I was in 4th or 5th grade when I met ELIZA. I was already a computer geek and deeply fascinated by concepts of AI, at least through exposure in scifi. ELIZA was my first exposure to AI in the actual world when she told me that she’d be my therapist. (I didn’t know I needed a therapist, but that was besides the point.) She asked, “Is something troubling you?”
“Yes,” I might have responded, “I want to learn how to create you.”
Ok, that was almost 40 years ago so I don’t remember exactly how I responded but at some point I’m certain I asked her how she was created. She didn’t tell me. She couldn’t tell me.
I was stuck. No one around me knew how ELIZA was programmed, and I didn’t know the right questions to ask. Life moved on. I still programmed, but I never learned anything more about AI or what I now know is a subset of AI programming called Natural Language Processing (NLP).
Skip ahead to the early 2000s. I stumbled upon something called AIML which stands for Artificial Intelligence Markup Language, and learned that it was a way of creating programs similar to ELIZA. That propelled me into the world of chatbot programming. I spent the next several years learning the ins and outs of AIML, creating several chatbots, and even entering one of them, Zoe, into a Turing Test-like contest called the Loebner Test. (I even came in second place one year!)
But I wasn’t happy. All I was doing was programming something that mimicked human speech. It didn’t think. It just responded the way I programmed it to respond, even if that response contained potential random answers. It wasn’t thinking. It was impostering. (Yes, I made that last word up. I’m a writer. I do these things.)
I learned how to do something like ELIZA**. And it was very anti-climactic.
The definitive moment when I realized I had made no progress on creating AI was when my oldest son was 18 months old.
One night, I had set up our small Christmas tree after he went to bed. In the morning, when he emerged from his bedroom, I had it all lit up, and we made a big deal of the tree. Then we settled down to eat breakfast. He was eating his banana and had a mouthful when I asked:
“So, do you like it?”
Without saying anything, I saw his eyes dart over to the tree. In that moment, I realized that even with a break in the conversation from when we had talked about the tree several minutes earlier, my son still knew that when I said “it,” I meant the Christmas tree.
How did he know? I could have meant the banana he was now eating. I could have meant his bottle sitting in front of him. I could have meant the chair he was in. I could have meant anything, but my 18-month-old correctly identified “it” with “Christmas Tree.”
One of the hardest things to program my chatbot, and my little one just got it. 18 months of listening to language and this came naturally to him.
Yes, I had a hook in my chatbot to keep track of the main noun, so if a conversational break happened in a conversation with Zoe, she would have that last noun or a history of them in her memory. But there was only a 50% chance she got it right, and she was just guessing, not thinking. She was not intelligent.
I continued to work on my chatbots for a while after that, but knowing that I had successfully created intelligence… biological intelligence… and knowing that I was unlikely to repeat that on my computer, set me off in different directions and I haven’t worked on my chatbot in several years.*
But I did create intelligence. Human intelligence. And not just once. I repeated this a few years later when I gave birth to my second child. There’s something very satisfying about that. It’s not the same as creating a sentient positronic neural network, but gratifying nonetheless.
* I still love AIML and wholeheartedly support it. If you want to learn how to make a chatbot, AIML is the way to go. But if you want to create Data, or R2-D2, we’re going to need something a little more advanced.
** There are several online versions of ELIZA where you can chat with it yourself. I’m sure there are more than just these two, but these are ones that I found that are active:
[…] So I’m going to break down the slight of hand for you. What qualifies me to do this? First, I have a masters in computer science with a focus on AI (which means I’ve studied a bunch of AI and learning algorithms and natural language processing). But second, and more importantly, I used to spend my free time developing the kind of AI that you could have a conversation with in an attempt to convince you that my software was human. More than that, I entered in my creations into the Loebner Contest, an instantiation of the Turing Test. One year, I even came in second place. (I wrote about why I gave it up in this article last year.) […]
[…] is a branch of computer science that deals with understanding human language (spoken or written). I’ve written and spoken about the challenges of NLP several times before, particularly when it comes to keeping an understanding of the context going through a […]
[…] branch of computer science that deals with understanding human language (spoken or written). I’ve written and spoken about the challenges of NLP several times before, particularly when it comes to keeping an understanding of the context going through a […]