I was inspired by a couple of articles on Artificial Intelligence I managed to coax out of my grandson, Arnav, for a local magazine I helped edit. Arnav in his current life breathes Python and exercises with AI as if the two letters were dumbbells. I wanted the articles written ‘For the wise Oldies’ who are my fellow residents. I was impressed with what he wrote, even felt educated, and was inspired to try out the most popular AI tool available on the Internet: OpenAI/ChatGPT.
I had been looking for an artist to illustrate some of my stories for quite some time. I thought, why not try OpenAI?
Clicking on the OpenAI site, after the bare minimum formalities, I typed:
Could you please make the image of a hunter being hugged by a large bear from behind. The man is middle-aged, fat and bald, wears North-Indian clothes, and has his hunting rifle in hand.
OpenAI barely took ten seconds.
The man was nearly as precise as I wished for, his expression showed shock, surprise, and abject fear. The lighting was of morning sunshine streaming through the trees, the back-lit ones in silhouette. That was something I hadn’t described, not even visualized, but the setting was perfect. You could see a fern and a few leaves on one side. The weapon in his hand was a 12-bore hunting rifle; you couldn’t ask for something more palpable.
I thought I could push things a little further by changing tracks
Can you please create an image of an Indian police constable trying to pierce the eye of a small-time criminal with a bicycle wheel’s spoke?.
Within four or five seconds, ChatGPT produced a colourful image of a Sikh policeman assaulting a poor man, blinded with a piece of cloth. Part of a bicycle stood in the background.
But I could ask for changes, if I needed any, said the AI.
I wrote : I want the Sikh replaced by a Hindu man without a turban, actually poking an eye of the victim.
Not exactly what I had in mind, yet the picture was flawless. The men had brown skin, and the policeman had a fierce mustache, but his cap was not of a constable, nor was his red belt. His shoulder star showed him to be a junior officer, not a low-ranking constable. The facial expressions – the rage on the face of the policeman, the pain in the face of the victim – were like the performance of a couple of good actors on a movie screen.
I was offered another chance to make more suggestions. I decided that if I gave the details more clearly, there could be the results that I had in mind. I wrote:
Police Constable:
Appearance: No cap, partially bald, with a small tuft of hair at the top, back of his scalp.
Uniform: Wearing a khaki uniform with a loosened belt around his waist.
Rank: Sergeant’s chevrons visible on his upper sleeves.
Weapon: Holding a bicycle wheel’s spoke aimed at the right eye of the victim.
Facial Expression: A cruel, intimidating look on his face.
Victim:
Appearance: A poor, small-time criminal.
Expression: Cringing with eyes closed in abject fear.
AI told me to try after 8:45 the next day. I had probably exhausted the memory allotted to me. I folded up for the day.
When I tried this exercise the next day. AI told me :
It seems I can’t create images directly from text descriptions, but I can guide you through the process of generating such an image. Here’s a detailed description you can use to communicate with an artist or an AI image generation tool which turned out to be an Add-on to ChatGPT,
That was a surprise. How did it create those images in the first instant?
DALL-E-2 or DALL-E-3
Now this is a paid application, at US$ 20 a month, not my cuppa, but my Non-resident daughter chipped in. The “ransom” (just kidding) was paid up.
I set off again, scouting out the DALL-E-2 (or 3, I’m not sure).
I pleaded : Please, I need an image of the following description:
Purpose : To expose the cruelty of a certain section of the Indian state police in their attempt to eliminate crime in their part of the Country.
Scene: a police constable is poised to pierce a bicycle wheel spoke into right eye of a poor, small-time criminal. The description of the characters are as follows:
Police constable
Facial expression is cruel, frighteningly intimidating.
Victim
Cringing in pain and fear.
Never mind that the Brahminical tuft of hair (his aerial to communicate directly with god) is absent because you cannot see the back of his scalp. The man has a star and a couple of stripes on his shoulder strap, which makes him a junior officer. The expression on his face is neither cruel nor menacing, The victim looks fatigued, but is neither frightened nor cringing in agony.
We had another try. Obviously, Open AI does not have many Indian Police officers and other ranks in its massive data store. The ‘Constable’ here is bipolar – he is both an officer (albeit Junior) and a Corporal as is evident from the chevron on his upper sleeves. I asked for a grayscale image with the corrections I suggested.
No, I am not frightened by the expression of the policeman (who now looks more senior than before with multiple stars on his shoulders; the chevrons are difficult to interpret. The victim lacks cornea in his eyes; his expression will not pass an audition for a movie.
Let’s have another go, please, Madam DALL.
Dear AI, you still haven’t got everything right. The policeman‘s look is fierce, he is poking the victim’s ear from behind, not his eyes. He has a tuft of hair on his head just above his forehead, not at the back of his scalp. The victim is certainly in distress, his eyes have lost cornea (probably from piercing), and they are bleeding – though not profusely enough.
Created only a few weeks ago, DALL-E-2 –has come a long way to DALL-E-3, like a good student who has done half his art course, but has quite some distance to go to get a degree. She will need to take all the descriptions into account at the first attempt.
The skin and features of the characters : are perfectly Indian after the significance was pointed out. Police uniform is guesswork; as it would be even of an American human artist assigned to do this work unless he had spent some time in an Indian police station and got to know officers, subordinate officers (like Sub-Inspectors), and constables. I insisted that stars on shoulder straps are only worn by officers; I was wrong – in India, every security guard sports a couple or more stars on his shoulders, no chevrons of any kind. I am sure my feedback should help OpenAI collect more data on Indian police.
Let us change track again, I thought. While OpenAI is collecting more data on Indian police and their uniform and rank badges, how about India’s major religion and its gods?
I requested a line drawing of Lord Shiva blessing a demon who had done penance in his (Shiva’s) name.
The image of Shiva was as if it was drawn by Raja Ravi Varma – except that the great Lord had a long, arrow-tipped tail!
There were other minor errors – this Shiva used his left hand to bless, which was uncharacteristic among Hindus. The Lord’s trident was perfect, only, instead of the double-headed drum (dumroo) that was supposed to be His favorite, we had a crudely tied bow. Open AI had assured me that there could be mistakes; I could ask for any change I desired. I could have asked for those changes, but I removed the tail and flipped the image horizontally.
Now that the image was passable, I wanted to change track once again. I requested for an image of Shiva and Vishnu in embrace.
This is what appeared in the next few seconds:
Perhaps OpenAI had my story among its tokens of data; it had, after all, reviewed my book
kindly; and even praised it. But, I thought, this is a far-too- friendly hug, not erotic enough to create a baby. So I asked for an image where the two gods hugged passionately. Here’s what emerged in the next five seconds:
Not the Best I could hope for, but this would do.
It took quite some effort, but Open AI cooperated with patience, and was willing to do final chiseling. Since it was past midnight, I gave in and thanked this marvelous tool for many human activities – among them, drawing and painting.
Can Open AI write a whole book ?
I typed: Can ChatGPT create a book of fiction if one were to give a detailed story?
Pat came the answer:
ChatGPT:
Yes, ChatGPT can help create a book of fiction if provided with a detailed story outline or plot. You can share key elements such as character descriptions, settings, major plot points, and themes, and I can assist in developing the narrative, dialogues, and detailed scenes. Feel free to provide the specifics of your story, and we can get started on crafting your fiction book!
But no, Sir. If ever I were to write another book, I would do it myself, in my own style. Just the same, you could be of much help in many ways – correcting my spelling and grammar, suggesting improvements, and adding fantastic artwork. Thank you.
