Beyond the Hype: Navigating AI Tools for Business Success
While generative AI has undeniably captured the spotlight with its widespread accessibility and innovative applications, it's crucial for businesses to understand when to leverage its capabilities and when to turn to other powerful AI tools, such as traditional machine learning. Less than five years ago, machine learning was the dominant form of artificial intelligence in business. However, the release of ChatGPT-3.5 in 2022 shifted many organizations' focus to generative AI, a subfield capable of creating new content.
Traditional machine learning is now a well-established technology within many businesses. Leading firms are now actively exploring use cases for generative AI, with a 2024 survey of senior data leaders indicating that 64% believe generative AI has the potential to be the most transformative technology of a generation. Yet, despite its widespread accessibility and novel applications, discerning when to utilise traditional machine learning remains a vital skill for businesses.
To shed light on this evolving landscape, we consulted with experts in AI regarding the interplay between generative AI and traditional machine learning, specifically exploring instances where generative AI might supersede predictive machine learning, when machine learning remains the optimal tool, and how these technologies can be effectively combined.
Unpacking Machine Learning
Machine learning, a branch of artificial intelligence, empowers computers to learn without explicit programming. Unlike conventional computing, which requires detailed, step-by-step instructions, machine learning programs can learn from examples. This technology finds diverse applications, from forecasting customer behaviour and detecting potential fraud in banking transactions to curating personalised search results on retail websites.
The data that fuels machine learning models, including generative AI tools, can take various forms, such as numerical spreadsheets, text, images, audio, or video. The accuracy of a machine learning model improves with the quantity of data it is trained on. For machine learning to function effectively, the data must contain identifiable and analysable patterns.
As one expert aptly put it, "The basic idea of machine learning is it's a lot easier to collect data than to collect understanding." For instance, it's simpler to provide a machine learning program with thousands of labelled images of cats and dogs rather than attempting to programmatically define all the intricate distinctions between them. This process of feeding labelled data enables the program to learn the differences autonomously. Another expert added that machine learning "makes decisions that generalise patterns that we would not have found otherwise." However, its effectiveness is contingent on the quality of the data and the models available. Consequently, machine learning is best suited for scenarios with abundant data, such as vast quantities of customer conversation recordings, sensor logs from machinery, or ATM transactions.
Demystifying Generative AI
Generative AI represents a more recent iteration of machine learning, capable of producing novel content, including text, images, or videos, drawing upon extensive datasets. Large Language Models (LLMs), which are AI programs designed to process and generate text, are a prominent example of generative AI. OpenAI's ChatGPT, released in 2022, rapidly gained traction due to its impressive ability to respond to user prompts in plain language and swiftly generate new content. Other widely used chatbots and LLMs include Anthropic’s Claude, Google’s Gemini, Microsoft’s Copilot, and Meta’s Llama, all of which have seen recent updates to enhance their accuracy and responsiveness.
An expert explained that while "machine learning captures complex correlations and patterns in the data we have," generative AI "goes further." Fine-tuned, specific generative AI models can uncover relationships within traditional datasets that conventional machine learning cannot, providing a distinct advantage.
Instead of merely making predictions or identifying patterns, generative AI excels at creating new content—it can answer questions, compose emails, or brainstorm ideas. One expert highlighted the numerous use cases for GPT models, noting that "You see a lot of companies trying to find a way in which they can use them within their own frameworks, be it to transcribe calls in a call centre, navigate policy documents, or help new employees learn the company’s existing software code." However, a word of caution was extended regarding potential issues like inaccuracies and bias when developing or utilising generative AI or machine learning.
Optimal Applications for Generative AI
Beyond its primary function of generating new content, generative AI is increasingly taking over tasks traditionally handled by machine learning. These scenarios include:
Handling Everyday Language or Common Images: LLMs, trained on vast amounts of text or images, can be used "off the shelf" for classification and detection. For example, a company wishing to analyse online product reviews for defects might once have built a custom machine learning model, a time-consuming and costly process. Today, product reviews can be inputted directly into an LLM to identify product improvement insights. GPT-4 and similar models can offer greater accuracy than custom-built machine learning models and enable quicker application deployment. Furthermore, generative AI models are becoming more affordable, increasing their accessibility to a wider range of companies.
