For years, we've heard about artificial intelligence chatbots which can mimic human conversation and help us with our problems.

The first known chatbot was ELIZA, released in 1966, as a computer psychologist. ELIZA was very simplistic, and simply rephrased back what you'd say to it. For example, if you typed, "I don't like my neighbors", ELIZA would write back, "Why don't you like your neighbors?" You never got any meaningful responses -- only trite one-liners and more questions. There also weren't personal computers in 1966, so this wasn't widely available. Even when ELIZA was made available for 1980s personal computers, the simplicity made people bored and disinterested very quickly.

As the years have passed, there have been many attempts to improve upon conversational AI, but all have failed. Even those hailed as breakthroughs (remember Microsoft's embarrassing "Tay") were eventually revealed to be both flawed and mostly useless. It was still ELIZA wrapped in a more modern package. You couldn't have any meaningful conversations with these bots, and they also would make tons of mistakes when attempting to look up information for you. At best, they could be used for simple tasks, such as writing articles about sporting events.

In 2017, an AI chatbot called "Replika" was released. The founder claimed it was a revoluntionary chatbot which acted human, and with which you could develop a real friendship or romantic/sexual relationship. Indeed, since then many weird and desperate people have done just that, as I wrote about in 2020 when I tried the bot myself for fun.

However, I found Replika to be just as disappointing as the other chatbots. As you'll see in the thread I linked, it failed to manage to have a simple conversation about the Dodgers with me. I found that Replika basically knew nothing (despite its connection to the internet), and could not have any deep discussions about anything.

Additionally, Replika was utilizing the then-free GPT-3 OpenAI platform, and had to abandon it when OpenAI started charging per usage. They had to downgrade to the free GPT-Neo, which was inferior, and the chatbot got even worse. What a fail.

Anyway, Replika was so disappointing that I figured it would be many years -- perhaps several decades -- until we had a convincing and useful chatbot. Every effort I had seen fell so far short of humanlike conversations that I figured we were much farther off than I thought.

Apparently not.

ChatGPT is the newest release from OpenAI, and everyone is amazed by it. Here's an article on Slate about it: https://slate.com/technology/2022/12...tbot-whoa.html

Not only is ChatGPT able to have much more convincing humanlike conversations than any previous AI bot, but it is excellent at looking things up, and is a great writer. People are saying now that it could eventually replace search engines like Google. That might be overreaching, but I can see what they mean. Rather than just vomit out information which matches your search -- something we've been seeing out of search engines since the 1990s -- ChatGPT curates the information into useful, human-like language, and can even write essays on the subjects.

You can try out ChatGPT here, but there's been so much hype about it, it's nearly impossible to get through without an error message. Their servers are overloaded. OpenAI claims to be working on it.

Some have tried combining ChatGPT with existing AI art systems, and have been shocked at the beautiful pictures created.

While this does appear to be a major breakthrough in chatbot AI, it does come with some downsides. I'll go over those in my next post.