In a recent conversation with Salesforce CEO Marc Benioff at the Dreamforce conference, Nvidia CEO Jensen Huang made a bold declaration that has sent ripples through the tech world: AI is advancing at a rate that far outstrips Moore’s Law, the long-standing benchmark for progress in computing.
Huang stated, “We’re at a stage now, we’re in an era now where we’re moving way faster than Moore’s law. Arguably, easily Moore’s law squared.” He went on to claim that while Moore’s Law would predict a 100x increase in computing power over a decade, AI is actually advancing at a rate closer to 100,000x.
As I reflect on Huang’s claims, I can’t help but feel we’re collectively missing the point. The tech world is abuzz with talks of an impending AI singularity, but I believe we’re chasing a mirage, conflating computational power with genuine intelligence.
The Fallacy of Exponential AI Growth
While it’s true that AI capabilities are expanding rapidly, we’re conflating two distinct concepts: computational power and genuine intelligence. Yes, our machines can process data faster and run more complex algorithms, but are they truly getting smarter at the same rate?
Consider this: despite the massive increase in computational power, we’re still grappling with fundamental AI challenges like common sense reasoning, transfer learning, and robust decision-making in novel environments. The GPT models, impressive as they are, still hallucinate and struggle with basic logic. A study by Anthropic found that even advanced language models like GPT-4 make errors in basic arithmetic 20-40% of the time.
The Hidden Costs of AI Advancement
What’s often overlooked in these discussions is the astronomical energy cost of these advancements. A study from the University of Massachusetts Amherst found that training a single large AI model can emit as much carbon as five cars would over their lifetimes. As we push the boundaries of AI, we’re also pushing the limits of our energy infrastructure.
Are we prepared to bear this cost in our pursuit of AI supremacy? More importantly, is this sustainable?
The Real AI Revolution: Efficiency, Not Raw Power
I propose that the next frontier in AI isn’t about exponential growth in computational power, but rather about radical improvements in efficiency. The true breakthroughs will come from developing AI systems that can do more with less – less data, less energy, less computational resources.
This shift towards efficiency could democratize AI, making it accessible to smaller players and reducing the monopolistic hold of tech giants on AI advancement. It could also align our AI development with urgent needs for sustainability and energy conservation.
Implications for Investors and Innovators
For investors, this means looking beyond the hype of “bigger and faster” AI models. The real value may lie in companies developing energy-efficient AI solutions, or those finding novel applications for existing AI capabilities in untapped markets.
For innovators, the challenge is clear: how can we make AI smarter without simply throwing more computational power at the problem? This could involve breakthroughs in neuromorphic computing, quantum AI, or entirely new paradigms we haven’t yet conceived.
Redefining Progress in AI
As we stand at this crossroads in AI development, it’s crucial that we redefine what progress means. It’s not just about outpacing Moore’s Law or achieving artificial general intelligence. True progress in AI will be measured by how well we can integrate these technologies into our world in a sustainable, equitable, and truly beneficial way.
The AI revolution is indeed upon us, but its shape may be very different from what we imagine. As we navigate this new landscape, let’s ensure we’re asking the right questions and pursuing advancements that truly matter for humanity’s future.
Further reading
- https://venturebeat.com/ai/why-jensen-huang-and-marc-benioff-see-gigantic-opportunity-for-agentic-ai/
- https://the-decoder.com/nvidia-ceo-ai-progress-significantly-exceeds-moores-law
- https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/