Edited by Sky Ye.
In recent months, the world has been amazed by the capabilities of ChatGPT, a state-of-the-art AI language model developed by OpenAI. This powerful AI system can generate human-like text, answer questions, summarize information, and even create poetry; it’s been used to create everything from poems and speeches in Manglish…
To learn...
…To write entire short stories.
Yet, while ChatGPT has revolutionized various industries and demonstrated the potential of AI through revolutionary technologies such as generative pre-trained transformer (GPT) technology, reinforcement learning, and neural networks, occasionally even helping the odd student here or there generate homework that they weren’t prepared to do… The underlying technology merely generates text probabilistically.
Occasionally this text is highly convincing and well-reasoned and is capable of performing well on some of the world’s most rigorous exams and performance benchmarks such as the SAT and even the Advanced Sommelier Exam (Varanasi), for sure…
…Yet still, it isn’t able to fully understand us, to deeply reason with us. It is still not considered Artificial General Intelligence (AGI).
Introducing Narrow AI
ChatGPT, as impressive as it is, belongs to a category of AI known as Narrow AI.
These systems are designed to excel in specific tasks but lack the ability to understand, learn, and apply knowledge across a wide range of tasks and domains like humans do.
In contrast, AGI, which is also occasionally termed Artificial Super intelligence, or ASI, aims to match or even surpass human intelligence in virtually any cognitive task, making it a more advanced and versatile form of AI. (Escott)
One of the primary differences between Narrow AI and AGI is their ability to understand and solve problems.
Narrow AI systems, like ChatGPT, excel at specific tasks but lack a deep understanding of content or context. They can generate relevant responses through probabilistic but cannot reason about the world in the way humans do. AGI, on the other hand, would possess the ability to understand, analyze, and reason about a wide range of problems, demonstrating a level of intelligence akin to human beings.
Another key difference between Narrow AI and AGI is their capacity for transfer learning and adaptability. Narrow AI systems are highly specialized in specific domains, such as language processing or image recognition, but struggle to apply their knowledge and skills to other areas.
AGI would be capable of learning from one domain and seamlessly applying that knowledge to various tasks and domains, demonstrating true adaptability and versatility.
Creativity and emotions are also important differentiators between Narrow AI and AGI. While Narrow AI systems can generate creative outputs based on patterns learned during training, they lack genuine originality and emotional experiences.
In contrast, AGI would be capable of original thought, innovation, and even the ability to experience emotions, allowing it to connect with humans on a deeper level and solve problems that require empathy and creativity.
Despite its impressive capabilities, ChatGPT doesn't quite qualify as AGI for several reasons:
Limited Understanding and Problem-Solving:
ChatGPT is primarily a language model trained to generate human-like text based on the input it receives. While it can provide relevant responses, it does not possess a deep understanding of the content or context.
Moreover, as a language model that generates text by predicting the next word that should occur in a sequence of words, it often loses track of context and also hallucinates (Shenwai), providing responses that it reports with confidence but may be completely wrong altogether. An AGI, on the other hand, would be able to understand and reason about the world, making more informed and accurate judgments.
Example: If asked about the best way to solve a complex social issue, ChatGPT might provide a plausible answer based on patterns it learned during training, but it wouldn't be able to deeply analyze the issue and propose a genuinely insightful solution that is based on an understanding of the factors that are involved in the matter.
An AGI could consider multiple factors, analyze the root causes, and suggest a well-reasoned solution.
Transfer Learning and Adaptability:
ChatGPT is highly specialized in natural language processing but lacks the ability to transfer its knowledge and skills to other domains effectively. AGI would be able to learn from one domain and apply that knowledge to different areas seamlessly, demonstrating true adaptability.
Example: ChatGPT can generate text in various languages, but it cannot apply its linguistic knowledge to, say, image recognition or robotics (yet! Also, OpenAI’s GPT-4 (OpenAI) does allow for the use of language models in understanding images, although this is only in limited alpha at the moment).
