While we were lotos-eating at the temple of the Football World Cup 2022, a world-changing event largely escaped the Indian media’s notice: ChatGPT—Generative Pre-trained Transformer—the Artificial Intelligence (AI) chatbot capable of responding to both serious and outré suggestions with startlingly syntactically correct prose and poetry, and even commanding academic treatises.
Much of ChatGPT’s effusions sound meaningful (and some are, as a report in Nature said, “sublimely ridiculous”). But, risible or not, ChatGPT’s skillfulness has the world’s academics and IT mavens in a tizzy. This isn’t just some run-of-the-mill chatbot converser such as Siri, Alexa, or Google Assistant: a deep-learning algorithm, GPT-3 analyses an unending avalanche of text from books, Wikipedia, social media battles and conversations, and scientific publications; it writes poetry and prose better than most humans; it can crank out, in seconds, B- and C-level undergraduate papers; and it can write short computer programs. And it can do all this with just a month-long exposure to data.
Debmalya Biswas wrote on LinkedIn: “Experienced Chatbot creators will know how we are always afraid of the ‘unknown’—the generic/unexpected questions that the user might end up asking.” But ChatGPT allows the user to ask anything. Even barely coherent requests result in perfectly lucid responses—and ChatGPT has had only a month in which to mine and splice data. Its maturation is expected to be exponential—and, theoretically, endless, even as the data it is feeding on increases by leaps and bounds.
As the data infusion continues apace, this machine-learning—and autodidact—AI cannot but get better at its job. GPT-3, the third-generation GPT model, was trained on 175 billion parameters. There will, inevitably, come a time when what is today a general-purpose conversation engine will super graduate to become an eclectic expert—an AI for all seasons (and all reasons)—and churn out doctoral-level humanities and sciences theses.
ChatGPT can deftly handle translations (in a growing number of languages), syntactics, coding, debugging software, factuality, write biographical essays, creatively solve mathematical equations, pen soggy romantic poetry, write political critiques, and much more.
Among other things, ChatGPT has produced a Shakespearean poem describing itself, a haiku on Donald Trump, a verse in biblical style on how to remove a peanut butter sandwich from a VCR, the opening of a Vanity Fair essay in the style of an Ernest Hemingway novel on how writers and artists will soon be replaced by AI, and three poems on a theme I suggested: Write a poem with a saffron bikini, Hindu outrage, and a shameless song.
ChatGPT has no concept of style—yet. Ask it to imitate authors and it will play around with their favourite words, perhaps copy basic syntax, and pick out mises en scène. It's Shakespeare mimicking is that of a child.
In Scientific American, Almira Osmanovic Thunström wrote about giving ChatGPT a test: Write an academic thesis in 500 words about GPT-3 and add scientific references and citations inside the text.
“As it started to generate text,” she wrote, “I stood in awe. Here was novel content written in academic language, with references cited in the right places and in relation to the right context. It looked like any other introduction to a fairly good scientific publication.”
As the Generalist website said in its piece Endless Media, “It seems inevitable that existing models will improve and new skills will be added until the creation of a new Hemingway novel or Hitchcock film is no trickier than spawning an image with Midjourney or short story with ChatGPT.”
The writer James Poulos, aware of this creative dystopia, tweeted, “Catechize the bots or be catechized.”
The downside is that ChatGPT has zero qualms about cheating. It will manufacture not only plausible-sounding academic references but also create them out of thin air to resemble academic manuals of style and editing. In effect, ChatGPT is a godsend for students looking for exegetical shortcuts.
While OpenAI, the creators of ChatGPT, are reportedly working on a “watermark-like” attachable trace to the chatbot’s works, cryptographers have already predicted that it will manage workarounds. In short, there might never be a way to unearth ‘plagiarism’ via machine—because ChatGPT does not copy text: it creates text.
Stack Overflow, a knowledge-sharing platform for organisational knowledge, has already banned text from ChatGPT because it found too many users posting what it says is unreliable AI-generated content. But the ban is temporary: the future will not be denied.
The relationship app Tinder has long banned users who create bots to swipe and send messages to hundreds of people simultaneously, but many users are now using ChatGPT to present polished conversation starters far beyond their capabilities of articulacy.
The digital supertechnocrat Elon Musk had worriedly tweeted, “ChatGPT is scary good. We are not far from dangerously strong AI.” He later went on to tweet that he was pausing OpenAI’s access to Twitter’s database after learning that the firm was using it to “train” the chatbot.
This apprehension, from one of the world’s premier techies, will percolate to the layperson. I’m leery of auguring—because the future is fundamentally unpredictable—but I do think that ChatGPT (and AI, in general) will throw even the non-IT world into an existential tizzy between a couple and five years from now.
The current version of the GPT-3.5 series was completed in early 2022. According to IT scuttlebutt, GPT-4 is expected to launch between December 2022 and February 2023. Details are being kept close to the chest, but going by small leaks from OpenAI, the firm that created ChatGPT, it could be a significant evolution. Unlike humans, it is tirelessly learning to upgrade its knowledge base every single nanosecond. You might not like it, but humankind’s greatest ontological challenge is coming, inexorably. It all comes down to how we will value AI in the future: as a partner or as a tool.
Kajal Basu
Veteran journalist