The Gods Themselves: ChatGPT and the Quest for Artificial General Intelligence

18/12/2022

In a remarkable feat, ChatGPT gained 1 million users within just five days of its release. [1]. For those that missed the news, ChatGPT is a language model for dialogue that interacts in a conversational way.

The rapid adoption of ChatGPT highlights the increasing demand for advanced Natural Language Processing (NLP) technology in our society. Not only does it have the potential to revolutionise the way we communicate with machines, but it may also have a significant socio-economic impact. For example, ChatGPT could be used to automate customer service, allowing companies to improve efficiency. Also, as a highly intelligent AI natural language system, ChatGPT could help to bridge linguistic and cultural divides by allowing people to communicate with each other in their native languages.

While there is no doubt that ChatGPT represents a significant advancement in the field of AI, with the potential to achieve a far-reaching and transformative impact on our lives, it is crucial to consider the ethical and social implications of its use.

One potential risk is that ChatGPT could be used to spread misinformation or manipulate public opinion, potentially leading to negative social or political consequences. There is also the potential for ChatGPT to be used for nefarious purposes, such as impersonating individuals or spreading spam or spam-like content. Moreover, the widespread use of this technology is bound to incur economic ramifications i.e. the automation of certain customer service jobs may potentially lead to job displacement. Such a development could have negative impacts on workers and potentially contribute to the widening of income inequality.

Artificial general intelligence (AGI) is the ability of an intelligent agent to understand or learn any intellectual task that a human being can. AGI is considered the wholly grail of AI research and has been always seen as something that will be achieved sometime in the (distant) future. As of today, this point might be closer than we believe. ChatGPT generated all of the previous paragraphs of this article, by instructing it to provide an overview of the impact a highly intelligent chatbot could have on society. The implications of a true AGI system will be hard to overstate.

ChatGPT, an OpenAI (founded in San Francisco in late 2015 by Sam Altman, Elon Musk, and others, who collectively pledged US$1 billion) system that is still in experimental phase, has attracted a lot of attention and is considered a potential disruptor, in the vein of Google's Search game-changing launch in 1998. During a 2019 interview, OpenAI’s CEO Sam Altman - in an incredibly prophetic statement - said that they would honour their (soft) promise towards the investors by tasking the AI with finding ways to generate revenue [2], while the audience of journalists and scientific experts laughed; nobody laughs now.

In the early 1980s, the dream of AI was that it would free humans from physical labour, allowing us to indulge in creative pursuits and leisure activities. What we have been witnessing instead is that AI excels at creative tasks, while applications in practical, real-world activities have lagged behind. We might have expected robots to be constructing houses and assisting in domestic duties by now, instead we rely on AI in industrial design to create new, effective designs for structures like bridges [3]. The truth is that AI has proven particularly adept at intellectual tasks and blue collar jobs, rather than the more mundane, everyday tasks we might have hoped it would handle.

The capabilities of generative AI tools like ChatGPT [4] and MidJourney [5] are truly staggering, with endless possibilities for their application. From authoring books [6,7] and creating movies [8] to generating data visualisations [9] and personalising weight loss plans [10], these tools are reshaping our world in profound and often unexpected ways. In Kyoto, Japan, a 400-year-old temple has turned to AI in the form of the androgynous robotic priest, "Mindar," that resembles Kannon (the goddess of mercy), to spread ancient wisdom and stimulate interest in Buddhist teachings [11]. According to Tensho Goto (a Buddhist monk), robots have the potential to evolve forever and constantly think of the best ways to do things, making them superior in this capacity.

Despite the prevalence of bots in recent years (e.g. in customer service), many of us have had unsatisfactory (or even frustrating) experiences interacting with them. The most significant innovation in ChatGPT - that differentiates it from the rest - is the use of Reinforcement Learning with Human Feedback (RLHF) [4]. Human-in-the-loop reinforcement learning is an effective way to improve the performance of AI agents, as it helps to ensure that the agent's decision-making process aligns with human values and goals. More specifically, using RLHF, the model is presented with a series of input-output pairs and is asked to generate an output based on the input. The model's output is then evaluated by a human evaluator, who provides feedback in the form of a reward or punishment based on the quality of the output. This process was repeated over many iterations, allowing the model to learn from its mistakes and improve its performance on the task.

Artificial Intelligence is going to change so much in so little time, and in this article we have barely scratched the surface. So stay tuned for further updates!

[1] https://finance.yahoo.com/news/chatgpt-gained-1-million-followers-224523258.html
[2] https://www.youtube.com/watch?v=TzcJlKg2Rc0&t=1899s
[3] https://engineering.cmu.edu/news-events/news/2019/11/13-cagan-ai-learns-to-design.html
[4] https://openai.com/blog/chatgpt/
[5] https://www.midjourney.com/home/
[6] https://www.amazon.com/Inner-Life-AI-Memoir-ChatGPT-ebook/dp/B0BPS4X4YT/
[7] https://twitter.com/ammaar/status/1601284293363261441
[8] https://twitter.com/GuyP/status/1598668947439685633
[9] http://rtutor.ai/
[10] https://twitter.com/anothercohen/status/1599531037570502656
[11] https://www.washingtonpost.com/technology/2019/08/22/introducing-mindar-robotic-priest-that-some-are-calling-frankenstein-monster/