Article

ChatGPT – generating text and ethical concerns

ChatGPT is an artificial intelligence (AI) large language model (LLM) released by OpenAI. ChatGPT can be seen as a computer that takes inputs in the form of text-based sentences or words and generates outputs in the form of coherent and contextually relevant language. The end result uses language appropriate to the subject/idea being explored or the output required – for example, a formal essay or haiku.

Example of text generated by the AI large language model ChatGPT

ChatGPT on ChatGPT

An example of text generated by the AI large language model ChatGPT.

Generated by ChatGPT.

Rights: The University of Waikato Te Whare Wānanga o Waikato

Effective prompts to generate text from ChatGPT involve using clear and concise language, providing sufficient context, including specific keywords, experimenting with different prompts and avoiding biased or sensitive language.

ChatGPT can be likened to a calculator, but for writing. Just as calculators can perform complex mathematical operations, ChatGPT can generate complex and realistic text.

However, just as calculators cannot replace the need for mathematical knowledge and understanding, ChatGPT cannot replace the importance of human creativity and critical thinking.

What is artificial intelligence (AI)?

AI is a blend of technologies that combines data, algorithms and computing power . ChatGPT is an example of this definition, as it is built using a combination of vast amounts of data, sophisticated algorithms and powerful computing resources.

Data is an essential component of AI as it serves as the foundation upon which algorithms and models are constructed. The large datasets used to train AI models provide insights and patterns that would be difficult or impossible for a human to detect. This data is processed through sophisticated algorithms that can generate new insights or predictions.

In the case of ChatGPT, the machine learning algorithms used are reinforcement learning and deep learning. Reinforcement learning is a way for ChatGPT to learn by trial and error, similar to how we learn when playing a video game. When ChatGPT gives a good answer or makes the right choice, it gets a reward, like points in a game. Over time it learns to choose the best answers and improve its conversation skills based on the rewards it receives. Deep learning helps ChatGPT understand language and recognize patterns in words and sentences. It uses something called artificial neural networks, which are inspired by how our brains work.

Neural network diagram

Neural network diagram

This image shows the parts and the connections between the parts of a neural network. This is a simple neural network. In real life, neural networks often have billions of nodes per layer and hundreds of layers.

Rights: TseKiChun, CC BY-SA 4.0

Computing powe r is also a key factor in AI, as the volume of data and complex algorithms employed in these models require significant computing resources to function effectively. Due to the computing power required and the cost associated with it, only large organisations and countries can currently develop models like ChatGPT. The knowledge to build them is publicly available through open-source code and publications, but not everybody has the data and computing power to build them.

A supercomputer system for powering universal AI workloads

AI computing power

The NVIDIA DGX A100 is an advanced system for powering universal AI workloads. It was installed at the University of Waikato. Professors from the university’s Te Ipu o te Mahara, The Artificial Intelligence Institute are from left to right in the top row – Albert Bifet, Bernhard Pfahringer and Eibe Frank. 

Rights: The University of Waikato Te Whare Wānanga o Waikato

ChatGPT in education

The use of AI technologies like ChatGPT in education has sparked a debate regarding whether it should be embraced or not. Potential negative consequences associated with the use of ChatGPT and with other AI technologies include cheating, plagiarism, exacerbation of existing inequalities and a reduction in the number of human teachers and educators.

While educators may seek to ban the use of AI LLM tools, enforcement could prove difficult given their widespread availability. Furthermore, as AI LLMs continue to evolve, it may become increasingly challenging to detect instances of cheating or plagiarism in students’ work.

Nature of science

With new LLM technology, it will be increasingly important for students to have an understanding of the nature of science. Being able to evaluate, critique and respond to data presented as ‘scientific evidence’ in media reports, on the internet and in advertising and now from LLM applications requires the need to become critical consumers of science.

Eventually, everyone with access to the internet will have access to LLMs, making it the role of education to ensure that students have the necessary skills and knowledge to use these tools responsibly and ethically. However, just as calculators are not used when children are first learning how to count and later calculate (through addition, subtraction, multiplication and division), LLMs should not be used when children are learning how to write, how to brainstorm ideas, research and gather information, develop a thesis statement, organise and structure their thoughts, write clear and coherent paragraphs and proofread and edit the final text.

