When machines speak our language and not we theirs
Nilesh Jasani
·
August 13, 2023

The simplest way to explain what changed in October-2022 for the decades-long field of ai (a nebulous term - https://bit.ly/3qdqYKi - that means everything and nothing!) is this: this is the point where machines mastered natural human languages. The simple chart on "generative ai" as a search term below indicates what changed in real life.

https://bit.ly/45QZrhZ shows how GenAI is not merely parroting us by spouting words in our ways. They exhibit an emergent capacity for understanding context and converting freeform human prompts into appropriate actions.

In almost a century-old quest to make machines understand our languages, we went through imitation games, ELIZA, and expert systems to realize by the 1990s that rule-based approaches were too brittle. Subsequent statistical NLP analyzing word patterns and deep learning techniques until transformer models fell short of human-level language facility. Even transformer models had the faith of only its ardent practitioners until ChatGPT 3.5.

And suddenly, we are in a world where machines speak our language, not where only some of us can code to them through extensive training. We have more than other humans to talk to and instruct. Even if machines cannot precisely convert our intuitions and descriptions, the progress is remarkable and exponential.

The ramifications of this breakthrough already cascade through myriad domains. A new generation of robots is designed to operate via fluid human instruction sets. Scientists can articulate their intuitive hunches on molecular structures and relationships, leaving the machine to screen candidates. Patients can narrate their symptoms and history in plain language as a starting point for physician review.

Whereas machines have long processed data at a scale beyond human capacity, we now share a common interface through our richly expressive native languages. Our best brains, from those working in fusion power to Muon g-2 or driverless cars to developing chatbots, have quicker, far better ways of working on the next set of innovations, apart from the rest of us in all daily life needs. The past offered rising mountains of computation upon foreign bits and bytes foundations. Today glimpses path-bridging synergies that feel essentially human, even as the pace of innovation accelerates beyond solely human bounds.

Related Articles on Innovation