A mental model for the current state of AI tech
In this post, I delve into my mental model of artificial intelligence, exploring the reasons behind its evolution that moves more towards chaos than a technological revolution. The crux of this issue lies in the unresolved challenges of safety engineering, which I will elaborate on here. Through this exploration, I aim not only to review my understanding of this mental model and its applicability but also to offer insights that might help shape your own views on the ever-evolving AI landscape.
Why not call it a “Technological Revolution”?
I find the term 'Technological Revolution' misleading for the state that we are in. My experience researching on a solution to the alignment problem led me to two striking problems:
The safety engineering problem remains unsolved.
We do not have a coherent path forward that can handle the technological chaos we are currently in.
The knowledge of the transformer model's architecture ( the image above) is not enough to fully understand GenAI’s inner workings, such as the specific functions of attention weights. For instance, we can't reverse engineer how an image is created in Midjourney or clearly explain how each attention weight contributed to ChatGPT's text responses. And despite this lack of understanding, the competitive drive of major tech companies to pioneer the first AGI - intensifies. It is evident that although AI is undoubtedly revolutionary technology, our current stage more closely resembles an attempt to comprehend and navigate through what seems like a chaotic, all-encompassing mega-dance that all of earth’s inhabitants are required to join.
Technology is a tool that “enables”.
Fire has been a pivotal element in our ancestral history. Unlike the majority of their existence, where our ancestors were at a disadvantage against formidable predators like lions, serpents, and crocodiles, occupying the lower ranks of the food chain, the discovery of fire marked a significant turning point. With fire, they could cook food, extract more nutrients, and fend off many predators, even in darkness. I believe the discovery of fire ranks as one of the most revolutionary technologies ever developed. It not only facilitated better access to food but also provided reliable protection against stronger animals, fundamentally altering our place in the natural world..
Physics has played a crucial role in our history. Our capacity to measure and understand the physical world has often determined the outcome of pivotal moments in history. From stone tools and wooden spears, which gave us an advantage in hunting and fostered cooperative skills necessary for tackling large game, to the longbows that decided many wars, and the use of gunpowder, each advancement represented a leap in our understanding of physics. In the realm of business, the advent of transistors revolutionized industries. Entrepreneurs who harnessed their potential reaped significant monetary rewards. Our ability to utilize revolutionary tools, driven by our understanding of physics, has consistently given us an edge, compelling us to maximize the benefits of each technological advancement physics offers.
Technology transcends mere power and capitalism; it is also a potent force for good. The Green Revolution powered by high-yielding varieties of crops of the 1960s was instrumental in making food more accessible and affordable, enabling nations to feed their populations effectively. The creation of electrical grids laid the foundation for the modern world we enjoy today. Since the end of World War II, a variety of technologies have fueled global progress. We have the ability to harness these technologies at our fingertips, crafting tools that not only serve practical purposes but also promote human flourishing. Our role goes beyond just wielding tools; we are the architects of a technology-driven future that enhances human life.
What does AI technology enable us then?
In my view, AI's potential as a truly beneficial technology is still unfolding. The promise of AI is evident with the development of transformative models like generative AI, but this is coupled with significant challenges like the safety engineering and coordination problems that still persist. Unfortunately, a clear answer to AI's full potential remains elusive, leading me to focus on the second word in the title of this post: 'Chaos'. This is to emphasize the current state of AI development, where the promise of innovation is intertwined with the complexities and uncertainties inherent in its advancement.
What is Chaos?
My interpretation of chaos might be the most challenging part of this post to grasp, but let's explore it. From the moment we are born, our understanding of the world is limited. As we grow, our minds, influenced by evolutionary underpinnings, gradually adapt to patterns that initially seem incomprehensible. These patterns, once obscure, eventually transform into ideas that we can comprehend. I define this process as 'chaos' - which mirrors the current experience with AI as a transformative technology. In both cases, we start with limited understanding and gradually adapt to make sense of complex, initially chaotic phenomena.
Order: When chaotic patterns become predictable.
As time progresses, we begin to recognize and understand patterns that were once chaotic, transforming them into 'Order'. This is the stage we have yet to reach with regards to AI as a technology and the amount of time we need to get there is rather unclear to me at this point. The solution to the safety engineering problem is crucial for this transition. As highlighted in a previous post, solving the safety engineering problem won't address all challenges associated with AI adoption.
I aimed to deconstruct my mental model here, and I believe I have succeeded with this post. Continuing to view AI's current state through the 'Technological Chaos' lens proves productive. This approach also drives my ambition to contribute to safety engineering, a crucial yet under-explored research area. To conclude, my quest to understand and articulate technological transformations, from chaos to order, persists. In exploring these complex realms, I hope to offer valuable insights that will lead to a safer, more structured future in AI technology.
Thank you for reading this post, if you find this useful - feel free to share it. Hope to see you in the next one!