Art, science and reality
It is 1850. We are in Paris surrounded by an angry mob outside the deep red building in Boulevard des Capucines in which Louis Daguerre perfected, a few years back, the recording of a scene as a permanent image. The address is now the studio of Felix Nadar, the foremost exponent of the new art – or is it science – of photography. Society is outraged, of course. It always is. Science has broken through the sacred walls of creative culture. A photograph by Nadar, whose portraiture is taking Paris by storm, is surely the end of Art; the end of human creativity as we know it.
Charles Baudelaire, the French man of letters who later became the subject of some of Nadar’s greatest pictures, railed against the new image-making machinery: “A revengeful God has answered the supplications of the multitude. His Messiah was Monsieur Daguerre. And the multitude said: ‘… Art means Photography’ .” Now instead, we listen to the fulminations of writers, scientists, pundits, politicians and even (if misquoted) its inventors against ‘Artificial Intelligence’ or AI.
I have been following the development of artificial intelligence for many years, and I can confidently tell you that the systems being hyped right now as AI bear as little relation to human intelligence as a chess playing robot does – even though it can easily beat me. Large Language Models (LLMs) like ChatGPT take a brute force approach to knowledge in an analogous way to Deep Blue, IBM’s first chess champion, which can look thousands of possible moves ahead and thus always beat a human player, even a Grand Master, who can think only 15-20 moves ahead.
An LLM simply chooses the most suitable next word from the mass of text it has absorbed, applying algorithmic rules to recognize the relationships and connections between words and to make the text coherent, relevant and structurally accurate. But ‘garbage in, garbage out’, as the old programmers’ dictum goes. If the LLM has encountered incorrect information, or not encountered the exact information requested, it will produce a bland mishmash of text, grammatically accurate but lacking insightful content. The outcome is not literature, just the findings of a super search-engine. Believing you will get mystic insights is like expecting a shovel to distinguish between the species of worm it cuts in half as you press it into the rich soil of knowledge. It is a tool, a useful one, but not the end of literature or the death of education or the destruction of the professional classes.
How can I be so sure ? Because of the way Art responded to photography. Art exists only within the human mind, which quickly found new ways to represent reality. Thus, impressionism was born – an entirely new creative approach, enriching Art in unexpected ways. Humanity as a creative force exploded into new fields. Vorticism, Abstract Expressionism, Pop Art. Would we have enjoyed Dali or Picasso, Pollock or Hockney without Daguerre and Nadar ? Photography was not the end of Art but a catalyst for the diversification of Art.
So, is ChatGPT and its fellows the end of literature or musical composition ? The analogy is clear. Anyone who has used LLMs in earnest as I have, will have recognised the value that such tools have. But they are just tools. They can be plain wrong, even if the grammar and syntax is perfect. But worse, LLMs tend to produce results which lack – what can we call it ? – oomph !
A piece of text, or a structure for a piece, generated by an LLM can give you a good start, but, like a mediocre photograph, it does not engage the human reader. Instead, just like photography, the skill is in setting up the scenario in the first place, like framing a shot, and then in taking the raw material and making something worthwhile out of it – the editing. When we look at the crisp moving images on a display of TV sets in a department store, those pictures are not the originals, snapped on a mobile phone out of a moving vehicle. No, they have been adjusted, squared up, colours enhanced, crispness perfected, the sweep of a green valley subtly changed by humans adding a picturesque barn here, a cow there and removing all trace of modernity. The edited version is the one that captivates you and seduces you into buying the latest screen technology that you could live without.
Exactly the same is true of LLM-generated text. If you ask a poorly framed question, you will not get a helpful answer. If you ask the right question but just present the answer raw, it will look like a raw photograph. All the telegraph poles and wires are still there. All the dull colours are left as they are. There’s no pretty cat at the focal point. The secret is in the framing and the editing, which are the parts that require human skill. The technology will likely spur literary creativity in new directions, just as photography did to Art. We underestimate the power of the human mind to adapt and innovate.
The touted ‘AI’ technology of today is a step forward in the use of powerful computing techniques and a valuable addition, if used sensitively, to the toolkit of the professional writer, which includes most professionals. But it is not intelligence in any human sense, nor is it threatening in the way some commentators suggest. Technology, especially information technology, has never resulted in less employment overall. This step is no different. It will take the drudgery out of some tasks. That’s about it.
Nevertheless, there is one risk to guard against and that is the protection of truth. Written material in circulation contains enormous amounts of unsupported opinion parading itself as truth. LLMs can be exposed to a huge amount of garbage. This is where reality comes in. If an LLM is not epistemologically accurate, it will come up with appalling distortions of reality. Furthermore, it must be kept fully up to date with scientific advance. Hence proposals recently passed by the European Parliament for objective review of LLMs before they are released on an untutored public are necessary and will likely influence global markets.
Meanwhile, the quest for ‘real’ AI will go on. It will be a long one.
Dr Gordon R Clarke
Managing Director
Monetics Pte Ltd, Singapore
First published in Singapore Business Times 4 November 2023