The Unclimbed Peaks
GUEST POST from Art Inteligencia
The pace of artificial intelligence (AI) development over the last decade has been nothing short of breathtaking. From generating photo-realistic images to holding surprisingly coherent conversations, the progress has led many to believe that the holy grail of artificial intelligence — Artificial General Intelligence (AGI) — is just around the corner. AGI is defined as a hypothetical AI that possesses the ability to understand, learn, and apply its intelligence to solve any problem, much like a human. As a human-centered change and innovation thought leader, I am here to argue that while we’ve made incredible strides, the path to AGI is not a straight line. It is a rugged, mountainous journey filled with profound, unclimbed peaks that require us to solve not just technological puzzles, but also fundamental questions about consciousness, creativity, and common sense.
We are currently operating in the realm of Narrow AI, where systems are exceptionally good at a single task, like playing chess or driving a car. The leap from Narrow AI to AGI is not just an incremental improvement; it’s a quantum leap. It’s the difference between a tool that can hammer a nail perfectly and a person who can understand why a house is being built, design its blueprints, and manage the entire process while also making a sandwich and comforting a child. The true obstacles to AGI are not merely computational; they are conceptual and philosophical. They require us to innovate in a way that goes beyond brute-force data processing and into the realm of true understanding.
The Three Grand Obstacles to AGI
While there are many technical hurdles, I believe the path to AGI is blocked by three foundational challenges:
1. The Problem of Common Sense and Context: Narrow AI lacks common sense, a quality that is effortless for humans but incredibly difficult to code. For example, an AI can process billions of images of cars, but it doesn’t “know” that a car needs fuel or that a flat tire means it can’t drive. Common sense is a vast, interconnected web of implicit knowledge about how the world works, and it’s something we’ve yet to find a way to replicate.
2. The Challenge of Causal Reasoning: Current AI models are masterful at recognizing patterns and correlations in data. They can tell you that when event A happens, event B is likely to follow. However, they struggle with causal reasoning — understanding why A causes B. True intelligence involves understanding cause-and-effect relationships, a critical component for true problem-solving, planning, and adapting to novel situations.
3. The Final Frontier of Human-Like Creativity & Understanding: Can an AI truly create something new and original? Can it experience “aha!” moments of insight? Current models can generate incredibly creative outputs based on patterns they’ve seen, but do they understand the deeper meaning or emotional weight of what they create? Achieving AGI requires us to cross the final chasm: imbuing a machine with a form of human-like creativity, insight, and self-awareness.
“We are excellent at building digital brains, but we are still far from replicating the human mind. The real work isn’t in building bigger models; it’s in cracking the code of common sense and consciousness.”
Case Study 1: The Fight for Causal AI (Causaly vs. Traditional Models)
The Challenge:
In scientific research, especially in fields like drug discovery, identifying causal relationships is everything. Traditional AI models can analyze a massive database of scientific papers and tell a researcher that “Drug X is often mentioned alongside Disease Y.” However, they cannot definitively state whether Drug X *causes* a certain effect on Disease Y, or if the relationship is just a correlation. This lack of causal understanding leads to a time-consuming and expensive process of manual verification and experimentation.
The Human-Centered Innovation:
Companies like Causaly are at the forefront of tackling this problem. Instead of relying solely on a brute-force approach to pattern recognition, Causaly’s platform is designed to identify and extract causal relationships from biomedical literature. It uses a different kind of model to recognize phrases and structures that denote cause and effect, such as “is associated with,” “induces,” or “results in.” This allows researchers to get a more nuanced, and scientifically useful, view of the data.
The Result:
By focusing on the causal reasoning obstacle, Causaly has enabled researchers to accelerate the drug discovery process. It helps scientists filter through the noise of correlation to find genuine causal links, allowing them to formulate hypotheses and design experiments with a much higher probability of success. This is not about creating AGI, but about solving one of its core components, proving that a human-centered approach to a single, deep problem can unlock immense value. They are not just making research faster; they are making it smarter and more focused on finding the *why*.
Case Study 2: The Push for Common Sense (OpenAI’s Reinforcement Learning Efforts)
The Challenge:
As impressive as large language models (LLMs) are, they can still produce nonsensical or factually incorrect information, a phenomenon known as “hallucination.” This is a direct result of their lack of common sense. For instance, an LLM might confidently tell you that you can use a toaster to take a bath, because it has learned patterns of words in sentences, not the underlying physics and danger of the real world.
The Human-Centered Innovation:
OpenAI, a leader in AI research, has been actively tackling this through a method called Reinforcement Learning from Human Feedback (RLHF). This is a crucial, human-centered step. In RLHF, human trainers provide feedback to the AI model, essentially teaching it what is helpful, honest, and harmless. The model is rewarded for generating responses that align with human values and common sense, and penalized for those that do not. This process is an attempt to inject a form of implicit, human-like understanding into the model that it cannot learn from raw data alone.
The Result:
RLHF has been a game-changer for improving the safety, coherence, and usefulness of models like ChatGPT. While it’s not a complete solution to the common sense problem, it represents a significant step forward. It demonstrates that the path to a more “intelligent” AI isn’t just about scaling up data and compute; it’s about systematically incorporating a human-centric layer of guidance and values. It’s a pragmatic recognition that humans must be deeply involved in shaping the AI’s understanding of the world, serving as the common sense compass for the machine.
Conclusion: AGI as a Human-Led Journey
The quest for AGI is perhaps the greatest scientific and engineering challenge of our time. While we’ve climbed the foothills of narrow intelligence, the true peaks of common sense, causal reasoning, and human-like creativity remain unscaled. These are not problems that can be solved with bigger servers or more data alone. They require fundamental, human-centered innovation.
The companies and researchers who will lead the way are not just those with the most computing power, but those who are the most creative, empathetic, and philosophically minded. They will be the ones who understand that AGI is not just about building a smart machine; it’s about building a machine that understands the world the way we do, with all its nuances, complexities, and unspoken rules. The path to AGI is a collaborative, human-led journey, and by solving its core challenges, we will not only create more intelligent machines but also gain a deeper understanding of our own intelligence in the process.
Disclaimer: This article speculates on the potential future applications of cutting-edge scientific research. While based on current scientific understanding, the practical realization of these concepts may vary in timeline and feasibility and are subject to ongoing research and development.
Image credit: Dall-E
Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.
