Online Reporter
Takaidza Mabuto, [email protected]
ARTIFICIAL INTELLIGENCE (AI) can best be described as the ability of a machine to perform complex tasks autonomously. The technology has evolved rapidly, shifting from analysing existing information to actively helping create new content, a transformation that brings both enhanced capabilities and new risks. Understanding this shift requires a clear distinction between what is commonly referred to as “Old AI” (predictive) and “New AI” (generative).
Under predictive or “Old AI”, systems analyse existing data to forecast outcomes — for example, predicting whether an email is spam or identifying an image as that of a cat. In contrast, generative or “New AI” can produce entirely new content, such as writing an email or generating an original image of a cat. Generative AI refers to technologies capable of creating new text, images, audio or video based on patterns learned from large volumes of existing data.
The current Generative AI revolution, driven by widely used foundational models such as ChatGPT, Gemini and Claude, clearly illustrates the shift from predictive analysis to content creation. Tasks once thought to be uniquely human — particularly creative ones — are now being handled by machines. Computers, traditionally strong in mathematical processing but weak in artistic expression, now demonstrate both analytical and creative abilities through generative AI.
Leveraging the potential of Generative AI: the next productivity frontier
According to a McKinsey report, Generative AI has the potential to reshape how work is done by augmenting individual productivity and automating specific tasks. Its overall impact could add trillions of dollars in value to the global economy. Central to this transformation is an understanding of how generative systems function, particularly the role played by foundational models such as Large Language Models (LLMs), image generation models and multimodal models.
Generative AI foundational models
Foundational models are pre-trained AI systems with broad capabilities that form the basis for a wide range of generative applications. They are designed for adaptability, enabling developers to build specialised tools on top of them. Their defining features include pre-training on massive datasets, versatility across multiple tasks and transfer learning, which allows previously learned patterns to be applied to new problems.
There are different types of foundational models. Large Language Models support text generation and comprehension, image generation models produce visual content, while multimodal models handle multiple data types, including text, images, audio and video.
Generative AI is activated through prompts — text, images, audio or video — and produces outputs ranging from written content to code or visual media. Despite its versatility, generative AI still requires human oversight to verify accuracy. As the principle goes: trust, but verify.
Large Language Models (LLMs)
Large Language Models are trained on vast collections of text, enabling them to understand and generate human-like language. Systems such as ChatGPT or Gemini operate by predicting the most likely next word in a sequence, based on contextual probability. The model processes input prompts, interprets context and selects word probabilities derived from extensive training data.
While often perceived as intelligent reasoning systems, LLMs function more accurately as highly advanced auto-complete tools, drawing on patterns learned from large portions of publicly available text. In practice, marketing professionals and other content creators use these tools to generate first drafts of emails, blog posts or press releases, which are then refined and edited by humans.
Image generation models
Image generation models use similar principles to produce visual content. They are trained on large image datasets to learn patterns, after which they generate new images guided by prompts or sketches. This technology has found wide application in fields such as design, gaming and fashion.
Graphic designers, for example, use tools such as Midjourney to quickly generate visual concepts, significantly reducing the time required for ideation while retaining creative control at the editing stage.
Multimodal models: bridging text, image, audio and video
Multimodal models represent a more integrated approach, enabling AI systems to interpret and generate multiple forms of data within a single framework. These models align information across formats — matching text to images or converting audio to text — allowing more natural interaction between humans and machines.
Their defining feature is cross-modality interaction, which enables information to flow between different data forms. Beyond experimental demonstrations, multimodal models are now embedded in real-world applications. In healthcare, for example, they assist in combining medical imaging with clinical reports to support diagnosis and treatment planning.
Generative AI challenges
The growing adoption of generative AI has also introduced significant challenges. One major concern is the rise of deepfakes and misinformation, where realistic but fabricated images, videos or cloned voices are used to deceive. This underscores the growing importance of verification.
Intellectual property presents another unresolved issue. Questions remain over ownership: if an AI system creates a piece of artwork, does ownership belong to the user, the developer or the creators whose work formed part of the training data?
Another challenge is the so-called “hallucination” problem. Because generative models rely on probability rather than factual understanding, they can produce outputs that sound confident but contain inaccuracies or false information.
What next?
Generative AI represents a significant leap in technological development, transforming machines from passive tools into active creators.
By learning patterns from extensive datasets, these systems can produce text, images, music and designs that resemble human creativity. However, their outputs are the result of statistical modelling, probability and computational design — not independent thought.
Understanding how generative AI works helps clarify both its potential and its limitations. At the same time, it raises critical questions about ethics, originality and human agency.
As societies increasingly integrate creative machines into everyday life, the challenge will be to ensure they enhance rather than replace human imagination.
The question remains: “If machines can create alongside us, how will that reshape the meaning of human creativity?”
*Takaidza Mabuto is a Certified Artificial Intelligence (AI) Expert and Risk Management Practitioner with extensive banking experience.
derick.khumalo