Skip to main content

Overview of AI and Law

By Quinten Steenhuis, September 7, 2023

What is Generative AI?

Generative AI refers to artificial intelligence techniques designed to produce new, unique outputs, be it text, images, or even music. Think of it as a type of AI that can channel its "inner artist." While humans rely on their experiences and inspirations, generative AI uses vast datasets and intricate algorithms. Tools like MidJourney have wowed users by generating hyper-realistic images, even conjuring up scenes straight from fantasy. Meanwhile, ChatGPT, since its introduction in November 2022, has been showcasing its knack for dishing out coherent and contextually fitting text responses.

At its core, generative AI employs machine learning, a subset of AI that learns from data. To grasp how it works, think of two main stages:

  1. Gathering Examples: Just like you'd show a child various shapes to teach them geometry, AI needs examples. The more data it gets, the more it learns.
  2. Training and Feedback: Here, the AI is repeatedly exposed to these examples, and errors are corrected. Imagine learning a new instrument: the more you practice, the better you get.

Now, imagine your smartphone keyboard predicting your next word. Generative AI is like an enhanced version of this, but instead of suggesting a single word, it can spin out sentences, paragraphs, or even full essays. And while it seems almost magical, remember that it isn't infallible. It's basing its guesses on the immense data it's seen, aiming to get it right or, at least, plausible.

Importantly, these models don't understand context or truth in the way humans do. Their "knowledge" is grounded in patterns and frequencies from their training data. As a result, while they're often accurate, they can occasionally misstep.

Behind the Scenes: How Does It "Think"?

Generative AI operates on neural networks, structures inspired by the human brain. Imagine a vast web, where each connection point is a piece of knowledge or a pattern. This intricate web is called a transformer architecture, which is particularly adept at handling sequences, perfect for language tasks.

Here's a quick overview:

  1. Layers and Neurons: Just as our brain has neurons, AI models have virtual "neurons" organized in layers. These neurons activate in response to input data, helping in decision-making.
  2. Attention Mechanism: Ever had someone say, "Pay attention!"? The AI has its own attention system, allowing it to focus on relevant parts of data when making decisions.
  3. Fine-tuning: Post the initial training, models can be tailored for specific tasks. Think of it as a musician first learning the basics and then specializing in a particular genre.

Generative AI is modeled on the way that humans think. The key difference is that it doesn't have a representation of what's true or correct. It only draws on examples of what is most probable based on its training data.

Large language models you can explore (for free or inexpensively)

Selected readings

Sample Syllabus Statement

Policy on use of generative AI and large language models (e.g., ChatGPT)

What are generative AI tools?

The line is getting fuzzier by the day, but in general, any tool that you provide a prompt and get back new or substantially revised text is one I would describe as generative AI. These tools all have a creative function that distinguishes them from tools that simply correct your work.

Examples of "generative AI" tools:

  • ChatGPT
  • Bing Chat
  • Google Bard
  • Google Docs's "Help me Write" feature

Examples of tools that are not "generative AI":

  • Spellcheck and grammar check
  • Microsoft's "Editor" tool
  • Grammarly
  • WordRake

Why we provide guidance on use of generative AI

  1. To ensure substantive accuracy of output;
  2. To help you understand how to avoid violation of client confidentiality and privacy;
  3. To help you build competency in the use of these tools and to be able to produce your own work in instances where the tools will fail;
  4. To provide transparency into your authorship so that I can appropriately assess your work.

Limits on use of AI

You may use generative AI tools on all assignments in this course. Where possible, I will provide instruction on how to appropriately use AI on a given assignment.

You should also always follow these general rules:

  1. Don't substitute the AI's judgment for your own. When I ask you to reflect, synthesize, or argue, I am asking for your thoughts, not those of the AI. You may use the AI as an "editor." For example: you can ask it if your reflection is missing anything.
  2. Don't put private or confidential information into commercial AI tools. I hope to provide you access to a Suffolk-owned tool which will be safe to use for this purpose.
  3. Follow the "50% rule": your original writing combined with any prompt to the tool should reflect at least 50% of the word count of the final content.
  4. Check your work for accuracy. Large language models are predictive tools and can make mistakes. Your job is to verify your work.
  5. Let me know how you used each tool. I'll occasionally ask you to provide a log of your conversation with an AI tool, but you should always use attribution.