New frontier: Generative AI is rapidly advancing and generating responses that resemble those of humans. – 123rf.com
Intelligence and learning are closely related in our modern, fast-paced world. Understanding how we learn is essential for education, human development and our pursuit of a knowledge-based economy.
With the advent of artificial intelligence (AI) and machine learning, we are gaining new perspectives on this timeless phenomenon.
There are three primary schools of thought on human learning: behaviourism, cognitive theory and social learning. Each school offers a unique perspective on knowledge and skill acquisition.
Learning occurs through interactions with the external environment, as emphasised by behaviourism. For instance, when children touch a flame and feel the searing pain, they instinctively withdraw their hands. Repeated experiences condition them to avoid flames. In this paradigm, learning is evidenced by observable behaviours, such as children consistently avoiding flames.
On the other hand, cognitive theory delves into the internal mental processes that occur when external information and stimuli are processed. For instance, when children touch a flame, their brains connect fire and pain. They use this information to develop a general understanding of all flames, including larger ones like those in a fire. This way, children learn about the hazards associated with fire. Their thoughts, described as “thinking about thinking”, guide their external behaviours.
The social learning theory emphasises the role of social factors in shaping behaviour. For instance, children may learn about the dangers of fire from their parents and observe their friends avoiding flames or experiencing distress after touching them. In this manner, children learn to avoid flames without necessarily having to interact with them directly.
It is important to note that these three learning theories – behaviourism, cognitivism and social learning – are not mutually exclusive and often overlap, depending on the learner, subject matter, environment and context.
AI levelling up
When we examine AI, specifically generative AI such as ChatGPT, it primarily aligns with the behaviourism learning theory. ChatGPT receives prompts and generates responses based on learnt patterns. The consistency in its responses indicates a form of learnt behaviour.
There are also hints of cognitive learning theory as ChatGPT can evaluate its answers, similar to “thinking about thinking.” However, it is still debatable whether it truly comprehends information in a human-like manner or whether this is an advanced form of behaviourism rather than cognition.
ChatGPT lacks the capacity for social interaction and contextual experiences, which makes it unable to engage in social learning, unlike humans.
When it comes to learning, it is not only about the method, but also about the content. One commonly used framework for classifying knowledge is Bloom’s Taxonomy, which categorises knowledge into various domains and levels of complexity.
The cognitive domain, for example, has six progressive levels: “Remember”, “Understand”, “Apply”, “Analyse”, “Evaluate” and “Create”.
According to current research, AI programs such as ChatGPT have achieved the “Apply” and “Analyse” levels of Bloom’s Taxonomy. However, there is still uncertainty about whether they truly understand information like humans do.
In his book The AI Delusion, Prof Gary Smith argues that computers can store and retrieve information but lack genuine comprehension of words and numbers. He compares this to Nigel Richards, a world champion in French Scrabble who doesn’t understand French but can memorise foreign words without understanding their meanings.
The evaluation of whether learning has truly taken place in AI depends on the perspective one adopts. According to behaviourism, if AI gives consistent responses, it has learnt. However, from a cognitive standpoint, the learning may seem incomplete without genuine comprehension.
Nonetheless, generative AI has made remarkable progress and appears to possess a rudimentary form of cognition. When asked its ability to respond based on Bloom’s Taxonomy, ChatGPT acknowledges its limitations.
Generative AI is rapidly advancing and generating responses that resemble those of humans. The Turing Test, a benchmark for human-like intelligence, is often surpassed by generative AI, compelling us to reevaluate our understanding of cognition, intelligence and humanity in this ever-evolving landscape. Machines may not learn in the same way as humans, but this does not negate the emergence of intelligence.
Dr Wong Teik Aun, a principal lecturer at the Centre of Australian Degree Programmes at INTI International College Penang, enjoys writing on subjects close to his heart and has published On the Beaten Track Nepal: The Himalayas, Symphony and Synchrony: An Orchestra of Ideas and Tales of Animal Lovers. He has also contributed numerous academic articles to internationally ranked journals. The views expressed here are the writer’s own.
