Ozeki Technical 001: The Beginning – What is Generative AI?

Welcome to the first of many technical blog articles from Ozeki. This series aims to explore the technology and products that drive electronic automated negotiations. We will delve into different topics, starting with the fundamentals and advancing to complex aspects of our systems and solutions. Specifically, how AI can revolutionize current negotiation methods without disrupting existing workflows. However, before we start building these exciting features, we need to lay down the fundamentals.

Regulatory compliance has always fascinated me, perhaps because it involves some of the most challenging software mechanisms, or because it represents real-world scenarios where finance and law intersect. Over the past 20 years, this interest has led me to fascinating discoveries and played a significant role in shaping Ozeki's technology and architecture.

Let's embark on this journey, noting that our audience includes non-technical individuals such as legal professionals, sales executives, and CEOs. These individuals are at the forefront of innovation and progress when it comes to utilizing software development in regulatory compliance.

Take your time to understand this topic. If it seems complex or tedious, break it down into smaller parts. If you're still interested and have questions, feel free to reach out.  If you are already familiar with these fundamentals, feel free to skip ahead to our next blog post.

Understanding Generative AI

Imagine you have a magic toy box. This toy box can make new toys all by itself. You give it some old toys, like a teddy bear and a toy car, and then it learns how to make new toys that look like the teddy bear and the toy car, but are different and unique.Now, let's talk about how this magic toy box works. Inside, it has a special brain called a neural network. This brain has many tiny parts called nodes, which are like little helpers. Each helper looks at the old toys, thinks really hard, and then decides how to make new toys. Some helpers look at the eyes of the teddy bear, others look at the wheels of the toy car.  When all the helpers work together, they can create new toys that look a bit like the old ones but are brand new and special. This is what Generative AI does—it helps computers make new things by learning from old things, just like our magic toy box makes new toys by learning from old toys.

Below is a little more technical explanation. Feel free to skip it if too much.

Before diving into Generative AI, let's first clarify the meaning of the term. "Generative" gained prominence around 2010 with the advent of Generative Adversarial Networks (GANs) by Ian Goodfellow. GANs introduced a game-like structure employing two neural networks (Goodfellow et al., 2014).

What is a Neural Network?

A neural network is an engineering representation of how the human brain processes information. The human brain contains neurons, and in software, these are typically referred to as nodes.

Here's a simplified explanation of how a neural network operates:

• Neurons (Nodes): Each node receives one or more inputs, processes them (usually by summing them), and then passes the result through an activation function to produce an output. ‍
• Layers: These nodes are organized into layers, typically consisting of an input layer, hidden layers, and an output layer.

Generative AI is like a magical machine that can make new things all by itself! Instead of just sorting or understanding what we show it, this machine can create all kinds of new stuff. Here's what it can make:

• Text: It can write stories, poems, or even messages just like how people do. Imagine a robot writing a bedtime story for you! ‍
• Images: It can draw pictures and create new artworks. Think of it as a robot artist that can paint new and exciting pictures. ‍
• Music: It can compose new songs and melodies. It's like having a robot that can create music just for you! ‍
• Videos: It can make new videos and animations. Imagine a robot filmmaker creating a new cartoon for you to watch. ‍
• Voice and Speech: It can talk like a human. This means it can create new voices and even read stories aloud. ‍
• 3D Models: It can make models that look like real objects. This is like having a robot that can create new toys or objects for your games. ‍
• Synthetic Data: It can create pretend data that looks real. This is helpful for scientists and researchers to study and learn new things.

When we talk about the magic machine making these different things, we call it different names based on what it makes. For example:

• If it makes text, we call it "Generative Textual AI."
• If it makes pictures, we call it "Generative Image AI."

In our work, we will primarily work with Textual, Image, and Synthetic Data AI.

The Leader in Generative AI: OpenAI

The current leader in the space of Generative AI is OpenAI. The goal/mission of OpenAI is to benefit humanity through the development of Artificial General Intelligence (AGI).

What is AGI?

Imagine you have a really smart robot friend. This robot isn't just good at one thing, like playing a game or answering questions. Instead, it can learn to do many different things, just like a person can. It can help you with your homework, tell stories, draw pictures, and even fix your toys.

This super-smart robot can also understand what you want and think about the best way to help you. So if you want to build a tower with blocks, the robot knows that's your plan and can figure out how to stack the blocks to make a tall tower. It's like the robot has a smart brain that understands your ideas and uses its own thinking to help you.

And here's something really cool about AGI: it has two special attributes that make it very smart:

• Intent: The robot understands what you want to do. ‍
• Reason: The robot thinks about the best way to help you get it done.

Intent vs. Reason in AGI

Now, let's clarify that intent and reason are not the same, and we can assert with solid proof that one is harder to achieve than the other.  If you had to guess, what do you think is harder to create in AGI: intent or reason? It turns out, intent is much harder to create.

Reason consists of how information is processed and what decisions are made based on rules, logic, and patterns. The entire field of computer science has been heavily involved in solving these aspects. We can refer to this process as polynomial (Goodfellow et al., 2016).

On the other hand, intent is challenging because it requires understanding goals, desires, and motivations. This involves delving deep into interpreting human behavior and intentions. In the world of software, the closest concept to this might be found in the ideas of Trygve Reenskaug and his book “Working with Objects,” where he expressed doubts that object-oriented programming can exist without behavior (Reenskaug, 1996). Understanding intent involves a level of complexity akin to grasping human consciousness, which includes empathy, context, and subjective experiences (LeCun et al., 2015).

Let's stop here with the fundamentals of Generative AI. I hope this session has helped you better understand the purpose of using it. In our next technical article, we will focus on understanding the semantics of a legal contract and how AI helps in this regard.

References

• Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative Adversarial Networks. arXiv preprint arXiv:1406.2661.
• Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.
• LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444.
• Reenskaug, T. (1996). Working with Objects: The OOram Software Engineering Method. Manning Publications.
• OpenAI (2015). Introducing OpenAI. [Online] Available at: https://openai.com/blog/introducing-openai/