At the dawn of the internet, nothing in economic or technological history could have prepared us for the changes it wrought. Similarly, the advent of Apple’s iPhone redefined the mobile phone industry, rendering previous market leaders obsolete.
Today, we stand at another such inflection point. In 2022, Generative AI, or GenAI, a narrow neural network method (that uses transformer models), revolutionized the field of artificial intelligence (AI). This method, in some ways a new type of heuristic or practical application of mathematics, could quickly unearth long-range dependencies in all kinds of data points in a way no statistical tool could before.
Transformers allowed machines to incorporate context in ever-increasing detail, not just in text, but in almost any form of data we can conceive.
Transformers have shown enormous scaling power: The first scaling achievement of this new model was its ability to understand human languages, ushering in an era of large language models (LLMs). Suddenly, the interface between us and our machines was no longer a specialized syntax, but our own languages. Getting the best out of machines is being democratized to an extent unimaginable a few years ago.
However, this revolution isn’t without its drawbacks. The software development industry is under immense pressure from GenAI’s productivity boom, with demand for developers dropping sharply. Some observers expect demand for AI developers to rise and compensate for job losses, based on a selective reading of historical patterns, but this is mostly wishful thinking so far.
To the extent one should use reason to prepare for a new world if machines are going to be operated increasingly in human languages, we face an urgent need to re-channel the talent developed for tools that are going obsolete. There are opportunities in the shifting balance towards hardware and innovative GenAI-spurred applications, but grabbing them will require design, purpose and investment.
Our machines can now ‘see’: At the second level of scaling, as AI models became more powerful, we discovered that the same transformer math could effectively conquer human vision. Several critical thresholds have been crossed in the advancement of machine vision, with radical and permanent implications for autonomous vehicles, robotics and other industries.
Machines can now learn through observation, accelerating the arrival of general-purpose or humanoid robots. Optimists expect the number of humanoids to exceed human populations in a couple of decades.
This may or may not come true, but two things are clear: most humanoids will not take the bipedal form factor popularized by movies and this is a new industry that could rise from near-zero to become multiple times the size of the global smartphone market in a decade or so.
Scientific exploration is AI’s third wave: At an even larger scale, transformer models unveil complex interrelationships in data across various fields, from genetics to quantum algorithms and material sciences.
As a result, many times more proteins or crystal structures have been unearthed by machines in a few short years than what we had discovered in all of pre-AI civilizational history. This capability is opening up new frontiers in scientific research.
For instance, in healthcare, GenAI models are already being used to decode genetic information and predict the efficacy of new drugs. In environmental sciences, they are improving micro-weather forecasts and even aiding in earthquake prediction.
GenAI is not just about chatbots: The transformer method is ushering the world into a broader machine era. This is not just about the arrival of chatbots and copilots, but about a fundamental shift in our relationship with technology. There will be missteps, mal-investments and market bubbles aplenty, but we need a genuine understanding—as against historical pattern fitting—at every level to benefit from these trends.
For nations, communities, families and corporates, GenAI is likely to consume decision-making more than almost anything else in the coming decades, with the gains disproportionately favouring those acting appropriately and early. Of course, this comes with all sorts of concerns, from the environmental footprint of building large global brains to the ethical implications of humanoid robots.
These worries are real, but in the world of competitive nations and corporates, the costs of indecision will be high. No path we take will be without risks and pitfalls, but some will cost society much more.
In other words, we are in an era of hyperchange, which may make the era of change between the early 90s and mid-2000s look like a mere trailer.
In columns to come, we will delve deeper into the specific implications of the transformer method for diverse aspects of our commercial and personal lives. The machine era is here, and it is for us to learn how it evolves.
#transform #world #ways #imagine
