What is Generative AI?
AI has a long history. Scroll through the timeline below to explore a selection of the key advances in AI. Click on the image in the top right hand corner to view fullscreen.
Generative AI is a type of artificial intelligence that uses machine learning to generate new content by analyzing and processing vast amounts of data from diverse sources. Generative AI tools can generate text, images, video, sound, and code. Different tools are trained on different datasets and with different training methods. The generated responses of these tools are probabilistic, which can result in errors in responses. Large language models (LLMs), for instance, specialize in analyzing and processing text and generating new text. Different LLMs have distinct datasets and employ unique training methods. GPT 3 and GPT 4 are examples of LLMs. OpenAI’s ChatGPT is a chatbot created on GPT 3 or GPT 4.
A useful glossary of AI terms can be found here and a great brief introductory video from the Wharton School is below:
While generative AI is not new, OpenAI’s launch of ChatGPT in November 2022 marked the fastest recorded adoption of a technology tool to date. Over the following months, the release of similar text-based generative AI tools from Microsoft’s Bing to Google’s Bard, in addition to improvements in tools have contributed to a perception of an explosion of AI.
Indeed, the rapid proliferation of tools and advancements in technology saw over 100 leaders in AI technology write an open letter urging a collective pause on AI developments more powerful than GPT 4 to give time for security and safety features to develop and for the creation of regulation and governance structures.
The need for such regulation or governance extends to full nations, but also to specific sectors, such as post-secondary education, and in turn, McMaster University. Broader issues related to generative AI include privacy of personal data, risks of misinformation, existential risks, concerns about job dislocation or loss, environmental costs, labour exploitation, and copyright.
The next section in this book will look at some of these risks in more detail as they relate to post-secondary education.
While each text-based generative AI tool has specific functionalities, some common capabilities include:
- Create informative, well-written text: prose, poetry, dialogue, code
- Provide examples and references *references may be ‘hallucinated’
- Generate outlines, questions, tables, long form text
- Summarize inputted text
- Provide feedback on text – both form and structure
- Explain concepts at different levels of understanding
- Translate between languages
- Remember within a chat thread – follow-up prompts
Limitations of text-based generative AI tools include:
- Hallucinations: confident declarations that are factually inaccurate (e.g., references to articles that don’t exist)
- Uneven access material after 2021: the free version of ChatGPT cannot access the web, though the subscription model can, as can Microsoft’s Bing
- Biases in training data are replicated in generated responses
- There is variation in responses based on the wording and framing of the user’s prompt
- If using ChatGPT 3 (free version) there can be lag times or delays in access if demand is high; Bing and ChatGPT 4 do not experience these delays.
the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in data.
based on or adapted to a theory of probability; subject to or involving chance variation.
a computer program designed to simulate conversation with human users, especially over the internet.