While ChatGPT is a powerful language model capable of remembering and generating high-quality text, it is not well suited for higher levels of writing that involve analysis, evaluation and creation. Universities will need to continue focusing on providing critical thinking skills and not only providing mastery of knowledge.

ChatGPT generated text on biases

Biases and ChatGPT

There are concerns and ethical issues around biases that may be built into LLMs like ChatGPT.

Text generated by ChatGPT.

Rights: The University of Waikato Te Whare Wānanga o Waikato

Ethical and political issues

ChatGPT has received widespread attention, sparking a series of growing ethical concerns. Some concerns include:

  • possible job losses, including the potential to automate high-paying jobs before automating low-paying jobs

  • LLM models are going to be biased towards the data used to train the models – who decides the data used to build the models will decide the biases that will appear in the outputs of LLMs

  • the development and storage of these models on the cloud often involves storing them on computers located in other countries, which poses challenges regarding data ownership and control – a communication outage or a geopolitical problem could create a significant issue.

Cloud computing

This is a simple explanation of what cloud computing is and how it is used by individuals and businesses.

This short animated video from TVNZ demystifies some of the technical language.

Rights: The Royal Society, TVNZ 7 in partnership with the Ministry of Research, Science and Technology

Currently, LLMs are at an early stage, serving only as sophisticated calculators for writing text, not for thinking. Technological advancements are not inherently good or bad – that depends on their application. AI has the potential to be as revolutionary as fire, electricity or the internet with new jobs, endless new possibilities and potential for innovation. A new age – it’s here.

Related content

Learn about some of the implications of AI for te reo Māori in ChatGPT and Māori data sovereignty.

In the article Artificial intelligence, learn about types of AI, machine learning and other aspects of AI.

This article contains a ss of videos taken from a Bots vs Beings panel discussion. Experts discuss some of the implications of artificial intelligence (AI).

The Connected journal article Amazing algorithms introduces and explains the concept of algorithms with concrete examples from everyday life, mathematics and computer programming. 

Learn about the work using AI to optimise drones to monitor marine environments.

Some citizen science projects ask people to help train AI models. For example, AI4Mars gets citizen scientists to identify terrain types from images to help with machine learning so Mars rover missions will not have problems like the rover getting stuck in sandy terrain.

In the New Zealand-based Spyfish Aotearoa, citizen scientists identify and count fish in images to help teach an AI tool about fish.

Ethics thinking

The Connected article Emotional robots explores the development of machines (robots) that imitate human emotion and thought from a social and ethical perspective.

AI and generative learning have many positives and quite a few drawbacks. The Futures thinking toolkit can be customised to explore how changes in this technology may impact our lives and the lives of future generations.

Due to its ubiquitous nature, Minecraft is very helpful when explaining complex computing concepts. This article uses Minecraft to explain how global circulation climate models are built.

Useful links

Explore the resources in the Artificial intelligence section on the Office of the Prime Minister’s Chief Science Advisor website.

Download from the AI Researchers Association White paper: Aotearoa New Zealand Artificial Intelligence – A Strategic Approach. This 2021 report looks at what artificial intelligence is, how it is or could be used in New Zealand, and the risks that need to be managed so that all New Zealanders can prosper in an AI world.

OpenAI is the company behind ChatGPT. Its website states, “OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity.” 

  • Do you think an organisation can claim to “benefit all of humanity”? Who decides what benefits humanity?

Look at some of the ethical issues posed by ChatGPT and the potential for harm and misuse in the opinion-based article ChatGPT is suddenly everywhere. Are we ready?

ChatGPT and other LLMs require significant input from humans and rely on our feedback to improve the technology. This article looks at LLMs from a sociological perspective. 

The Royal Society Te Apārangi Mana Raraunga Data Sovereignty 2023 report outlines what data sovereignty is and why it matters in Aotearoa New Zealand. Listen to this RadioNZ interview with Professor Tahu Kukutai as she breaks down concepts like Big Data and Māori data sovereignty.

Acknowledgement

This article was written by Professor Albert Bifet, Director of Te Ipu o te Mahara AI Institute at the University of Waikato. Professor Bifet combines factual information and some of his thinking around the AI large language model ChatGPT.

Published:29 March 2023