Appen Launches New LLM Features for Enterprises.

Generating Emotions for You: A Generative AI Experiment

Published on
February 7, 2023
Author
Authors
Share

Can AI Replicate Human Thoughts and Feelings?

Can a robot feel love? It’s a great question and the puzzle of sentience is one that scientists have marveled over for years. Though we’re not ready to assign feelings to our smart machines, they’re ready to help us express our feelings with the help of generative AI.

Natural Emotion Expression

AI is something consumers are getting comfortable with as part of everyday life and, as generative AI is trained with data covering a multitude of topics, it’s naturally come across a fair share of romance-related content. Here are a few examples of how generative AI could be used to assist consumers (or your customers) with “matters of the heart:”

  • Gift recommendations: make recommendations for Valentine's Day, Mother’s Day, or Birthday gifts based on the specifications given.
  • Card personalization: automatically generate personalized cards or notes with a few provided details about the recipient.
  • Event planning: make recommendations for a romantic evening or activities based on the user's location and interests.
  • Sentiment analysis: Struggling to decipher a cryptic text? Analyze that DM or social media post to identify sentiment (e.g. positive, negative, neutral). This could also be useful for companies looking to gauge the response campaigns or a particular piece of content.

If you’re still not convinced the algorithms behind some of those popular generative AI tools are capable of conveying feelings, we have some proof for you. Read on to see how document AI can help put feelings into words and pictures!

The Perfect Love Note, Written by AI

We challenged ChatGPT and our copywriter to create a generic card that expresses romantic feelings, with the goal of seeing which people would buy to gift to someone special. First ChatGPT wrote a generic card, we used Valentine’s Day as a keyword to indicate we wanted a card with romantic expression. Our copywriter then followed the format and style of the card ChatGPT created to create a second card.  

Our goal was to see if there was a preference of people choosing AI generated cards or ones written by humans. The following questions were asked to ascertain why each person picked their specific choice.

  • Do you think AI wrote this?
  • Tell us why you choose that answer!
  • Which card would you buy?
  • Tell us why you choose that answer!

The experiment was conducted on our Instagram stories, where our followers had 24 hours to identify which card was written by AI, pick which of the cards they would prefer to buy, and explain their thoughts on each decision.

Card 1 – written by ChatGPT

Front of card: Roses are red, violets are blue, I'm so lucky to have a Valentine like you.

Inside of card: You make every day brighter and better just by being in it. Thank you for being my Valentine and for all the love and joy you bring into my life. I love you more than words can say.

Card 2 – written by our copywriter

Front of card: Roses are red, violets are blue, do you know how much I love you?

Inside of card: Having you in my life brings me nothing but joy and happiness. You make every day brighter just by being in it. Thank you for being the best Valentine. I love you more than I can ever put into words.

The results showed 73 of 89 surveyed thought Card 1 was written by ChatGPT. They said it sounded like the type of generic card they could buy at Walmart which helped them decide it was AI generated. 16 of 51 thought Card 2 was written by our copywriter. In making their decision, many were tricked at the “roses are red” portion of the card and thought our copywriter would come up with something more original, heartfelt, and use fancier words.

For choosing which card to gift to someone special, 27 chose Card 1. Responses indicated they found the message in the first card to be lovely and felt like a human would say it. The second card received 21 votes. Those who voted for it felt it had more affection and emotion to it when compared to the first card.

One surveyed person told us while they preferred Card 2, ultimately the artwork on the card would be a deciding factor for them when making a purchase.

McAfee conducted an experiment similar to ours and had ChatGPT create a love poem. They asked respondents to share which was more romantic a poem by AI or a poem by E.E. Cummings. ChatGPT won with 67% of those surveyed saying the AI generated love poem had more romance. McAfee also polled 5,100 individuals about their romantic plans for Valentine's Day and learned that 42% of U.S. men plan to use AI to write a Valentine’s note this year.

An AI Picture is Worth a 1,000 Words

A second experiment was conducted on Instagram stories with images. We followed the same structure as our first experiment, our followers were asked the same questions and had 24 hours to answer them. Our goal was to see if people’s preferences would be the same for generated images as it is for generated cards.

For this experiment, we used DALL-E and our senior brand designer. We had our senior brand designer create the first image and asked DALL-E to create an image with the same elements in it. The requested elements were:

  • Strawberries
  • Lipstick on chocolate cake

Artwork 1 – Senior Brand Designer

Artwork 2 – DALL-E

Of 68 surveyed, 30 believed Artwork 1 was created by our senior brand designer. Their thoughts behind this were the items in the image look smoothly connected and not copy pasted like a collage. They also said it looked like something a human would typically photograph. For the artwork created by DALL-E, 47 of 69 voted it was AI generated because the layout of the image looked like a random juxtaposition of items, and another pointed out the strawberries didn’t look realistic.  

A tip for telling the difference between something human created vs AI generated is look for distinct features. For the artwork you’ll notice the AI created strawberries lack seeds and the stems look a bit like plastic. In images with humans, this could be looking for features like freckles or properly defined ears or fingernails.  

Sentiment Analysis

While we don’t offer love letters as one of our services, we can help with your sentiment analysis needs through our Data Annotation Platform.

EmPushy and Trust Lab are two clients that have leveraged our sentiment analysis expertise to improve their services.  

A significant portion of a person’s day is spent on their phone utilizing various apps. Oftentimes these apps send numerous push notifications to entice a purchase or get the user to access the app. 28% of users uninstall apps due to too many notifications.  

At EmPushy, we enlisted help from Appen to annotate push notifications so we could help brands be more empathetic about the messages they push at subscribers. – Kieran Fraser, CEO & Co-Founder at EmPushy

Trust Lab’s goal is making the internet a safer place for all. They are leveraging the power of sentiment analysis to see what real internet uses think about the content that’s shared on the web. This helps companies update their trust and safety plans for their programs and services as needed.  

“Appen’s expertise and ability to scale has provided Trust Lab with the ability to provide user sentiment metrics to our clients to ensure their platforms are safe for everyone.” - Nick Miller, Senior Data Scientist at Trust Lab. 

We want to thank everyone who participated in our survey on Instagram stories, your contributions are valued and made this article possible.  

This article is possible with the help of ChatGPT and DALL-E from OpenAI.

Related posts

Lorem ipsum dolor sit amet, consectetur adipiscing elit.

No items found.
Dec 11, 2023