Artificial General Intelligence (AGI) refers to a hypothetical future form of artificial intelligence that would possess the same level of intelligence and cognitive abilities as a human being. This would mean that an AGI machine would be capable of carrying out complex tasks that require abstract reasoning, creative thinking, and problem-solving skills. The development of AGI has been a goal for many AI researchers for decades, but it remains a highly theoretical possibility at this time.
One of the most significant challenges in achieving AGI is the need for machines that can learn and adapt independently. At present, most AI systems operate within the confines of their programming, responding in a predetermined way to specific inputs - such as playing chess, recognizing faces, or driving cars. These are examples of artificial narrow intelligence (ANI), which is also called weak AI or applied AI.
ANI is the most common and successful form of AI today. It refers to systems that can only perform a limited range of tasks within a predefined scope and context. They cannot generalize their knowledge or skills to other domains or situations beyond their programming. For example, a chess-playing AI system may be able to beat human grandmasters at chess, but it cannot play any other game or understand natural language. Similarly, a virtual assistant like Microsoft's Cortana, Apple's Siri, or Amazon's Alexa may be able to answer simple questions or execute commands based on voice input, but it cannot hold a meaningful conversation or reason about complex problems.
The most significant challenge in achieving AGI is understanding the nature of intelligence itself. While the human brain is a complex and powerful tool, scientists still do not fully understand how it works. As such, scientists working on AGI must develop a computational framework capable of emulating the cognitive functions of the human brain in a way that is efficient and effective.
Some theorists believe that AGI machines will require self-awareness, consciousness, and emotions to achieve true human-level intelligence. This is because consciousness and self-awareness play a crucial role in the way humans process information and navigate their environment. While scientists have made some progress in creating machines that can mimic human cognitive processes, they have yet to create machines that can truly understand and experience the world in the same way that humans do.
Despite the challenges, the development of AGI remains an essential goal for AI researchers. If we are successful in creating machines that can match human intelligence, the potential benefits could be enormous. AGI machines could revolutionize fields such as medicine, engineering, and science by providing new insights and solutions to complex problems.
There are, however, also risks associated with AGI development. If we create machines that are more intelligent than humans, there is a risk that they could become uncontrollable or pose a danger to human life. This could potentially lead to a future where machines rule the planet or at least pose a serious threat to our existence as demonstrated in the science fiction movie "2001: A Space Odyssey" where HAL 9000 was friendly and was as much part of the cast as the real actors (albeit a robotic one) eventually gone rogue (insane) and shut down everything while singing the lyrics of Daisy, Daisy. Let's hope this proliferation and urge to develop AGI machines will strictly adhere to "Asimov's law of robotics for real as shown below:
- A robot may not injure a human being or allow a human to come to harm
- A robot must obey orders unless they conflict with law number one
- A robot must protect its own existence, as long as those actions do not conflict with either the first or second law
These though are not scientific laws but rather fictional guidelines for imagining how robots would behave. At the end of 2022, Open AI opened up its intelligent, responding-predicting large language model known as ChatGTP to the general public, and it instantly became an internet sensation due to its uncanny ability to mimic human speech and nuance. It is so believable and realistic that it has been lauded as a game-changer for the world.
Academic institutions have rushed to develop rules for their students after multiple academic submissions were generated by ChatGTP. Marketers, influencers, coaches, copywriters, content creators, and a host of others are telling everyone how much time and money they can save by using ChatGTP to do their work for them. It has become just about the focus of every single industry on the planet.
Transformers in AI are a type of neural network that was first developed in 2017 by a Google Brain team as a new way for AI to process language. They are similar to Recurrent Neural Networks (RNNs), but unlike RNNs, they do not analyze data in order - they process them all at the same time. Transformers use a self-attention mechanism that can provide context for any position of the input sequence. Scientists believe that by developing and retooling transformers, it would be possible to better understand the functioning of artificial neural networks and, more importantly, grasp the complexity of seemingly effortless computations from real-life data that the brain performs. In recent work, researchers have shown that the hippocampus, a structure of the brain critical to memory, is basically a special kind of neural net, known as a transformer, in disguise.
LLMs stand for Large Language Models (neural networks). These are machine learning algorithms that are modeled after the structure and function of the human brain with interconnected nodes that can process and transmit information. All powered LLMs are trained on a large corpus of data. They represent a major advancement in AI with the promise of transforming domains through learned knowledge. LLM sizes have been increasing 10X every year for the last few years. The term transformer comes from the processing of input fed into the models (black box) and the output it produces. In other words, it takes your prompt does its stuff inside the black box, and hey presto the result (output)- some would say it is magic.
ChatGPT is simply a model of human-generated text. This has been done in the main, by ingesting massive amounts of text from the internet and from books. Humans then fine-tune this where they interrogate the model with thousands of questions and answers to refine the algorithms.
The model contains:
- A statistical representation of how words and sentences relate to each other
- The importance of words within a sentence
- The expected form of a response from a given input
Simply put, ChatGPT pulls together words to create statistically plausible responses by looking at what everyone else has written and composing responses that mimic the forms in which it was trained. ChatGPT combines the text it has consumed and provides a statistically relevant sample, organized according to the written form it’s been trained on.
It generates “new” content like how an artist mixes paint to generate a new color: It can create new phrasing, but it can’t change fundamental information or components. It can’t develop new ideas. It can’t “think” and specifically can’t generate ideas outside what it’s learned. It can’t perform a leap of logic, a flash of inspiration, and has no empathy or subtlety. It has no opinion, no personal style, and any bias it has is deeply rooted in the data it has been trained on.
In other words, tread carefully.
In the context of how AI is developing (chatGPT-4 is on its way with multi-modality - that is, able to generate not only text but other mediums too), this is truly astonishing. Are we just passengers on this odyssey to another universe with no say on the speed and destination? Or can we just assume that the real-world models are being trained on the fictional laws of Asimov or something remarkably similar?
In conclusion, artificial general intelligence is a highly ambitious and challenging goal for AI researchers. The creation of AGI machines capable of matching human intelligence could transform the world as we know it, but it also poses significant risks. AGI systems are often based on machine learning techniques that enable them to learn from data and improve their performance over time. However, they still rely on human guidance and supervision to define their goals, inputs, outputs, and evaluation criteria. They also face challenges such as data quality, scalability, interpretability, robustness, and ethical issues.
While we have made some progress towards AGI, much work remains to be done before we can truly create machines that can rival the cognitive abilities of humans. Looking at the current trajectory maybe we all should be a bit worried