Something is rotten in the state of Denmark.
Hamlet, Act-I, Scene-IV
This non sequitur, put into the mouth of Marcellus by the ‘the Bard’, goes beyond the mere observation that something is not right. It is a literary device, spoken by a run of the mill guard, alluding to the notion that even those who are not part of the ruling elite are still acutely aware of the problems and corruption which abounds in the upper echelons of society. A similar state of affairs has long stalked academia and is evident in conversations with those who are not closely involved, some of whom have never attended a university, yet are nonetheless aware that not all is well on campus.
I have mused on this topic before in Journey’s End and Reading More Efficiently And Effectively and will not rehash these elements of the argument here — though will return to them again one day. Rather it is to surface the landscape with which I am regularly faced when contemplating AI — most famously at present the impact of chat bots such as ChatGPT.
First up, it is perhaps worth examining what ChatGPT is and what it is not. While many are classifying it as AI, it is after all built by OpenAI, it is not artificial intelligence. This is because ChatGPT is not able to reason. No need to take this assertion from me, if you place your faith in technology, you can have the answer right from the Chatbot:
ChatGPT is a machine learning model that has been trained on a large dataset of text. It can generate human-like responses to text input, but it does not have consciousness or self-awareness like true artificial intelligence. It is a sophisticated tool that can perform certain AI-like tasks, but it is not truly AI in the sense that it is not capable of truly understanding or experiencing the world.
ChatGPT Response
This is a critical piece of the puzzle to grasp before we go down the rabbit hole, particularly in the space of academia and exams. For the ends of learning are not to see how much we have remembered, much less to discipline the appetites. Rather, as Evelyn Waugh would have it, ‘the purpose of education is to train the moral perceptions’. Yet all too often, students grasp for that which gets them through the course — and only that. Given engaged students only retain about 20% of the material they study, unless we can lift expectations of our students above exam results, we are not educating humans but, in a sense, building chat bots.
This feeds in to why ChatGPT is not true AI, because while it can use statistical and language models to generate text based on patterns it has learned, it is not able to understand what it generates, much less apprehend the implications of its responses. The sad correlation is that institutions, rather than turning out intelligent graduates capable of truly understanding their subject matter, are increasingly training people who have some recall of information and who use statistical and language models to generate responses based on patterns learned. Ultimately, with little understanding of what they generate.
What makes a human intelligent is not that we are able to learn the language pattern ‘fire, fire!’, but that we are also able to comprehend what the implications are of us shouting fire in a crowded cinema — when there is no fire danger. This enables us to articulate language from which we can draw moral conclusions and to make value judgements about future behaviour.
These two drawbacks, that ChatGPT cannot reason, nor can it draw inferences from the answers it generates, should be enough to dispel the near hysteria that is growing about the platform. Yet, that so many student essays display, at best, poor reasoning, and when cold called, students often have an inability to draw value judgements on the topic in quo, explains why some are lamenting that The College Essay Is Dead. This is because declining standards cause some professors to deem the output of Chatbots to be graduate-level, or at least akin to what their students produce, and thus creates an environment of fear that:
A student can generate an entire essay in seconds, at a cost of around 50 US cents. Equally worrying, would-be academic researchers can call on these systems to generate articles for submission to conferences and journals.
Sharples, 2022, p. 1120
I have shortened the output considerably, lest your soul ebb away in the face of inane prose, but to give a flavour of what ChatGPT is able to generate when asked to write an essay:
The construct of “learning styles” is problematic because it fails to account for the processes through which learning styles are shaped. Some students might develop a particular learning style because they have had particular experiences. Others might develop a particular learning style by trying to accommodate to a learning environment that was not well suited to their learning needs. Ultimately, we need to understand the interactions among learning styles and environmental and personal factors, and how these shape how we learn and the kinds of learning we experience. Learning Styles and Learning Modes In the 1970s, learning styles became an important feature of educational practice. In the 1990s, “learning modalities” became a major focus of learning style research. This is the idea that we process information in different ways…
In a review of this research, Dunn and Dunn (1997) found that the evidence does not support the idea that people have a particular learning mode that is dominant or that they process information in a particular way.
