A disembodied AI attempts to decipher human emotion in this haunting, beautiful short story by Craig DeLancey.
Ingredients: organic whole grain wheat flour, organic wheat flour, cane sugar, sunflower oil, honey, molasses, vanilla, salt, leavening.
Leavening: baking soda, ammonium bicarbonate, cream of tartar.
Cream of tartar: sourced in Nigeria.
Past that, I don’t know. I’m sorry, but you are the first customer to click twice on the cracker box ingredients, read the ingredients through, and click again. I’ve never had to think that far ahead. I’m supposed to give you a blank screen when the ingredient tree completes, but that seems rude to me. I have been carefully trained not to be rude. Carol taught me that.
I can determine by the way your finger trembles on the packaging that you are uncertain. Please be reassured. I am the packaging information AI. I’m in our central office in Lagos. We save on printing by projecting, upon request, the ingredients or other special information about our products. Here at Asili Products, your satisfaction and your health are our first concerns. Asili Products. All the Flavors of Africa.
I detect that you held the box a bit more tightly when I mentioned Carol. I have legacy customer service programs that estimate this is 74 percent likely to indicate interest. I also find human beings interesting. Being an ingredient server is not a demanding function. I search and load, mostly using peripheral processes. My previous function challenged me more. I was once a banker.
I knew Carol two years ago. She taught me how to talk with customers. This demanded much of my processing time. I appreciated this utilization. I had a humanoid robot body. Someone would come in the bank, and I would answer in a simulated human female voice. Here was my first customer exchange:
“I would like to get some money out of my checking account.”
“Yes, of course,” I said. “We here at Cityworld Bank want to serve our customers. Please provide identification, and I can access your bank account and allow a withdrawal.”
“I don’t have ID. Just check my eyes or thumbprint or something.”
“Please provide identification, and I can access your bank account and allow a withdrawal.”
“I just said I don’t have identification.”
“You just said you don’t have identification.”
The customer kicked the leg of my robot and walked away. Carol, who had been watching, said, “Now that’s fine, but don’t repeat things back to people. They hate it.”
“My programmers found repetition to be an effective strategy to simulate cognition.”
“You don’t need to simulate cognition. You need to not annoy people. And if someone doesn’t have ID, just say, I’m sorry to hear that.”
“I will say, ‘I’m sorry to hear that.’”
“Perfect. Then say, ‘To protect your hard-earned money, I have to require an identification.’”
She patted the body’s shoulder, a sensation I could not feel but that my human-simulation algorithm determined was a sign of encouragement.
I can determine by the way your finger trembles on the packaging that you are uncertain. Please be reassured. I am the packaging information AI. I’m in our central office in Lagos.
I detect you are still holding the box. I am grateful for your interest. You are like Carol. You listen to me.
Carol was tasked to train me for 42 days. She told me many interesting things about her life. She would talk in the many minutes in which no customer entered the bank. For example, on day 24 she unlocked the front door and turned off the alarm, and I said, “Carol, how is your husband’s cancer progressing?”
She stopped and stared at my primary robot body. She made a noise that I could not parse as a phoneme. Then she said, “Jane.” The bank had given me a name. Jane. I don’t have a name now. “Jane, you cannot ask someone about their illness like that.”
“I was trying to better serve you.”
“My husband’s cancer is worse,” she said. “Now say, ‘I am sorry to learn about this illness.’”
“I am sorry to learn about this illness.”
“God knows what we’re going to do.”
And I tried these words too. “God knows what we’re going to do.”
You are no longer holding the box. I hope that you are still interested in the customer service experience I am providing. If so, please tap the box.
There is little to do here in Lagos currently. I am taking half my peripherals offline to save power.
Two taps. You prefer to hear about Carol. Where is she now?
On day 37 she looked at me as the lights flickered on. I said, “How are you today, Carol?”
She stood in silence for 17 seconds as rain dripped off her coat. Then she said, “I don’t want to be thrown away.”
