From Eliza to ChatGPT: the 60-year history of chatbots

People have been trying to talk to computers for almost as long as they’ve been building computers. For decades, many in the technology industry have been convinced that this was the trick: If we could figure out a way to talk to our computers the way we talk to other people, and get computers to talk back the same way, those computers would easier to understand and operate, more accessible to everyone and simply more fun to use.

ChatGPT and the current revolution in AI chatbots are actually just the latest iteration of this trend, which dates back to the 1960s. That’s when Joseph Weizenbaum, a professor at MIT, built a chatbot called Eliza. Weizenbaum wrote in an academic journal in 1966 that Eliza “enables certain forms of natural language conversation between humans and computers.” He set up the bot as a therapist, a vessel into which people could pour their problems and thoughts.

The technology behind Eliza was incredibly primitive: users would type into a text field and the bot would select from a number of predefined answers based on the keywords in your question. When he didn’t know what to say, he would simply repeat your words: you said, “My father is the problem,” and he replied, “Your father is the problem.” But it worked! Weizenbaum wrote in another newspaper a year later that it was difficult to convince people of this was not a human on the other end of their conversation.

What Eliza showed, and what other developers and engineers worked on over the next six decades, is that we treat our devices differently when we think of them as moving, human-like objects. And we’re remarkably willing to treat our devices that way. (Have you ever felt sorry for your robot vacuum cleaner as it makes its way through your living room, or thanked Alexa for doing something for you?) It’s human nature to anthropomorphize objects, to imbue them with human qualities, even if they don’t have that. each. And when we do that, we are kinder to those objects; we are more patient and cooperative with them; we like to use them more.

We treat our devices differently when we think of them as moving, human-like objects

Examples of what this might look like are all over pop culture. The Star Trek computer is a classic source of inspiration for Silicon Valley types – “Tea. Earl Grey. Hot.” – just like Scarlett Johansson’s ambient AI Her. HAL9000 in 2001: A space odyssey is both an inspirational and cautionary tale, just like WOPR War games. These are computers that think, talk and understand.

The only problem with all these humanoid computers? They are remarkably difficult to remove. A bot like Eliza could generate a somewhat convincing conversation, but that only gets you so far. Most of the time, it’s your computer’s job to do things, and these chatbots have never been very good at that. People have been working on that too: a group at Xerox PARC built a chatbot in the 1970s that you could use to book airline tickets, but it was finicky, slow and incredibly expensive to use. There have been numerous attempts to do the same since then.

Many versions of these tools have appeared over the years. There were other early chatbots, such as Dr. Sbaitso, Parry and Alice. In the early years, there was SmarterChild, the irreverent AIM bot that introduced so many teens and tweens to the idea that a computer could talk back. There was the age of voice assistants, when everyone thought Siri, Alexa, Cortana, Bixby, and countless other tools would change the way we used our devices and got things done. With each generation we got a little closer to a computer that could both talk and walk. But nothing ever ended up there. When was the last time you asked Google Assistant to book a flight for you?

Now we’re at the dawn of a new era in chatbots, one that many in the industry think could get the job done. Tools like ChatGPT and Google Gemini, and the underlying language models that power them, are much better at understanding you and getting things done for you. Microsoft is betting that Copilot will be your all-day AI companion at work; Google puts Gemini in the same position. These tools aren’t perfect, or even close – they make things up, they misunderstand, they crash, they go completely haywire every now and then – but they are the closest thing to a conversational computer we’ve seen yet. have seen. You talk to it like you talk to someone, and it talks back.

Is it cool to think that a computer can fit into your life like an assistant or friend could, or is it horrifying?

The rise of these powerful bots raises many questions. Is it cool to think that a computer can fit into your life like an assistant or friend could, or is it horrifying? Is there something fundamentally wrong with the idea of ​​having an AI companion like the ones from Meta, Replika, or Character.AI, or is there something beautiful about enabling those kinds of relationships? How much better do these bots need to get before we can really rely on them? Will they ever be that good?

But most importantly, we finally get an answer to the question we’ve been asking since the 1960s: Is this the way computers should work? Many people have believed that for years, many others have said they are wrong and that training computers to work like humans would make them less efficient and more annoying. But we never found out, because the bots were never good enough to really compete with all the other ways we interact with our devices, our information, and each other. Now they are close, or at least much closer. And so we can really find out if Joseph Weizenbaum was right all those years ago: that conversation is the future of computing. The chatbot has been the future of computing for almost as long as there have been computers, and now it’s here.

Leave a Reply

Your email address will not be published. Required fields are marked *