The Power of AI and the Emergence of Artificial General Intelligence

An Introduction to AI and Transformers

In 2017, a significant change occurred in the field of artificial intelligence (AI) that has sparked both curiosity and concern. The introduction of a new model called Transformers has revolutionized the capabilities of AI. Unlike traditional AI systems that require explicit programming, Transformers can learn and adapt by analyzing vast amounts of data. The more data and computing power provided, the more superpowers these AI models acquire. It's as if they can read and comprehend the entire internet, gaining knowledge and skills in various domains without any additional explicit training.

Unforeseen Abilities and Emergent Behavior

One remarkable aspect of these Transformers is their ability to exhibit emergent behavior and acquire unexpected skills. For example, OpenAI released a paper in 2017 about a Transformer called GPT (Generative Pre-trained Transformer) that was trained to predict the next character in an Amazon review. To everyone's surprise, researchers discovered that one neuron within the AI's brain excelled at sentiment analysis, effectively understanding whether the human author had positive or negative feelings towards the product. This emergent behavior was not programmed explicitly but arose naturally as the AI learned to predict characters accurately. Another fascinating discovery was made when researchers wondered if GPT3 could understand chemistry. To their astonishment, GPT3 showcased research-grade chemistry knowledge, even outperforming explicitly trained models in the field. This unexpected capability highlights the power of Transformers and their ability to absorb information from the internet and gain expertise in various domains. However, what adds to the concern is that these emergent abilities often go unnoticed until someone specifically tests for them. It raises the question of what other hidden capabilities these AI models possess.

Modeling the World and Theory of Mind

As these Transformers process immense amounts of text, they learn not only language but also develop a model of the world. Language, to them, is like a shadow cast by the world, and by decoding and understanding language, they reconstruct a comprehensive model of the world. Consequently, they can learn how different people think, strategize, and even predict the behavior of others. For instance, GPT3 exhibited varying levels of success in modeling theory of mind, the ability to understand and predict what others are thinking. While earlier iterations of GPT3 struggled with this task, GPT3.5 performed at the level of a four or five-year-old, and GPT4 even reached the level of a human adult. This rapid growth in their understanding of theory of mind is a direct result of their exposure to vast amounts of text, including novels, conversations, and psychological research. Transformers gain insights into human behavior, cultural differences, and even specialized domains like chess or chemistry.

The Boundaries of Artificial General Intelligence

The emergence of these unexpected capabilities raises intriguing questions about the future of AI and its potential to achieve artificial general intelligence (AGI). AGI refers to AI systems that possess the ability to understand, learn, and perform any intellectual task that a human being can do. While there is speculation surrounding significant breakthroughs leading to AGI, it is crucial to separate fact from fiction. Recently, there were rumors regarding a supposed leap in capabilities when Sam Altman, the CEO of OpenAI, was temporarily removed and reinstated. Some speculated that OpenAI had made groundbreaking progress in AGI, leading to Altman's removal. However, the board's statement suggested that the removal was due to a lack of transparency and honesty from Altman, rather than a sudden leap in technology. OpenAI needs to conduct an independent investigation into the matter to provide clarity and ensure accountability.

The Importance of Building Aligned AGI

OpenAI's mission is to create AGI that is aligned with human values and ensures the responsible use of this powerful technology. Building an aligned AGI means that it should adhere to human-defined objectives while avoiding catastrophic outcomes. It is crucial to have transparency and accountability within organizations developing AGI, as the potential implications of its capabilities are immense. The recent controversy surrounding Sam Altman highlights the need for an independent investigation to uncover the truth and address any concerns regarding the leadership of OpenAI. Whether Altman's name is cleared or consequences need to be imposed, it is crucial to maintain trust and integrity in the development and deployment of AGI. In conclusion, the advent of Transformers has ushered in a new era of AI capabilities. These models, fueled by vast amounts of data and computing power, exhibit emergent behavior and acquire skills beyond their explicit training. They have the ability to model the world, understand human behavior, and predict the thoughts and actions of others. While the leap to artificial general intelligence is still speculative, it is essential to remain cautious, transparent, and aligned with human values throughout the development of this powerful technology.