References
Dunn, R., & Dunn, K. (1997). Learning styles: Research and practice. Journal of Research in Education, 7(2), 139–151.
Yet dig below the journalistic clickbait and the researchers and professors examining the first rumblings of the AI Essay phenomenon acknowledge:
But look more closely and the paper falls apart. It references “Dunn, R., & Dunn, K. (1997). Learning styles: Research and practice. Journal of Research in Education, 7(2), 139–151.” There is a journal named Research in Education, but no issue 7(2) in 1997. Dunn & Dunn did publish research on learning styles, but not in that journal. GPT-3 has fashioned a plausible-looking but fake reference. The program also appears to have invented the research study it cites. We can find no research study by Dunn and Dunn which claims that learning styles are flexible, not fixed.
Sharples, 2022, p. 1122
And that is to say nothing about the appalling — I shudder to dignify the passage with the word argumentation. However, when a qualified academic spends more time on the ChatGPT output, a much better ‘fake’ of an ‘academic paper’ can be achieved.
If you did not read the link above to the better ‘fake’, I can vouch that the more time the operator spends honing the ChatGPT output, the closer they can get it to the sort of turgid output that bodies forth from so many academic journals. Yet for all the bad prose, there is one critical element needed in this faux-literary process: a person to refine the AI output. Without this, without an understanding of what can and cannot fool an editorial panel, the whole endeavour would fail at the first editor’s review.
In time, of course, the process will improve to the point at which people who are functionally illiterate will be able to produce academic papers, but what of it? Most output sinks without trace, and that which is dug up in the future and cited is not true or false because of the authorship — at least it should not be. A universal truth is a universal truth, whether rendered on tablets of stone by the hand of God or vomited out by an amoeba. However, such speculation is unlikely to still the fluttering pay cheques of the academic Cosa Nostra who eat or starve by the quantity of their output.
A more ‘progressive’ argument that has been put forth for the benefit of chat bot generated essays is that they could democratise cheating. Given 15.7% of students admit to paying to have their assignments written, ChatGPT represents the opportunity to level the field for those who cannot afford to support the $100 million dollar a year industry that is contract cheating. I find that a nihilistically ‘levelling’ view which, like most outlooks prizing equality of outcome, benefits few and ends up failing everyone else.
Hopefully, and more optimistically, there will be the unintended consequence of forcing professors to rethink how they assess students. It might even, hold on to your seats, result in more student engagement because instead of set and forget essay topics, increasingly imaginative ways to gauge student capabilities will be needed. Spoken examinations, or viva voce, could be more than the preserve of PhD candidates, and used to demand a level of rigour that would benefit society inestimably by turning out graduates who can actually contribute. Something that would be in stark contrast to the growing crop of employables who merely sit, collect a paycheque, and hope no one sends a question or anything approaching real work in their direction. A sentiment which most hiring managers I speak with hold even more intemperately.
Power Exchange also, in these Foucauldian times, surfaces much agitation. Bringing to the fore ways in which it is feared AI will reinforce power imbalances:
When the field of AI believes it is neutral, it both fails to notice biased data and builds systems that sanctify the status quo and advance the interests of the powerful. What is needed is a field that exposes and critiques systems that concentrate power, while co-creating new systems with impacted communities: AI by and for the people.
Kalluri, 2020, p. 169
I will skip over the moronic notion that those who are powerful are somehow not part of the ‘the people’, and instead observe the canard that bias only exists when power is concentrated or when X group is not in power.
The reality is that truly equal systems, or systems that provide no bias or benefit to specific groups or individuals, will ALWAYS benefit the powerful. This is because of the Matthew Effect, in that if you line up 10 people and Usain Bolt in his prime, and give them an equal opportunity to run a 100 meter race, Bolt wins every time. This is not because the race is biased, it is because in any given endeavour there will always be people who are better equipped to succeed. In these endeavours, running, writing, climbing, making money — pick any field you like— those with the requisite skill sets will tend to succeed more often than those without.
