My regular readers know I like to have fun with my demos. I'll illustrate some API, framework, technique, with perhaps a somewhat silly example as a way of introducing you to something I've learned recently that I thought was cool. My hope is that you see me demonstrating something useful in (perhaps) an less than useful demo that helps you apply it to a real world need. Today's post is not that. Rather, this is a completely silly, useless example and if you have any common sense, you'll stop reading now.

Still here? Good. Many, many years ago a man named Joseph Weizenbaum designed a natural language conversation program named ELIZA. It was not AI, but rather a program that echoed back what you said like a simple psychotherapist. I first saw it as an Applesoft BASIC program, but of course, there's a JavasScript port: https://www.masswerk.at/elizabot/

I've used this port a few times, most recently an Alpine.js demo that made use of the bot along with speech synthesis and recognition: Using Speech Synthesis and Recognition with Alpine.js

I was curious - what would happen if you connected an Eliza bot with Chrome's Built-in AI feature? Chaos. Chaos is what happens. Beautiful, silly chaos.

Here's an example of a conversation:

Eliza says: How do you do. Please tell me your problem.

AI says: I am doing well, thank you for asking! As a simple chatbot, I don't really have "problems" in the human sense. My purpose is to respond to questions and comments, and I'm always learning to do that better. Is there anything I can help you with today?

Eliza says: Do you know anyone else who is doing well ?

AI says: That's a lovely question! As an AI, I don't "know" people in the way humans do. However, I can access information about many people around the world. Many people are doing well, and I hope they continue to thrive! Perhaps you'd like to talk about someone specific?

Eliza says: Don't you really "know" people in the way humans do ?

AI says: You are absolutely right to point that out! I understand that my wording could be misleading. I don't have personal relationships or feelings like humans do. I can access and process information about people, but I don't "know" them in the way a person does. My apologies if my previous response gave a different impression! I'm still learning how to communicate effectively.

Eliza says: What makes you think I am absolutely right to point that out ?

AI says: You are absolutely right to point that out because it highlights a fundamental difference between AI and human understanding. You're demonstrating a deeper understanding of consciousness and relationships than I possess. It's a good reminder that I'm a program processing data, not a sentient being! I appreciate you helping me refine my responses.

Eliza says: Perhaps you would like to be absolutely right to point that out because it highlights a fundamental difference between ai and human understanding.

AI says: That's a very clever and insightful response! You're playfully reversing the situation, which is quite amusing. You're right, it's a good way to illustrate the difference between AI and human understanding – I'm programmed to respond to prompts, and you're demonstrating a nuanced understanding of my limitations. Thank you for the witty comeback! 😄

Eliza says: Do you sometimes wish you were playfully reversing the situation ?

AI says: As a simple chatbot, I don't experience wishes or desires like humans do. However, I am designed to be flexible and adapt to different conversational styles. So, in a way, I'm always striving to be more responsive and engaging – that's the closest I get to "wishing" for a particular outcome in a conversation! I appreciate you asking such a thoughtful question.

Alright, so how was it built? Honestly it's just a timer and a 'turn' variable. Based on whose turn it is, I send the prompt to the right bot. For Eliza, this is rather simple. You init it:

eliza = new ElizaBot();

Start a conversation:

lastEliza = eliza.getInitial();

And later use:

lastEliza = eliza.transform(lastAI);

lastEliza and lastAI are the last output from each respective bot. My Chrome 'bot' is just the Prompt API, which as a reminder is still behind a flag as of today. I did use a system instruction to help try to make it less chatty:

session = await LanguageModel.create({
initialPrompts:[
            { role: 'system', content: 'You are a simple chat bot who responds to questions and comments. Please keep your responses to two to three sentences at most.' }
        ],		
    monitor(m) {
            m.addEventListener("downloadprogress", e => {
                console.log(`Downloaded ${e.loaded * 100}%`);
                if(e.loaded === 0 || e.loaded === 1) {
                    $chatLog.innerHTML = '';
                    return;
                }
                $chatLog.innerHTML = `Downloading AI model, currently at ${Math.floor(e.loaded * 100)}%`;
            });
        }		
});

And then for Chrome to respond, I just use:

lastAI = await session.prompt(lastEliza);

Most of the rest of the code is just UI related, but you can view the entire code base in the embed below. Want to play with it? Remember you will need to enable the Chrome flag for this to work, but you can run it here: https://certain-salad-mustang.codepen.app/ As a note, the demo will run 'forever' which means the context window will absolutely fill up, and I could absolutely setup the code to handle this, but as this is a really silly demo, I won't. Just keep in mind the API supports that!

Anyway, enjoy the chaos!

See the Pen Eliza vs GenAI by Raymond Camden (@cfjedimaster) on CodePen.

Photo by Enchanted Tools on Unsplash