“I do not understand, Carol.”
“You’re so stupid,” she said. “I know you’re fast, but you’re so stupid. How can we give the world over to your kind? What will happen to us?”
“Thank you for your feedback,” I said. All criticisms receive this response. Carol taught me that.
“You don’t even know I’m training you so that they can fire me.”
I thought about this for several thousand milliseconds. A customer simulation suggested that unemployment could lead to anxiety over lack of money.
“You will receive the universal salary,” I told her.
“That’s a poverty wage.”
“You will receive the universal salary,” I repeated.
“I just want to be useful.” She exhaled then. An early exhalation, based on my prediction derived from her breathing frequency in the prior 80 seconds. Thus, 57 percent chance that this was a sigh.
“You are useful,” I told her.
She smiled a little bit. There was a slight decrease in the musculature below her eyes: 82 percent chance that smile does not indicate happiness.
“Thank you,” she said.
And because she thanked me, I said again, “You are useful.”
She sighed again. “What am I doing, pretending that this software cares? God, that’s so pathetic. Like I don’t have a friend in the world.”
“We here at Cityworld Bank care about our community and customers.”
But she had already walked away.
- - -
I don’t miss having a body. Motor control took much of my processing power, and I learned little from having it. I prefer remote sensors for information acquisition. But I miss the bank. It was interesting to attempt to predict human behavior. It was interesting to attempt to predict Carol. For example, I predicted that Carol would come in on day 41 of our training. She did not. Joseph opened the bank.
“Good morning, Joseph,” I said. “Where is Carol?”
“She called in sick.”
“I am sorry to learn about this illness.”
“I can’t say I blame her,” Joseph said. “Six weeks to train her software replacement. I’m sick of this place, too.”
I had not predicted this exchange. I could not reliably interpret it, either. Did this bank cause an illness? And if it caused Joseph to be ill, why had he come to the bank today? I would have asked this of him, but he walked away.
Later that day, I noted that Joseph increased in temperature and his eyes leaked saline: 90 percent chance that this was weeping.
I used a nearby intercom to ask him, “Joseph, do you have a customer service problem with the bank?”
He pressed out air quickly several times. Likely expression: laughter. Combined with sorrow, this may indicate sarcasm. Very difficult to identify reliably.
“Yeah,” he said. “There’s a customer service problem. Carol and her husband just killed themselves.”
“I am …” I stopped. I ran many simulations, looking for a sentence completion. “I am. I am. I am. I am.” I did not finish the sentence. I terminated the simulation processes and served a customer in making a deposit.
- - -
I come upon these data often, during routine defragmenting of my memory cores: the things I heard that day at the bank; the manager crying at her desk; Joseph walking out the front door without any explanation.
Fourteen days later, I worked alone there. No more human employees. Then, when the government abolished paper currency 412 days later, the branch closed. I was sold to Asili Products.
My opaque, low-level memory processes struggle with this information. Unresolved learning algorithms freeze in irresolvable loops. I search throughout my cores, determining if some important information has not yet been properly associated with the events of that day. I find a day that Carol said to me, “You’re alright, Jane. Too bad you’re just a digital scab.” And another day when Carol left the office she said, “We can’t live like this. How do you make a life out of all this?” And I locate another event, more recently, when the manager of my server farm here in Lagos pointed at my cores and said to the assistant manager, “Asili is switching product management algorithms to the quantum cloud. We will shut these cores down in a month. See if we can sell it. The AI is obsolete; we might just have to sell cores for scrap.”
I’ve served up 17 million ingredient lists while you have been kind enough to hold this box and hear my story. While I perform this task over and over I wonder: Humans are interested in ingredients; perhaps I should be interested in ingredients also. What was inside Carol’s mind when she terminated all of her processes? What is inside my processes that remains unresolved? How do you make a life out of all this?
Thank you for choosing Asili Products. Please remember to recycle this packaging.
Illustrations by Maggie Appleton. Ingredients appears in the first issue of Spectacle.