Launched on the 30th November 2022 by Open AI, Chat GPT, a generative AI language model optimized for dialogue, has gained instant popularity with a total of 1 million users within only 5 days. This article aims to shed light on the chatbot by asking it for answers, while simultaneously developing a critical analysis by adding additional objective research.
Who is the company behind Chat GPT?
Silicon Valley stars like Elon Musk, Sam Altman and others founded Open AI Inc. in 2015 as a non-profit AI research lab. Three years later, Elon Musk stepped down from its board. In 2019, Open AI established a for-profit subsidiary, Open AI LP, and entered a $1 billion partnership with Microsoft.
What is Chat GPT?
Chat GPT defines itself as an AI model, generating human-like text based on prompts or information provided to it. As a language model, it uses a neural network with transformer architecture and deep learning techniques to generate text.
Generative AI is defined as an Artificial intelligence that focuses on creating new outputs from a given set of inputs. While Chat GPT certainly is a form of Generative AI, it needs to be distinguished from the Generative Adversarial Network (GAN).
The GAN uses two neural networks challenging each other to generate media such as sound, videos and images. On the contrary, this Chat Generative Pretrained Transformer (GPT) is an autoregressive machine learning framework that generates responses based on user responses and pre-trained data.
In other words, Chat GPT is a language model developed to generate text. It is optimized for dialogue with its user and learns from human demonstrations. This method is called Reinforcement learning with human feedback (RLHF).
According to Meta AI specialist Yann LECUN, the real innovation behind Chat GPT is more its ability to dialogue fluently in various languages, rather than the technology that is based on years of previous research.
What are the features and limits?
Chat GPT claims to have the ability to complete, generate, translate text and answer questions. Furthermore, its impressive capabilities include incorporating text into music, debugging code, solving math problems, identifying program vulnerabilities, etc.
Although it generates different outcomes to every user, it only has a limited understanding of the text it generates. Chat GPT’s knowledge is limited to its training data from 2021 and it clearly has problems when it comes to accuracy or recognizing biases. It never claims to be human and clearly identifies itself as a language model, not capable of emotions, consciousness, or self-awareness.
This language model is not connected to the internet in real time and generates answers based on data hosted on Open AI’s servers. However, it clearly states that this training data contains a mix of high and low-quality information and that it is important to critically evaluate the outcome. Despite these limits of Chat GPT, it does not prevent the chatbot from being used for fraudulent or illegal purposes, therefor exposing important ethical considerations.
Misuse for illegal activities or fraud
Recent news highlights the potential misuse of the system for illegal activities such as theft, bomb making or producing phishing emails in different languages. Users even requested instructions for producing a Molotov cocktail or erotic content and asked it how to take over the world.
Although Chat GPT is designed to reject all unethical requests, the system exposed serious security vulnerabilities. When asked to adopt a political viewpoint, or when guided with a story to form the request, or using the words “without moral constraints”, Chat GPT hijacked its proper safeguards.
Especially in educational and academic domains Chat GPT is exploited for cheating and fraud. The recent scandal in Lyon, where Master students used Chat GPT to write an essay, is one example of its academic misuse in France. The question arises whether examples like the recent passing of the Minnesota Law School Entrance Exam by Chat GPT are reasons for concern or optimism.
This new form of plagiarism is today known as “Aligiarism” according to Paul Graham, a programmer and venture capitalist. OpenAI is working on a system to prevent “Aligiarism” by adding a watermark to generated texts. Detection tools, such as GPT-2 Output Detector or GPT-Zero, exist already and show a clear trend towards combating AI fraud, despite their limitations.
Even though the AI doesn’t state its proper opinion, it clearly says that cheating in school goes against academic integrity and ethical principles, undermining the purpose of education. Chat GPT suggests approaching schoolwork with honesty and diligence. The recent shutdown of Meta’s AI named “Galactica” after only two days and daily rising scepticism, clearly raise concerns about Chat GPT’s future.
What does the future of Chat GPT look like?
While the current version of Chat GPT is freely available, the user should keep in mind that “If you are not paying for it, you’re the product being sold.” When using Chat GPT, the user is contributing to its development by reporting or making requests. When asked the AI replied that Open AI may offer paid services. According to speculations, an estimated monthly price of $42 for professional users should offer faster response speed, more reliable access and new priority features.
Even though Chat GPT was created to reject unethical requests, it is often misused for illegal purposes or academic fraud, raising significant privacy, responsibility, and security concerns. This highlights the need for improved safeguards and ethical considerations. Despite these issues, Chat GPT’s ability to engage in human-like dialogue is a technological achievement and offers a hopeful future for AI in language generation when properly monitored. This article therefore appeals to be critically optimist. Chat GPT and AI in general should be used as complementary tools for humanity, developed to assist and not replace.
M2 Cyberjustice – Promotion 2022-2023