AI Open | The Frontline Newsletter

Dear reader,

In the late 1960s, kids around the world tuned in to watch Johnny Sokko and His Flying Robot. For the uninitiated, this was a Japanese tokusatsu (live-action, in my kids’ glossary) series about a boy, Daisaku Kusama (Johnny Sokko in the American version) who controlled a towering mechanical guardian with his wristwatch. Giant Robo was a metal behemoth. He was armed with missiles, lasers, and more.

The humble Doordarshan had aired a dubbed version of Giant Robo in the mid-1980s and later years. In one unforgettable episode, Giant Robo, who’s fiercely loyal to Daisaku, disobeys—after a lot of suspenseful hesitations, of course—a “destruct” command from the enemy (who’d just got hold of Johnny’s watch) and decides to act independently to protect his master from danger. This was a goosebump moment that 1980s kids cherish even today!

In hindsight, this was my first brush with a thinking machine—an idea that, at the time, belonged squarely to science fiction.

Fast forward. Today, artificial intelligence (AI) is no longer confined to television screens or science fiction. The rise of large language models (LLMs) like GPT-4, Claude, Gemini, DeepSeek-R1, and Llama has turned autonomous decision-making from fantasy into reality. But the evolution of digital intelligences has been long, winding, and full of both triumphs and controversies.

The idea of artificial beings is older than modern computing. The word “robot” itself dates back to 1921. The Czech playwright Karel Čapek introduced it in his play R.U.R. (Rossum’s Universal Robots). Quite interestingly (or ironically?), the word robot is derived from the Czech word robota, which means “forced labour”. Clearly, robots were then envisioned as mechanical servants.

In the 1950s, Alan Turing, the “father of theoretical computer science”, released his paper “Computing Machinery and Intelligence”, which laid the groundwork for AI, introducing the famous Turing Test.

Turing proposed that a machine could be deemed “intelligent” if it could mimic human conversation in a convincing fashion. Around the same time, Claude Shannon’s “A Mathematical Theory of Communication” (1948) set the foundation for natural language processing. This work is called the Magna Carta of the information age, as it formalised how information could be encoded and transmitted. These theoretical advances heralded to the world that AI is no longer a possibility; it’s happening, and it’s here to stay.

But the hype would soon give way to disillusionment.

In 1956 came the Dartmouth Conference, which officially launched AI as an academic discipline, and early attempts at natural language processing included simple rule-based systems like Joseph Weizenbaum’s ELIZA (1966). ELIZA mimicked a Rogerian therapist (you can look up the psychologist Carl Rogers if it doesn’t ring a bell) by responding with scripted phrases like, “Tell me more about that”. Some users were convinced they were speaking with a real person, but the illusion quickly shattered under sustained interaction. The field faced growing scepticism.

By the 1970s and 1980s, AI entered a period known as the “AI Winter”—a time when enthusiasm gave way to funding cuts and, more importantly, disappointment. Early AI models struggled with the unpredictability and the natural complexity of human language. Rule-based systems couldn’t scale. Machine learning was still in its infancy. It seemed like AI would remain a pipe dream.

The “winter” thawed in the 1990s, thanks to a shift from rule-based AI to statistical methods. Researchers at IBM pioneered statistical machine translation, using probability and vast amounts of data instead of manually coded rules. This era saw the rise of probabilistic models (which deal with uncertainty) like Hidden Markov Models (HMMs) and n-gram models. These helped power some of the early breakthroughs in speech recognition and machine translation.

A turning point came in 2013. The Czech computer scientist Tomáš Mikolov and his team at Google introduced Word2Vec, a method that allowed words to be represented as vectors in a multidimensional space. This technique captured semantic relationships—so that “king” minus “man” plus “woman” approximated “queen”. This was quite a revelation. Meaning could now be learned from data rather than pre-programmed.

The next major leap came in 2017. Vaswani et al (researchers led by Indian-origin Ashish Vaswani) introduced the Transformer architecture in the now-famous paper “Attention Is All You Need”. This revolutionary neural network architecture took the world of artificial intelligence by storm. Simply put, Transformers replaced sequential (one-by-one) processing with self-attention mechanisms, allowing models to weigh the importance of different words in a sentence dynamically.

This innovation enabled the rise of modern LLMs.

Google’s BERT (2018) showed the power of bidirectional contextual understanding (lesser mortals like us can call this, well, a kind of holistic context awareness), improving natural language understanding tasks. OpenAI’s GPT series, beginning with GPT-2 (2019) and now in GPT-4 (2023), pushed the limits of generative AI. GPT-3, boasting 175 billion parameters, stunned researchers and the public alike with its ability to generate coherent, human-like text from simple prompts.

You know the rest of the (hi)story.

Despite their impressive capabilities, LLMs are not without issues. Among the most pressing concerns are bias and fairness. Since LLMs learn from vast datasets “scraped” from the internet, they often absorb biases present in that data. This can lead to problematic outputs that reinforce stereotypes.

Another worry is the “Black Box” problem. Neural networks, mainly deep learning models, are notoriously opaque. Understanding why an LLM makes a particular decision continues to remain a challenge. Also, AI companies are pretty tight-lipped about what’s going on behind closed doors. There is absolutely no transparency here.

The third concern includes misinformation and hallucinations. LLMs sometimes generate false or misleading information with complete and impressive confidence. This has led to concerns about their reliability in critical applications (law, education, health analysis, recruitment, etc.).

Then there is the environmental impact. Training massive LLMs requires enormous (read in bold and capital) computational power. Sample this: In 2020, a study estimated that training a model like GPT-3 could produce as much carbon dioxide as five cars over their lifetimes.

That said, the future of AI lies not just in making models bigger but also in making them more efficient, interpretable, and aligned with human values. Researchers are actively looking at concepts such as multimodal AI (integrating text, images, audio, and video to create richer and more context-aware AI systems), explainable AI (developing methods to make AI’s decision-making more transparent), and ethical AI, meaning developing better guidelines and regulations to ensure AI benefits society rather than worsens existing inequalities.

Johnny Sokko’s robot is “cringe”, as my 9-year-old son ruthlessly put it when I showed him my favourite show recently. For their generation, LLM-powered AI has pushed sci-fi dreams into colourful realities. And now we are looking at Artificial General Intelligence, where machines think exactly like humans.

But as we stand on the precipice of AGI, we must ask ourselves: Are we ready? AI is no longer just a tool; it is a collaborator, a challenger, and, at times, an unpredictable entity. As we continue refining this technology, we must make sure that it understands and intelligently serves humanity’s best interests. After all, as Johnny Sokko’s Giant Robo showed us, true intelligence is not just about raw power—it’s about making the right choices.

Will AI be a true creative force in history, or will it just end up as an “AI”tem number? Only time—and a few million algorithms—will tell.

These “musings” struck me while I was editing Eshwar Sundaresan’s sharp and thought-provoking essay in Frontline last week—right after Chinese AI upstart DeepSeek sent shockwaves through the tech world with its impressive LLM. “Where is India’s DeepSeek?” Sundaresan asks, tossing out questions bound to make India’s IT community sit up and take notice. Dive into the essay, and don’t forget to write back about your own AI adventures!

Wishing you a super-intelligent time ahead,

Jinoy Jose P.

Digital Editor, Frontline

We hope you’ve been enjoying our newsletters featuring a selection of articles that we believe will be of interest to a cross-section of our readers. Tell us if you like what you read. And also, what you don’t like! Mail us at [email protected]