Generative Artificial Intelligence

This welcome video was created using generative AI tools in the summer of 2023. It was made with the intent to show both the capabilities and limitations of these emerging technologies.

It was created using the following tools: 

• DALL-E 2 generated the image of the library ram, using the prompt: “a photo of ram dressed like a librarian, wearing a bowtie and glasses, background full of books”
• MyVocal.AI was used to synthesize a voice, reading a human-generated script. 
• HeyGen was used to create the video animation by bringing all the materials together. 

What do you notice about the video? 

While the resulting media might seem a little surreal or silly, the aspect we wanted to highlight was the speed with which it was generated. All of the media materials were created within a matter of minutes, much longer than it took to write the initial script. 

GenAI tools, like ChatGPT, can be incredibly powerful in the research process for tasks like brainstorming, writing abstracts, or even getting feedback on your writing.

However, the technology is designed to make things up based on data that it has been trained on and should not be trusted to generate accurate, truthful information. This is why critical evaluation of the content is required. If you are looking for well vetted information or resources, the library search and our databases are a better bet. 

Artificial Intelligence (AI) can be tricky to define because of its position in the popular imagination, where it is often understood in science fiction as a type of machine that thinks like a human.

This explains why Merriam-Webster provides two definitions for AI: 

1. a branch of computer science dealing with the simulation of intelligent behavior in computers (the technical one) 

2. the capability of a machine to imitate intelligent human behavior (the popular imagination one)

Within computer science there are different ways of understanding the term “AI”:  

1. The first is a colloquial definition that applies to any system or application, using an element of artificial intelligence (machine learning, deep learning, Large Language Models, etc.) [Artificial Intelligence and the Future of Teaching and Learning, Department of Education]
    
2. The other definitions are more technical and relate to how the AI system is built  [Glossary of Artificial Intelligence Terms for Educators – CIRCLS]:
   • through the use of rules provided by a human (rule-based systems); or 
   • with machine learning algorithms, for which there two broad categories: 
     • Supervised: Sometimes referred to as "traditional" machine learning, this is a technique in which the algorithm learns to make predictions based on previously labeled data in the training dataset. One of the most common methods is called classification, in which the model tries to predict the correct label for a new given input (i.e. email filtering, book recommendations, etc.) 
     • Unsupervised: Sometimes referred to as "deep learning," this is a newer technique, in which the algorithm learns to identify hidden patterns in the training dataset, and how it identifies these patterns is often unknown to the user. One of the most well known methods is called neural networks, which is way that generative AI works. 

General vs. Generative AI:

• With regards to the popular idea of “AI,” computer scientists define this as “Artificial general intelligence (AGI),” which is a form of AI that does not yet exist and “would be when an AI system can learn, understand, and solve any problem that a human can.” [Glossary].

• Generative AI, on the other hand, is simply a system built with a neural network approach to AI, in which content is produced (LLM, text-to-image generators, etc.).

• How exactly does generative AI learn? There are three main phases:

  • Pre-training: Pre-training the model is the initial learning phase, in which an incredible amount of data is used to train the entire foundation of the generative AI model. 

  • Fine-tuning: Fine-tuning is the phase in which humans intervene to help refine the results generated by the neural networks. 

  • Embeddings: In short, an embedding is a method in which a synthetic material generated by the machine is fed back into the neural network in order to fine-tune it.