INTRODUCTION

The story I want to tell can be found throughout stories in history, stories about man-made artificial intelligence have been present in pop culture for centuries – from Frankenstein to The Wizard of Oz, eventually evolving into modern-day media with series such as Westworld (2016-2022) and the recent AI horror movie M3GAN (2023), stories about manmade intelligence, specifically computer generated artificial intelligence, have slowly infiltrated into pop culture. This phenomenon reflects an important trend in STEM areas where AI is advancing at exponential rates. In recent decades, however, exponential progress in AI has made it a reality. In 1950, Alan Turing published his paper, Computing Machinery and Intelligence. In 1956, the first artificial intelligence program was presented at the Dartmouth Summer Research Project on Artificial Intelligence (DSRPAI) hosted by John McCarthy and Marvin Minsky (1). Only a few decades later, intelligence forms such as OpenAI’s ChatGPT, a large language model trained to follow an instruction in a prompt and provide a detailed response, ‘Sydney’, Microsoft’s AI chatbot, Stable Diffusion, an AI image generator, and more, are readily available to the public. 

Since the beginning of AI representation in media, plots have frequently represented the appearance of other intelligent entities as ending in catastrophe, from human workers being replaced by automation, which has only grown in probability as we entered the 21st century, to the existential crisis of mass AI-committed genocide to purge mankind. Similarly, concerns and anti-AI sentiments have also risen parallel to the rate of its growth. The ethics of services such as ChatGPT have been the subject of rising concern. The recent rise of anti-AI sentiments have highlighted ethical issues associated with AI usage in the real world. 

Where does your mind immediately go when you encounter the concept of AI in the coming decades? Maybe you think of the danger of an AI apocalypse, or perhaps of the underlying philosophy of intelligence, but the current focal point of AI discussion is ethics of AI usage, as highlighted today by chatbots such as ChatGPT. 

ETHICS

The first aspect of ethics is transparency and sourcing. Chatbots often do not involve citations in their answers, meaning that not only are researchers not as recognized for their work, but also it is difficult for learners to fact-check and fully understand the material they are working with. 

Stanford Online High School junior Benjamin Klieger, creator of CheckforAi, a nonprofit tool to identity AI-generated work, and Groupters, a social media platform with a focus on ethical design, commented that “AI language models, while powerful, are known for their ability to hallucinate information, creating fake citations or quotations.” 

The second aspect of ethics is deceptive and over-reliant usage of AI chatbots and its academic dishonesty both in theory and in practice already in the status quo. Tomás Guardia, a professor in mathematics at Gonzaga University as well as OHS, and a leading researcher in the fibonacci sequence and number theory in relation to real life application such as rithmomachia and game theory, says that he is concerned that “students [may] use [ChatGPT]as a cheating tool instead of a supporting tool to improve their learning” and that AI prevents students from thinking critically by keeping them in their comfort zone. 

In order to combat this concerning trend, many AI detection services have been created, such as CheckforAi. In an interview with Klieger, he described in detail how different models have been cross-applied to ensure maximum efficiency. This technology is originally built on “OpenAI’s original roberta-base detector for GPT output, with a proprietary model built on top in order to help detect edited text” as well as input from existing models from the open source community and labeled datasets with millions of sentences. The newest model also responds to two of the main concerns regarding AI detection: 1) detecting paraphrasing and small edits to the text, 2) preventing false positives. 

“This new analysis uses a total of 3 models: The original model combination, Model Amelia, a perplexity based model for additional accuracy, Model Bonnie, and a third confirmatory model, Model Charlie,” he explained in response to the concerns. “Through the combination of these three models, text edited by and from AI is better identified as high risk, and false positives are limited.”

Furthermore, the most recent combination of models examines both the sentence perplexity (i.e. how well a model can predict text) and word combinations within a sentence by sentence and full context. This means the analysis is resistant to small changes, while also providing detailed insight on a sentence by sentence level. In terms of false positives, a low-cost model to reframe results is by presenting them as likelihoods instead of absolute truths. In addition to this mechanism, Klieger notes that the purpose of these tools should be assisting human judgment rather than replacing it entirely. 

Moreover, according to Klieger, the long term goal of AI detection software is to continue to guide educator discussions, and help provide transparent and accurate insight into the nature of AI generated text. 

Gallery • 3 Photos

Benjamin Klieger

How human-generated text appears once tested by three different models, where each model serves a different purpose in the process, designed by Klieger.

EDUCATION

Educators at Stanford Online High School have also spoken about AI in education, whether it be through words or actions. Dr. Meg Lamont, an instructor of English, has preliminarily integrated ChatGPT into some English assignments and class discussions, the Honors Physics class with Mr. Shahram Mostarshed features a final project which consists of comparisons between work done by ChatGPT and students, respectively. 

College professors such as Dr. Ethan Mollick at the Wharton School of the University of Pennsylvania and Dr. Philip Yuan at Tongji University of Architecture & Urban Planning have begun requiring the use of ChatGPT in their class. 

Professor Yuan serves as the Council Member of Architects Sector and Digital Fabrication Sector at the Architectural Society of China (ASC), Director of the Academic Committee at the Digital Fabrication Engineering Technology, and the Principle and Co-founder of Archi-Union Architects and Fab-Union Technology. He summarized that “for architects, generative AI is revolutionizing the traditional process of architectural design.” 