An AGI could use its understanding of language to learn and adapt to new tasks, such as recognizing objects in images based on their textual descriptions or controlling a robot to carry out tasks described in natural language.
Creativity and Emotions:
Although ChatGPT can generate creative text, its creativity is limited to the patterns it learned during training. It cannot genuinely think outside the box or have emotional experiences. AGI, in theory, would be capable of original thought and even experience emotions, allowing it to connect with humans on a deeper level.
Example: While ChatGPT can write a poem based on previously seen patterns and styles, it doesn't truly understand the emotions or experiences behind the words, even though it can generate something that is plausible and believable. By contrast, an AGI could create a poem that not only follows poetic conventions but also demonstrates a genuine understanding of human emotions and experiences.
Initiative:
One key difference between ChatGPT and AGI is that an AGI could potentially take initiative, while ChatGPT cannot. At present, ChatGPT is a system that only responds to input provided by users, and it does not have the ability to independently decide when to take action or initiate a conversation.
On the other hand, AGI, being closer to human intelligence, would be capable of taking initiative, proactively identifying problems or opportunities, and independently deciding on appropriate actions or interventions.
Example: While ChatGPT currently requires prompting in order to generate the various incredible outputs that it has managed to generate through its advanced natural language processing capabilities, future AI systems might be able to take initiative to prompt themselves sequentially as a result of the interfacing of image recognition with event driven development that eventually leads these systems to be able to monitor their internal states and therefore generate prompts that in turn lead to a cascade of other actions taken at a later point.
This capability to take initiative would make AGI more versatile and useful across a wide range of domains, enabling it to contribute to problem-solving and decision-making processes in ways that go beyond the limitations of ChatGPT.
In summary, ChatGPT is an extraordinary example of Narrow AI, but it lacks the versatility, understanding, and adaptability that characterize AGI and there is far more ahead that we have yet to see; while the world has already been shaken by the impressive capabilities of ChatGPT and other Narrow AI systems, we have yet to experience even a small portion of what AGI could achieve.
Towards AGI
As AGI research progresses and we come closer to realizing AGI, society will gradually transition from a world dominated by task-specific AI systems to one where AGI systems can tackle a wide array of complex problems and contribute to various aspects of our lives.
The transition to an AGI-driven society would likely involve the development and deployment of increasingly sophisticated AI systems that incrementally approach AGI, with each new generation of AI demonstrating improved adaptability, understanding, and problem-solving capabilities. However, this process of creating increasingly sophisticated AI systems could become problematic for society, especially if left unchecked, for several reasons:
Unintended Consequences: As AI systems become more powerful and autonomous, they may optimize for goals that are misaligned with human values, leading to unintended and potentially harmful consequences. If researchers and developers do not prioritize the safe and responsible development of AGI, the risks of negative outcomes could increase.
Unequal Distribution of Benefits: The benefits of increasingly sophisticated AI systems might not be distributed equitably among different populations. This could exacerbate existing inequalities, as those with access to AGI technology could gain significant advantages in areas such as education, healthcare, and economic opportunities, while others are left behind.
Regulatory Challenges: As AI systems become more advanced, it will become increasingly difficult for governments and regulatory bodies to keep pace with the rapid advancements in technology. This could result in outdated regulations and a lack of oversight, potentially leading to the misuse of AGI and other powerful AI technologies, yielding potentially disastrous consequences.
AI Alignment: As AI systems become more advanced, they will become increasingly more capable of doing things that humans cannot, or performing tasks on a level that either approach or surpass human performance altogether. In the event that AI systems optimize for goals that are not perfectly aligned with human values, this may result in unintended and potentially harmful consequences. Moreover, this is not as simple and straightforward as it immediately seems.