Seeking More Accessible Options: Many software engineers can utilise generative AI models without extensive additional training, whereas building machine learning models demands specialised technical expertise. This makes generative AI a "democratising force" by significantly increasing accessibility. If a problem or opportunity involves everyday information, the advice is to "try generative AI first" and avoid defaulting to traditional machine learning.
When Traditional Machine Learning Remains Best
Despite the rise of generative AI, traditional machine learning remains the superior choice in certain situations:
Addressing Privacy Concerns: Caution is advised when feeding proprietary, sensitive, or confidential information into LLMs due to the potential for data leaks. While building private models is an option, it requires specialist technical skills that may not be readily available within an organisation. In such cases, reverting to "the old-fashioned way" of machine learning might be preferable.
Utilising Highly Specific Domain Knowledge: LLMs are trained on widely available data and are adept at handling everyday information. However, their accuracy may diminish for highly technical or niche tasks, such as medical diagnoses based on MRI images. For domain-specific problems involving extensive technical knowledge, jargon, or issues unique to a particular company, the traditional machine learning route is generally recommended. It's worth noting, however, that generative AI models are continuously improving, and this dynamic may evolve over time.
Leveraging Existing Machine Learning Models: Organisations have invested considerable effort in developing machine learning programs for specific applications, such as detecting potential credit card fraud. In these instances, there is often no urgent need to replace them with a generative AI system. The key decision point arises when considering new use cases and opportunities.
Synergies: Combining Machine Learning and Generative AI
In several scenarios, machine learning and generative AI can be integrated to achieve enhanced outcomes:
Augmenting Machine Learning Models: Algorithms are limited by the models provided to them. By offering more context through generative AI, these algorithms can be improved. For example, a machine learning model might predict cardiac fitness from a dataset of names, heart rates, and running speeds. Generative AI-augmented machine learning could extract additional insights from the person's name, inferring age and other demographics by drawing on external context.
Simplifying Machine Learning Model Design: Generative AI tools can be used to facilitate the design of machine learning models. By inputting data along with desired functions and techniques, a generative AI tool can build and evaluate models on different datasets and report on their accuracy. While generative AI is transforming the workflow for machine learning practitioners, continuous analysis and critique of model outputs are essential to prevent the compounding of hallucinations and errors.
Generating Data for Machine Learning Models: When insufficient data exists to adequately train a traditional machine learning model, generative AI can be employed to create synthetic data, which replicates the statistical properties of real-world datasets.
Preparing Structured Data for Machine Learning Models: Tabular data, particularly in industrial settings, often contains errors like missing values that require preprocessing before model training. Instead of manual cleaning, this data can be uploaded to an LLM with a prompt to identify anomalies or mistakes. Generative AI streamlines the entire traditional machine learning workflow, from data procurement and cleaning to actual modelling, acting as a "turbocharger." However, this efficiency comes with a caveat: the need for constant vigilance to ensure the accuracy of LLM-generated outputs.
Given the diverse array of AI tools available, mastering the art of selecting the appropriate tool is becoming another essential skill for AI practitioners. As one expert concisely summarised, "If you want to generate stuff, use generative AI. If you want to predict things but with everyday stuff, try generative AI first. If you want to predict things on domain-specific stuff, do predictive stuff, [use] traditional [machine learning]. It’s as simple as that."
Disclaimer: The content provided herein is for general informational purposes only and does not constitute financial or investment advice. It is not a substitute for professional consultation. Investing involves risk, and past performance is not indicative of future results. We strongly encourage you to consult with qualified experts tailored to your specific circumstances. By engaging with this material, you acknowledge and agree to these terms.