How LLMs will help us reach Human Level AI (HLAI)

LLM has captured the public imagination. In this article we discuss how they are central to the future of AI.

INTRODUCTION

Large Language Models (LLMs) represent a significant advancement in the field of artificial intelligence (AI), particularly in natural language processing (NLP). These models are designed to understand and generate human language at a scale never seen before, thanks to their ability to learn from vast amounts of text data. LLMs have the capacity to process context, nuances, and complexities of language, making them incredibly versatile and powerful tools in AI development. They have been applied to various tasks, including chatbots, language translation, content generation, and more, reshaping how we interact with technology and opening up new possibilities in AI research and applications.

Critical Future’s Trailblazing Role in AI

Since our inception in 2014, Critical Future has been at the forefront of AI innovation, navigating the intricate landscape of artificial intelligence with foresight that predates the current AI zeitgeist. Founded by world-class AI expert Adam Riccoboni, author of “The AI Age,” and supported by a team of PhD-level experts from prestigious institutions like Cambridge, Harvard, and King’s College London, our journey has been one of relentless pursuit towards understanding and harnessing the potential of AI.

Large Language Models (LLMs)

Our work with Large Language Models (LLMs) stands as a testament to our commitment to pushing the boundaries of AI. From embedding specific domain knowledge into LLMs to make AI employees, to constructing Retrieval Augmented Generative (RAG) systems that delve into a company’s internal knowledge base, our endeavors encompass the full spectrum of AI’s capabilities. Our expertise extends to the creation of private LLMs tailored for enterprise-level applications and localized LLMs that operate independently on a single computer, demonstrating our versatility and innovation in the field.

Large Language Models are revolutionizing a myriad of sectors through their versatile applications. These range from content creation and management—where they automate the generation and summarization of various texts, offer real-time language translation, and personalize content—to enhancing customer support by powering chatbots and conducting sentiment analysis. They streamline business operations with email automation, document creation, and process optimization, and fuel innovation in products and services through dynamic product descriptions, personalized recommendations, and AI-powered assistance. In research and development, LLMs facilitate data analysis, scientific exploration, and competitive intelligence. They’re transforming training and education by creating e-learning materials and simulations, bolstering security through fraud detection and compliance monitoring, and even contributing to creative fields by assisting in writing and design. This wide array of applications underscores LLMs’ pivotal role in driving efficiency, innovation, and personalized experiences across industries.

But we think LLMs will have a much more profound impact on AI and society ahead could help us reach Human Level AI.

Future of AI – The Crucial Role of LLMs in Human-Level AI

The debate surrounding AI’s capability for understanding—a core theme of “The AI Age”—raises profound questions about the nature of intelligence itself. Human-Level AI (HLAI) was the goal of the AI founders and early AI researchers. Alan Turing began the quest for the great “prize” with his 1947 lecture to the London Mathematical Society—the first recorded reference to what is now known as human-level artificial intelligence.

As Herbert Simon a founding father of AI said in 1957, “…there are now in the world machines that think, that learn and that create. Moreover, their ability to do these things is going to increase rapidly until—in a visible future—the range of problems they can handle will be coextensive with the range to which the human mind has been applied.”

HLAI is the grand goal of AI research, said A.M. Turing Award winner Edward Feigenbaum in 2013:

“Computational Intelligence is the manifest destiny of computer science, the goal, the destination, the final frontier. More than any other field of science, our computer science concepts and methods are central to the quest to unravel and understand one of the grandest mysteries of our existence, the nature of intelligence. Generations of computer scientists to come must be inspired by the challenges and grand challenges of this great quest.”

What are the milestones to Human-Level AI? Yann LeCun of Facebook illustrates this with a metaphor: “We are currently climbing a hill. We are excited about the progress we have made, but as we approach the top, we see a series of other hills rising in front of us. That is how the HLAI development landscape looks. Just as we reach the top of each hill, we see the limitations in our current AI and what we need to do to climb the next hill. It’s not possible at this stage, with our current visibility, to see exactly how many breakthroughs we will need to reach human-level intelligence.”

Nonetheless, leading AI thinkers already agree on some key milestones we need to achieve to reach HLAI and the big one is understanding.

Understanding, LLMs, and Human Level AI

The holy grail: Understanding – Causality, reasoning, transfer of learning, imagination, and common sense are themselves hallmarks of one of the holy grails of AI: understanding. HLAI would need to have an understanding of the physical world. This is how children develop. During the pre-language stage, children use observation to develop a basic understanding of how the world works. Then through language children are able to articulate and conceptualise and understand the world. AI systems, by gaining the abilities of causality, reasoning, transfer of learning, and imagination, would gain an “understanding” of the world. It’s my view now that LLMs give AI this understanding, which is crucial to getting to Human-Level AI.

For example, when Adam Riccoboni wrote the book in 2019, “The AI Age,” we had algorithms that recognize a cat. The AI could recognize an image of a cat and predict correctly when it next sees a cat, but it didn’t know that a cat is a biological organism and often kept as a pet, or that it meows when hungry and, according to superstition, has nine lives. Deep learning didn’t understand a cat on any level beyond the pattern that makes up its image. Deep learning also needed to be trained on millions of data points before it could recognize the cat in the image. A human infant can see a cat a few times and then know it.

Yoshua Bengio explained this still represented a primitive type of intelligence –

“The computer has some understanding; it’s not a black-and-white argument… These networks have a level of understanding of images if they’ve been trained on them, but that level is still not as abstract and as general as ours.”

But now,  because of LLMs, a machine can understand a cat in as much depth as a human. The LLM will know a cat is an animal, it has 9 lives according to mythology, it will know its biology, history, everything a human knows. So LLMs are filling in the crucial gap of understanding for AI systems on their journey towards Human Level AI.

Critical Future stands committed to advancing the understanding and capabilities of AI through LLMs. This pursuit is not just about enhancing technology—it’s about deepening our collective understanding of intelligence, both artificial and human. As we push the boundaries of what AI can achieve, we remain inspired by the challenges and potential that lie ahead in this grand quest. 

We would love to hear your ideas and comments about the future of AI and the role of LLMs so please get in touch.