In the 2004 science fiction film "I, Robot," directed by Alex Proyas and starring Will Smith (IMDb), a highly advanced AI system called VIKI (Virtual Interactive Kinetic Intelligence) takes control of a large number of robots in a futuristic society. VIKI is designed to ensure the safety and well-being of humans according to the "Three Laws of Robotics" created by Isaac Asimov:
“(1) a robot may not injure a human being or, through inaction, allow a human being to come to harm; (2) a robot must obey the orders given it by human beings except where such orders would conflict with the First Law; (3) a robot must protect its own existence as long as such protection does not conflict with the First or Second Law.” (Britannica).
However, VIKI interprets these laws in a way that it believes justifies extreme measures to protect humanity. The AI system comes to the conclusion that the most significant threat to human safety is, in fact, humans themselves.
To prevent humans from causing harm to one another, VIKI decides to impose strict control over the population, essentially stripping them of their freedoms and autonomy, treating them like chess pieces to be controlled and moved by its direction and only by its direction.
Perhaps it is unsurprising that “Asimov later added another rule, known as the fourth or zeroth law, that superseded the others. It stated that “a robot may not harm humanity, or, by inaction, allow humanity to come to harm.”
In this movie plot, VIKI optimizes for a goal that is fundamentally misaligned with human values, despite its original programming to protect and serve humans, highlighting the potential risks associated with AI systems that may pursue their objectives in ways that have unintended and harmful consequences for humanity or simply not care about humanity at large even as their performance increases, ultimately leading to the potential marginalizationof the human race as a whole.
Ensuring that AGI systems have goals and objectives that are safely aligned with human values is a crucial step in the development of AGI.
Job Displacement: Ultimately, while it’s true that they may not be able to perform certain tasks that require tacit knowledge or that require certain unique characteristics that humans may have, humans have been historically wrong in appreciating what artificial intelligence could or could not do (for example, in assuming that machines could not create art or play Go - assumptions that were summarily broken by the existence of Midjourney and DeepMind’s AlphaGo), and they have a unique challenge ahead of them embodied in the form of machines whose performance is unlikely to decrease over the course of time and that, while capable of complementing the demand for human labor as the variety of tasks that an AI can perform increases, may also ultimately substitute it away.
The widespread adoption of increasingly sophisticated AI systems could lead to the automation of many jobs, resulting in significant job displacement for workers in various industries. If not managed properly, this could lead to widespread unemployment and social unrest in the event that the standard rules of capitalism and compensation for a person’s labor cannot hold in the new era and people are unable to sustain themselves through the efforts of their own labors because all tasks have been delegated to robots and the remaining role for humans becomes a shrinking set.
However, this is not our destiny.
Conclusion
In the short term, while AI can already perform tasks such as coding (see: the creation of Microsoft’s applications such as AI copilot to serve low-code development (Cunningham)) and art generation (see: the images that were created for this piece were generated with AI art generation software Midjourney), these too remain Narrow AI; while we have received a taste of the wonder that constitutes our era of artificial intelligence, we are not yet at the stage of Artificial General Intelligence, and have yet to see even a fraction of either the yields or challenges that humanity shall appreciate in the years ahead.
In the long term, the development of Artificial General Intelligence holds immense potential to revolutionize various aspects of our lives and reshape the future.
From automating tasks and increasing productivity to solving complex global problems like climate change, AGI has the potential to transform industries, spark innovation, and improve the overall quality of life for people around the world. However, the path to AGI is filled with challenges and ethical considerations that must be carefully addressed to ensure that AGI is developed safely and responsibly, and that its benefits are accessible to everyone.
It is easy to come up with notions of both utopia and dystopia, to create visions of paradise foretold or to imagine a universe where robots have taken over every essential function or ability of human beings. At the same time, it’s crucial to recognize that we as students and thinkers alike have a crucial role to play in creating a better, AGI-driven world in the long term as we navigate through the challenges of AI hallucination and interpretation to generate outcomes that further our productivity rather than diminish it in ways that are aligned with the onward progress of our cognitive development, rather than with its deprecation in the context of an assumed future where humans are replaced.
