rigour, reliability, and interpretability make it deserving of just as much attention, if not more, than its generative counterpart.
Rather than pick sides, organisations should embrace both aiming for smart, problem-driven decisions that balance innovation with impact.
Filippo Sassi is Head of the AI Labs at Version 1.
For more insights on how our AI expertise is driving innovation and helping businesses around the world transform, click here.
Latest AI blogs
-
Blog
Bridging the AI gap