The only way to shift the ‘power imbalance’ is to handicap the powerful. In other words, to build that which people seem to be campaigning against — a system that is biased. But one which is biased against anyone who would, unchecked, pull ahead of the crowd. Only in this way can equality of outcome be achieved. While this is often considered optimal when those handicapped are deemed ‘undesirables’ — billionaires, white colonialists, etc. — the problem with ‘others’ is the revolution soon runs out of them and turns to us, ‘the people’. Don’t believe me? Please see pretty much any revolution in history — French, Russian, Chinese, Hitler’s Nacht der langen Messer.
The reason I cite these examples is because we do better in our thinking about AI once we acknowledge that it, like all the other technologies in history, is unlikely, no matter how much thought and care is put into its development, to redress the millennia old problem of haves and have nots. We are on a better path to look at the real underlying societal problem. A problem that already has and, as more people jump on to the AI driven writing bandwagon, will more pressingly emerge — poor literacy.
It is ironic that world literacy is at the highest level in human history, yet this literacy is at best functional and thus something that is primed to be replaced by AI. Simply speak into your device, assuming you are privileged enough to have one, and the crude output of the current crop of chatbots will already exceed what passes for bog standard student essays these days.
My old professor, who began lecturing more than half a century ago, spent a career monitoring the literacy levels of post-graduate students. The results, which are almost impossible to get published in an age that prefers to celebrate progress — so long as it is progress for approved in-groups — showed a steady decline in linguistic capability among post-graduate cohorts since the 1970s. It is no coincidence that this is also the period which has seen the rise of the machines and development of word processors that do to words, to quote Stephen Fry, ‘what the KRAFT company does to cheese.’ Other relatively marginalised studies which examine contract cheating, show that in almost lock step with the general decline in literacy is an uptick in the ‘Essay Mill’ industry.
This is alarming not because students are gaining qualifications with only the most rudimentary understanding of the course material, but because students are not achieving that which is the purpose of the course — experience of subject. As Burke observed: ‘to learn for show is like painting for complexion, it looks tawdry, lasts not long, and is no better than a Cheat’.
Ultimately, I do not think the problem is that we are building machines so ‘smart’ that their output is indistinguishable from human compositions. The problem is that we are educating humans whose output is so illiterate, and devoid of experience, it is indistinguishable from computer generated text. In such a context we seem to be failing our race by educating generations of people who are, to quote the failing that ChatGPT observes about itself, ‘not capable of truly understanding or experiencing the world’.
Pablo Picasso noted: ‘The trouble with computers is that all they give you is answers.’ I for one hope the fear about AI taking over our jobs, and the world at large, will prompt the realisation that the purpose of language is not to give answers, but to convey ideas, glean experience, and make value judgements. A process that enables us to experience the world as never before, because we go beyond experiencing as individuals and instead experience as a society. For it is not our answers that make us human, it is our collective experiences.
Good night and good luck.
Photo by Patrick Fore on Unsplash.
References
Kalluri, Pratyusha. “Don’t Ask If Artificial Intelligence Is Good or Fair, Ask How It Shifts Power.” Nature 583, no. 7815 (July 2020): 169–169. https://doi.org/10.1038/d41586-020-02003-2.
Newton, Philip M. “How Common Is Commercial Contract Cheating in Higher Education and Is It Increasing? A Systematic Review.” Frontiers in Education 3 (2018). https://www.frontiersin.org/articles/10.3389/feduc.2018.00067.
Sharples, Mike. “Automated Essay Writing: An AIED Opinion.” International Journal of Artificial Intelligence in Education 32, no. 4 (December 2022): 1119–1126. https://doi.org/10.1007/s40593-022-00300-7.