According to Dr. Yuan, it has been proven that art is often more than just a product of humans from the popularity of GAN models in 2014 to the development of ChatGPT and StableDiffusion today. In the curriculum, ChatGPT is often used to write copy for projects and algorithmic frameworks like StableDiffusion generates conceptual images for architecture. However, he notes that both ChatGPT and StableDiffusion, as well as derived algorithms like Dalle and MidJourney, are still far from actual architectural projects as AI still cannot understand the meaning of structure and materials, nor can it understand how materials will affect structures. 

Dr. Yuan implemented these technologies into the curriculum to provide students with opportunities for creative thinking instead of limiting them. 

“Our course is highly experimental and our team has developed an artificial intelligence algorithm platform called FUGenerator,” He explains. “In the early stages, we set a design theme for students and had them use AI tools to generate design solutions around that theme. We then guided students to further use graphics and semantics to understand architecture, creating their own design lexicons to guide AI generation. Throughout the process, students continuously modified or added key terms, experimenting with different materials and construction forms to generate novel and interesting images which were perhaps attempts that architects had never tried before.”

During the entire process, AI could generate dozens of concept images within 10 minutes. This was an incredible benefit because human designers were able to focus on decision-making and critical thinking. In the course, students learned how to quickly screen and evaluate solutions based on intuition through rapid iteration and decision-making.

Klieger responds that “it is incredibly exciting and must be paired with constructive engagement on how and when is the best time to leverage generative AI in the classroom.” 

CONCLUSION

What has been covered, astonishingly, is only the status quo. Looking towards the future, Dr. Yuan also shared his predictions and hopes for the development of AI, particularly its effects on architecture, and how the underlying trends can be applied to general society. First, Dr. Yuan believes that it cannot be denied that in the coming years, a singularity will occur where the development of a certain technology will bring revolutionary changes.

Furthermore, he anticipates two possible types of these transformative technologies in his field. The first is a comprehensive large model based on architecture, similar to ChatGPT. This large model would begin to approach artificial general intelligence that can understand and process various data related to architecture in a holistic manner, such as combining the era of architecture’s origin with its components, form, and layout. The main specialty would be the possibility of human-like thinking as well as multipurpose utility. The second type is a group model, which would likely manifest as an integration of multiple intelligences. It would decompose the problems of architecture itself into many small parts, which would be trained and used by architects in a distributed and interactive manner.

But not everyone is as optimistic about the future AI as Dr. Yuan. 

Professor Guardia argues that we do not yet fully understand the implications of the usage of a tool as advanced as ChatGPT which can be extremely dangerous. While he agrees that ChatGPT is changing lives at an extreme pace, he argues that this is precisely why we have a lack of control, which eventually could ricochet and harm us instead. To finish his argument, he states that AI usage should be heavily regulated so there is a higher likelihood of using this intelligence for our benefit.

“We have to be conscious about the limitations of AI, if we do, there is hope” He says grimly. “Otherwise, I don’t know what our future will be.”

Dr. Guardia is not alone in thinking this. As many share these beliefs, AI alignment has become an increasingly important field that aims to ensure artificial intelligence systems achieve desired outcomes, ensuring that AI systems work for humans, no matter how powerful the technology becomes. 

Although Professor Yuan is generally optimistic about AI development, he acknowledges and agrees with some of the concerns presented by Dr. Guardia. 

“To be honest, AI has already posed a threat to human employment in some industries,” he says.

While he notes that AI is not developed enough to cause a significant impact on the field of architecture, he argues that it does not mean that preparation for the arrival of intelligence that could trigger many systemic changes is not needed. According to him, the use of AI can optimize building design, improve building quality, reduce costs, and directly reduce human intervention. The use of these technologies may make certain positions obsolete, such as manual draftsmen and certain scheme designers. At the same time, this will also become an opportunity to create new jobs and improve efficiency and production capacity. Therefore, how society or the architecture industry establishes a sound mechanism for architects to collaborate with AI, how to train people and provide reemployment opportunities, and how to transition from extensive development to intensive architecture design processes will be the key focus of future discussions.

Stanford OHS freshman Eden Li, who is working on a long term independent project writing a science fiction novel about the impacts and reception of advanced AI, agrees with this hypothesis. He argues that AI models are more efficient and profitable. The current version of the GPT model GPT4 already supports a vision of hyper-efficient AI that is able to communicate effectively with humans. 

Li argues that while AI would likely replace a lot of the value humans put into society, there will eventually be a balance with two possible paths. The first possibility follows the trend in human history: revolutionary changes in technology did not lead to mass unemployment. With the introduction of more powerful production tools, society eventually reforms and maximizes its asset utilization, meaning that people would be still employed with the help of such powerful AI tools. The result would be that production and work efficiency increase, making society more prosperous, and ideally, not leaving a trail of unemployment. 

The second possibility is that AI eventually becomes powerful enough to replace human decision making as a whole. In this future, he argues that humans will pursue the “inherent value of being human”, just like how humans pursued the art of horse riding after the invention of the automobile. He believes that humans will continue to pursue intellectual expansion, although now as an artform and not a corporate duty. 

This is not the beginning of AI research/creation, and it certainly isn’t the end – but it is undeniable that we have arrived at a crucial juncture in the development of sophisticated technology and advanced intelligence. Regardless of whether you believe in a dystopian AI apocalypse, a utopian post-materialistic society where the goal of life is uniquely individualized, or anything in between the two extremes, what is certain is that recent advancements in AI have shaken education and daily life. Only time will tell what will be unveiled as we tread further down this road of discovery, novelty, and the inevitable reformation that will come with it; for now, let’s hope that our story doesn’t unfold like the movies.