The surge in the development of AI based on LLMs (Large Language Models, a type of artificial intelligence that can process, understand, and generate human language) is a massive trillion-dollar experiment.
Two and a half years after the launch of ChatGPT, AI tools are being used as creative writing professors and software coding companions. Cool new products like Synthesia have come on the scene, with embedded technology to create videos with realistic AI avatars who speak in many languages and sophisticated lip-syncing algorithms. Big Tech is generating revenues embedded in search engines and data analytics to discover correlations that are hidden from the human eye.
Language Mimicry: The Surface of Understanding
LLMs’ ability to generate human-like text is undeniably impressive. However, this eloquence of language has concealed a fundamental limitation: a lack of genuine understanding. These models excel in pattern recognition and statistical correlation but fall short of true comprehension – the “intelligence” and “common sense” that humans apply in decision-making.
The current capabilities of AI, while astonishing to creators and users, are mimicry of our language skills and should not be confused with understanding or intelligence–that will require entirely new breakthroughs and mindsets.
Gen-AI’s mimicry capabilities came as a big surprise to its creators, lighting up their eyes which caused them in turn to prematurely predict and propagate its potential rewards and dangers. While their excitement in the moment is understandable, I would say their lack of experience and ability to grasp the greater picture is an example of human immaturity, something success can create.
Today the very same creators are doubling down to increase the size of training data to be made available to LLMs, and to increase the number of Graphic Processing Units (GPUs, chips used to process information for LLMs to learn.) They believe that teaching them more through the same process will make the models intelligent.
Is that the way to create intelligent models? Or do we need new technical breakthroughs and new approaches?
I would argue the latter. We need newer startups to make new breakthroughs with new approaches. For starters, they don’t have old technology to protect their vested interest — their revenues from current products. “Cannibalizing your own product/s” is a difficult proposition, but it is a well-known need in Silicon Valley. Sometimes you have to in order to reinvent yourself. We need a fresh mindset to look at the problem and not be stuck in our old frame of mind. Rarely can new technology be taken from A to Z in the same lab.
Double Helix and Cures:
Gen-AI LLMs can be compared to the discovery of the double Helix model of DNA. While Watson and Crick’s discovery of the DNA double helix in 1953 was revolutionary, it took nearly 60 years for CRISPR to emerge as a groundbreaking gene-editing technology in 2012. Even now, 70 years later, its applications remain limited to approved treatments, mainly for inherited blood disorders like Sickle Cell disease.
60 years is a long road. Likewise, AI will need a long runway. Technological innovations have historically also taken much longer to reshape industries and have an economic impact on society at large. We need applications that can drastically improve human decision-making through understanding, not mimicry, and that will take time.
Discovery to Delivery:
The technology inventors are rarely the ones to identify groundbreaking applications of their inventions. AI is no different. Behavioral economists may be our answer. Economists tend to pragmatically look at the ripple effect. They, along with psychologists, may question how cognitive biases, emotions, and social influences shape human decision-making for economic gains, impacting humanity at large.
They, not the inventors, are the ones asking how human intelligence and machine intelligence can come together for the greater benefit of society. This marriage could prove to be true intelligence.
Much like the discovery of DNA’s double helix, the journey of AI’s promises and transformative potential will need patience, perseverance, and new breakthroughs. Frans Johansson’s book, The Medici Effect, shows us how “stepping into the Intersection created remarkable, surprising and groundbreaking ideas — intersection of field, ideas, people and cultures.” That is likely to happen in AI as well to discover the real machine intelligence. Let’s not forget, Watson was a biologist and Crick a physicist.
Until then, the trillion-dollar double helix experiment of AI will continue.
4 responses
Great article Vinita, enjoyed reading it. Yes, with so many technologies it takes a very long time to bring practical applications, and enough of them to really impact our daily lives.
I READ A LOT ABOUT AI BEING A SHAREHOLDER OF NVIDIA, MICROSOFT, ALPHABET ETC. HOWEVER I AM STILL WAITING TO SEE THE FINAL RESULTS OF THIS INVENTION. MY SON-IN-LAW RECENTLY TRANSLATED AN ARTICLE ON ME IN A MALAYALAM NEWSPAPER TO ENGLISH! I WAS SHOCKED AS TO HOW HE COULD DO IT! IT IS STILL TO MY MIND A WORK IN PROCESS!
SREEDHAR
Both LLM/ChatGPT and the discovery of the DNA structure (double helix) were major breakthroughs. But the similarity ends there. The rate of progress in medical sciences is nothing comparable to that in computing and information technologies. While computing power has been growing exponentially (doubling every year), the US life expectancy today is still less than what it was at its peak in 2014. Moreover, contrary to the common misconception, most common diseases are caused by unhealthy diet and lifestyle, not by faulty DNA. Perhaps a better analogy for the LLM/AI breakthrough would be the Manhattan Project which instantly created the powerful new technology for both nuclear energy and nuclear annihilation.
My son works in AI (Anthropic). According to him, the pace of growth in AI today is so fast that any country with a 6-12 month lead has an insurmountable advantage in critical areas such as military intelligence. The major concern, nay, obsession of the US intelligence community is to ensure that the US maintains an unbeatable lead over its adversaries, especially China.
Some AI experts now predict that in 3-5 years (not 60-70 years) the impact of AI could transform society in unimaginable ways. For example, AI doctors will be better than most human doctors. Moreover, they will be available 24×7 so human doctors simply can’t compete. Same for lawyers, etc. Consequently, AI’s impact on the job market, college enrollments, etc. will be truly unprecedented. It will be like a massive socioeconomic earthquake.
Interesting!