Instruction

Looking at Chatterbots and Chatbots

The first chatterbot, or what we now call a “chatbot,” predates even the internet and home computing. Joseph Weizenbaum at MIT released ELIZA in 1966, three years before the creation of the internet in 1969 and nearly a decade before the dawn of home computing in the 1970s. ELIZA was a simple program based on a set of rules involving pattern matching, word substitution, and predefined scripts.

The most famous script was DOCTOR, in which ELIZA mimicked a Rogerian psychotherapist by reflecting a user’s statements back to them.

For example, here’s a simplified conversation:

• User: “My head hurts.”

• ELIZA: “Why does your head hurt?”

• User: “I don’t know.”

• ELIZA: “Do you often feel unsure about why your head hurts?”

In this case, ELIZA would first identify the keyword “hurts” and then use a pre-programmed script to prompt the user for more details. The script would apply transformations, such as changing “I” to “you,” to form a response. Other keywords, such as “don’t know,” would trigger different scripts, leading to other prompts and transformations.

In this regard, ELIZA didn’t truly understand inputs from users. The interaction was entirely determined by how well the scripts were written for the text from the user. To change the interaction, the scripts needed to be updated manually.

While True, Do Chat

You may have created a simple version of a chatbot early in your software career. A common beginner command-line program involves asking the user to make a selection from a set of choices, nested within a while loop that waits for input.

For example, you may have written a program like this:

Program output:

What do you need help with today?
Enter one of these numbers:
1. Sales
2. Technical Support
3. Something else

User input:

4

Program output:

I'm sorry, I don't know that input.
  
What do you need help with today?
Enter one of these numbers:
1. Sales
2. Technical Support
3. Marketing

User input:

3

Program output:

No one is here. All is dark. You are likely to be eaten by grue

Setting aside the geek joke (it’s from Zork, a text-based adventure game that’s similar to a chatbot!), this type of command-line program represents a basic form of a chatbot.

Then Came ChatGPT

While ELIZA and simple command-line programs laid the groundwork for conversational systems, ChatGPT represents a significant leap in how machines can understand and generate human language. ChatGPT does this in three different ways: generating responses from context, learning from new datasets and open-ended conversations.

Generating Responses from Context

Unlike ELIZA, which relied on rigid scripts and pattern matching, ChatGPT uses deep learning models trained on vast amounts of text data. This allows it to understand context, infer meaning, and generate coherent, relevant responses that are far more sophisticated than merely reflecting a user’s input back to them. ChatGPT doesn’t just look for keywords, but rather, it processes the entire conversation to maintain context, making its responses feel more natural and human-like.

Learning from New Datasets

While ELIZA required manual updates to its scripts to change its behavior, and simple command-line programs are static unless reprogrammed, ChatGPT continuously improves by learning from extensive datasets during its training phase. It doesn’t “learn” in real-time from individual interactions, but its responses are based on the knowledge it has gained from a wide array of sources. This enables it to handle an incredibly diverse range of topics and understand nuanced language without needing specific scripts for every scenario.

Open-Ended Conversations

The command-line example shows a program with a limited set of choices and responses. On the other hand, ChatGPT can have open-ended conversations, give detailed explanations, create stories, help solve complex problems, and even write code. Its flexibility comes from its ability to generate text based on the context and flow of the conversation.

Using ChatGPT in an iOS App

You can also easily integrate ChatGPT into your own iOS apps. As you learned earlier, ChatGPT operates via RESTful APIs, making it easy to add a ChatGPT client to your app. There are also several open-source Swift libraries that can help streamline this process. Two of the most popular are:

• OpenAISwift provides a simple interface for interacting with ChatGPT. It supports various models, including GPT-3 and GPT-4 and offers features like streaming completions and structured concurrency.

• OpenAIKit is another popular library for interacting with ChatGPT. It provides extensive customization options, including SSL certificate verification for secure API calls, custom endpoint support, timeouts, and more.

Using one of these libraries can save you time and effort compared to building and maintaining your own client from scratch. However, for learning purposes, you’ll continue using the ChatGPTClient you created earlier in this course.