Creativity is one of the most remarkable human traits. In 2016, a World Economic Forum report had concluded that creativity would be the 3rd most crucial skill in the workplace in 2020, catapulting in importance from the 10th position in 2015. The speed with which machines and Artificial Intelligence are permeating skills so far considered the forte of human beings, makes one wonder if it will steal the spark of human creativity? Also, as we advance, will AI only augment or replace human creativity?
Machines have come a long way from being catalysts to pushing human creativity. Advancements in Artificial Intelligence have enabled machines to become creative literary entities. Machines are now writing poetry, screenplays, and even stories. Superhuman Innovation: Transforming Business with Artificial Intelligence is written in a dialogue form between author Chris Duffey and artificial intelligence called Aimé.
Natural Language Processing gives machines the ability to read and understand human languages. Language models use an enormous & diverse pre-trained corpus of text via datasets in a process called Generative Pre-training. These are autoregressive language models that use deep learning to produce human-like text. Generative Pre-training model- GPT-3 that launched in May 2020 was one such model that operated with 175 billion parameters.
Besides drafting legal tenders, generating resumes & financial statements, GPT-3 could generate Tweets, create memes, and write podcasts. Tinkered Thinking is a podcast that uses GPT-3 for podcasts in both audio & text format. #MERZeye is yet another podcast written and performed by AI.
In October this year, in partnership with Microsoft, NVIDIA launched one of the most significant transformer language models, The Megatron-Turing Natural Language Generation (MT-NLG) model, with a whopping 530 billion parameters. DeepSpeed and Megatron power MT-NLG. Compared to existing large language models like GPT-3 (175 billion parameters), TURING NLG (17 billion parameters), Megatron-LM (8 billion parameters), MT-NLG has thrice the number of parameters.
Very recently, Google has introduced the GLaM- Generalist Language Model. This trillion weight model that uses sparsity has improved efficiency across 29 public NLP benchmarks in inference tasks and language completion categories.
Rapid advancements in language models are significantly improving AI-driven creative writing. In the future, AI will write more appealing & intriguing plots once the software gets better at synthesizing language. A significant advantage AI will always have over human authors is its ability to keep track of characters and stories at all levels.
AI has been writing prose with impressive structure and vocabulary but poetry requires more emotions based on experiences. Google Arts & Culture Lab, in collaboration with coder Ross Goodwin, has built a software that generates poetry. He trained a deep learning neural network on 25 million words of 19th-century poetry. Named PoemPortraits, the web app that takes a word, combines it with a selfie, and generates a few lines of AI-driven poetry. The software seeks patterns in the data and offers poetry to match the user’s inputs.
A Chinese publishing company, Cheers Publishing, published ‘Sunshine Misses Windows’, a collection of 139 selected poems written by AI named Xiao-Ice. In 2,760 hours, Xiao-Ice wrote more than 10,000 poems. Not an easy feat to accomplish by man.
Deep-Speare is an AI poet that generates verses that resemble Shakperean sonnets. It uses three components to generate content- rhythm model, rhyme model and language model.
Yet another example is that of Ai-Da. It is the world’s first ultra-realistic humanoid robot artist. In Nov 2021, she gave a public performance to celebrate the great Italian poet- Dante. In response to Dante’s “Divine Comedy,” Ai-Da, used algorithms from her data bank of words and took inspiration from Dante’s speech patterns to produce her work. Ai-Da’s unique AI language model works with “restricted editing.”
So, even when AI innovates using generative methods, it has managed to break the creativity barrier. It remains to be seen if AI will catch up and completely replace humans. The Future of Life Institute’s AI Impacts project predicts that Artificial Intelligence will be capable of writing a best seller by 2050.
There is no doubt in the quantity of output but what remains to be seen is whether AI can match the emotion and feel of human creativity.