Human artists feel threatened by AI – perhaps they shouldn’t just yet

Artificial intelligence (AI) is quite the rage these days – and predicting it means the decline of everything from critical thinking to the art of photography seems to be even more so. This Cybernews reporter decided to give AI art generator DALL-E2 a try, to see what all the fuss was about.

I won’t lie, I was excited. OpenAI’s text-in, art-out tool sounded like a lot of fun. As someone whose artistic career went into meltdown upon being diagnosed as color-blind around the age of 11, I was genuinely keen to see if a machine could generate the kind of picture I could only dream of crafting with these all-too-human hands.

My excitement built as I signed up for a free trial of DALL-E2. For absolutely not a broken penny, I was free to generate thirty images using OpenAI’s tool, released a couple of years ago. The only limitation would be my vocabulary – and unlike images, words are definitely firmly in my wheelhouse.

Now, while DALL-E2 might claim to “create original, realistic images and art from a text description,” OpenAI also seems quite keen to get users to opt for the surreal. An Astronaut Riding A Horse In Photorealistic Style is the chosen flagship image for its website, and it does indeed show just that.

AI fumbles the basics

But I wanted to get DALL-E2 back to basics first. When I was growing up, and so-called modern art was making waves, there was a counter-movement that suggested that before one learns to run, the art of walking should first be mastered. That is to say, before spraying a canvas with a smörgåsbord of seemingly random colors Jackson Pollock-style, the artist should prove themselves by demonstrating familiarity with the basics: conventional paintings that demand attention to pesky but important things like composition, perspective, and realism.

I decided to pose this one to DALL-E2. A teddy bear riding a bicycle in Brooklyn is all very well, I thought, but what about something more old-school? I was still keen – surely an AI machine capable of rendering the surreal could easily give me something as obvious and established as, say, a print of daily life in Georgian London in the style of contemporary visual satirist Hogarth?

Alas, I was doomed to be disappointed. Typing in “An 18th-century black-and-white print in the style of Hogarth,” I waited feverishly while DALL-E2 went about its work. Four images were returned shortly after. First off, none of them remotely resembled Hogarth’s celebrated style – a bit of a downer considering this is an illustrator who lived and died some three centuries ago.

Worse still, the figures portrayed – although broadly dressed accurately in contemporary style – all seemed to lack faces. My efforts to get DALL-E2 to give me something acceptable from the same period by refining my search revealed ‘art’ even more offensive to the eye.

When I typed “A Georgian gentleman in a coffee house in London”, I was hoping to get, well, something along those lines. What I did not expect to receive was “A Georgian gentleman recovering in a coffee house in London after undergoing a botched facial vivisection.”

Let’s face it, AI can’t paint

The problem of the human face – by no means exclusively but so often the central subject of paintings, sketches, and photos – recurred throughout my experiment with DALL-E2.

The first return for “A cinematic still of a period film from the 1950s” was a black-and-white number of, at first glance, a fairly convincing Humphrey Bogart-type tough guy in a suit. But on closer examination, the poor fella appeared to have suffered a stroke: the left side of his face drooped, arguably making him look more menacing, but certainly not what I’d asked for.

The next image was even more horrendous. An otherwise ordinary-looking young woman appeared to have suffered third-degree facial burns, leading me to wonder at this point whether DALL-E2 hasn’t in fact arrived at the Singularity by way of cultivating a pretty sadistic sense of humor.

My request for “A black and white photo of a classic rock band from the 1970s” met with similar results that made the Ramones in their prime look pretty. Sticking with the postwar counter-culture theme, I went for “Hippies at a music festival in the 1960s,” and got more mutant men and women, in an apparent sick twist on the old dropout mantra of “be in the place and get off your face.” This time, even the backdrop was wrong: despite my date-specific request, the tents in the background were clearly of modern design.

“A horde of bloodthirsty Saxon warriors invading Celtic Britain” (I was having fun in an altogether different way by now) brought back a ragtag garishly colored band of Munchkin Men wearing oversized helmets that instantly put me in mind of Rick Moranis’s turn as the Darth Vader parody in Mel Brooks’ Spaceballs, and had me wondering how the British Isles ever became Germanic in the first place.

“A color rendition of Rick Deckard in the film Blade Runner” presented me with a graphic-novelesque illustration of someone who wouldn’t look out of place on a panel of pundits on British football commentary show Match of the Day, while “Wartime London during the Blitz” presented me with a cityscape so repulsive you’d probably have thanked the Luftwaffe for bombing it had it really existed.

Stick to the surreal

Only when I followed OpenAI’s gentle manipulation and reverted to the surreal and impressionist did DALL-E2 begin to deliver something like convincing imagery. “A picture in the Cubism style favored by Picasso” returned some pretty decent AI-generated art, and “A jungle scene painted in the style of Henri Rousseau” turned up one or two decent attempts.

But then Picasso spurned realism, and the less well-known artist he later championed was roundly criticized by contemporaries for his lack of grasp of the aforementioned basics of painting. Now, I’m not saying I agree with them: I went to an exhibition of Rousseau at the Tate years ago and loved every minute of it. But clearly a machine that can only handle one or two movements in art, that is to say human image creation, and the downright surreal is clearly a machine that has some way to go before it supplants us.

Of course, this will change, and probably alarmingly quickly. But for now, perhaps we should continue to celebrate human creativity in all its diversity and deviance – because, as OpenAI’s abortive efforts seem to show, we are not so easily replicated. Or not yet, anyway.