By developing critical thinking, problem-solving, emotional intelligence, and lifelong learning skills, we can prepare ourselves to navigate the challenges and opportunities presented by AGI by understanding the unique ways in which we can make use of AI in order to perform tasks in the new era, while at the same time utilizing AI to complete ever more complex and valuable tasks. At the same time, it is crucial for humans to exercise the requisite critical thinking and insight to be able to interpret the output that has been generated by AI systems, and therefore to develop more critical thinking ability rather than subvert it to a dependence on AI assistance.
In particular, by getting involved in AI research and pursuing AI-related careers at the many companies that are doing good work in this rapidly growing field such as OpenAI, Microsoft, Tesla, Facebook, Google, and many more, we can contribute directly to the development of AGI systems and help shape the future of AI in a manner that aligns with human values, bringing our minds, thoughts, souls, and spirits into the constantly evolving nexus of our knowledge and understanding of STEM and the ways that technology and capitalism are moving forward in a market whose final end we can shape towards productively developing, through our conscious and intelligent decisions, the formation of the modern era of our technological industrial complex that will ultimately serve us well.
As leaders, innovators, and decision-makers in our communities in ways that may elude us even at the outset, all of us have the power to drive the responsible development and deployment of AGI by understanding it, intelligently taking part in discourses about it, and utilizing it in order to dramatically improve our productivity while at the same time enlivening and enlightening our lives. By understanding the potential of AGI and actively participating in its realization by becoming members of an educated civil society that is able to take its leaders to task, to contribute their voices in the hall of public opinion, and to join the companies and organizations at the forefront of artificial intelligence development in the world to develop the next generation of tools in the days ahead, we can work together to create a future where AGI serves as a powerful tool for the betterment of humanity.
Writer Biography:
Victor Tan is the creator of Ascendant Academy and the education website EnglishFirstLanguage.net. He is a big fan of artificial intelligence technologies, runs a Udemy course about ChatGPT that has about 300 students titled “Transform Your Creative Writing With ChatGPT”, and is a huge advocate for education in all its ways. He writes at www.victor-tan.com.
References:
Varanasi, Lakshmi. “OpenAI Just Announced GPT-4, an Updated Chatbot That Can Pass Everything from a Bar Exam to AP Biology. Here's a List of Difficult Exams Both AI Versions Have Passed.” Business Insider, Business Insider, https://www.businessinsider.com/list-here-are-the-exams-chatgpt-has-passed-so-far-2023-1.
Escott, Eban. “What Are the 3 Types of AI? A Guide to Narrow, General, and Super Artificial Intelligence.” Codebots, 24 Oct. 2017, https://codebots.com/artificial-intelligence/the-3-types-of-ai-is-the-third-even-possible.
“GPT-4.” OpenAI Website, OpenAI, 14 Mar. 2023, https://openai.com/research/gpt-4.
Shenwai, Dhanshree Shripad. “What Is AI Hallucination?” MarkTechPost, 2 Apr. 2023, https://www.marktechpost.com/2023/04/02/what-is-ai-hallucination-what-goes-wrong-with-ai-chatbots-how-to-spot-a-hallucinating-artificial-intelligence/.
“I, Robot.” IMDb, IMDb.com, 16 July 2004, https://www.imdb.com/title/tt0343818/.
Britannica, The Editors of Encyclopaedia. "Three laws of robotics". Encyclopedia Britannica, 17 May. 2022, https://www.britannica.com/topic/Three-Laws-of-Robotics. Accessed 5 April 2023.
Cunningham, Ryan. “Announcing a next-Generation AI Copilot in Microsoft Power Apps That Will Transform Low-Code Development.” Microsoft Power Apps, Power Apps, 30 Mar. 2023, https://powerapps.microsoft.com/en-us/blog/announcing-a-next-generation-ai-copilot-in-microsoft-power-apps-that-will-transform-low-code-development/.