We probably remember the discussions around GPT a couple of years ago that it cannot even solve simple math problems such calculating the time taken by 4 cars to cover a specified distance at a given speed. The GPT multiplied the time taken by each car, and gave an answer four times the correct answer. Its capacity for quantitative reasoning was poor.
Over the next two years, the developments have been tremendous. The GPTs can now perform complex calculations, write thousand lines of codes, write essays, and even ‘guide’ learning! While teachers have been impressed by these developments and are breathlessly trying to keep up to ever-increasing sophistication of AI and the applications that leverage AI, it is natural that there would be a number of questions in many minds: “Given the advancements in AI, what is my role in education as a teacher? Will there be no teachers a few years from now? What will be the future of education, educational institutions?”.
While there is no doubt that educational institutions will undergo a sea of change in their modes of functioning and the modern learning process will undergo a paradigm shift, in this article we outline the limitations of AI and why a human teacher will always be quintessential in the educating of the young. It is true that the slogan that AI cannot replace teacher has been around but without much explicit articulation and has not kept the ‘fears’ of a teacher at abeyance.
This article is an attempt to alleviate the fears of a teacher and distinguish between those functions of teachers that AI can perform better than teachers, and those functions that it may not be able to function in the immediate future. This would mean a careful consideration of the function of teachers in classrooms and outside the classrooms.
GPT models have been trained on vast amounts of data, primarily textual, to learn ‘patterns’ in the data. An oft quoted example to explain this is as follows: Take the sentence “The King and the ______ left for their summer palace”. It is easy to guess that the most likely word in the blank is ‘queen’ because this phrase almost always occurs as such in many stories and books. It is such correlations between words and co-occurrence of words in the textual corpora that the GPT models ‘learn’.
The example above is only meant to develop an intuitive understanding of how the GPT model learns – in reality, the surface patterns of correlations and co-occurrences it learns are much more complex, thanks to the developments in the AI algorithms. It is with this it is able to synthesize a large amount of text and ‘organize’ it when given a prompt.
In AI, an intent is the user's underlying goal or purpose behind their query or action. The way current AI functions, it has no intent. Only an entity with mind has an intent and goal. And intent is what we humans bring in when we provide prompts.
Imagine you are teaching a class of 60 students and are aiming at designing a learning task that requires the students to think critically about what you are engaged with. In designing your task, you keep in mind possible responses of the students, your immediate educational goals embedded within an educational philosophy, the background knowledge and understanding of your students, your prior experience in teaching (the same set of students and other sets of students).
This is a complex design task – it is complex because there are multiple goals and constraints to be satisfied simultaneously. The level of creativity and strategic thinking required in the design task, drawing from prior teaching experience, aiming at a novel educational goal, anticipating student responses is beyond the reach of AI simply because it is only trained to synthesize from existing data, not design creatively with intent. This is where it cannot replace a human – the human touch in teaching is a creative and strategic role, not simply that of delivery a textbook. This is also a wakeup call for teachers to move from simply donning the hat of deliverers of knowledge to that of developing the higher-order capabilities of students.
While there is considerable research in the world to model how the mind-brains of humans interpret other people’s thoughts, make moral judgments, and comprehend their belief systems [1], at least for now, it does not seem that the existing AI models are capable of replacing a teacher.
[1] https://betterworld.mit.edu/spectrum/issues/fall-2009/theory-of-mind/
