Eliza old talk, praise you for singing

Eliza: How are you? Please tell me your problem.

Me: I feel sorry.

Eliza: What does feeling sad remind you of?

Me: My college days.

Eliza: Your college days?

Me: Yes, when I had lots of friends and I used to go out with them every day.

Eliza: Good.

Me: And now I’m alone.

Eliza: Do you believe it’s normal to be lonely?

Me: Not always, but there are times when it’s not normal.

Eliza: What event are you thinking of?

Eliza, as you can tell, and if you don’t know, is a healer of sorts. She comes across as encouraging and slowly ready to probe deeper. Even with those few lines, you can understand how people can find it soothing and even therapeutic to talk to him.

So, who is this thoughtful lady? Well, to start with, she’s not a woman. He is a programmed figment of the imagination of a certain computer scientist. His name is Joseph Weizenbaum, and in the 1960s he created ELIZA as a demonstration of what could be considered artificial intelligence. His idea was to answer Eliza as a psychiatrist. In particular, she seems to consider the patient’s questions and often return them so that the patient may consider them in return.

When Weisenbaum sent Eliza out into the world, many people believed she understood exactly what they were saying. But of course Elijah did no such thing. Looking for patterns in what patients typed in, and using some artful word and phrase substitutions, Eliza churned out responses that seemed appropriate, even sympathetic, and intelligent.

For a while. Chat with Eliza, and you’ll quickly realize that she’s just a bot. In our “conversation” above, I have dispensed with the following few lines in which Eliza says increasingly incomprehensible things.

However, the point is this: ELIZA was the first attempt to program a certain kind of intelligence into a computer. Or at least, to suggest to a casual user that the computer was displaying some intelligence.

In the nearly 60 years since then, those efforts at artificial intelligence have grown steadily more sophisticated. For example, when I dabbled in AI in the late 1980s, my small research group was very excited by the system we’d built that could handle basic queries about restaurants in our area, return recommendations . Although “basic” is the operative word here. The lesson from that time may seem obvious: any “intelligence” displayed by our system was only as good as the knowledge we fed it. Any hope of understanding a question depended on that bank of knowledge. That’s why another AI group in our company was completely focused on building a vast store of knowledge about the world.

Today, there is an AI system that you must have heard of by now, ChatGPT. It’s making serious waves because it does much more than just pattern recognition and phrase replacement. It’s a lot more complicated than ELIZA. ChatGPT can produce tracts of prose—or poetry—to the point that you or I can sing the praises of a human being without blinking. If it only takes a few lines for Elijah to display his essential bot-ness, ChatGPT is good enough to keep you guessing for a long time. Maybe even forever.

I don’t say this lightly. Already students are using ChatGPT to write their assignments. Scientists use it to write their grant applications; Others use it to improve their academic papers. Programmers have used it to help them write their code. And countless casual users like me have thrown questions—mundane, esoteric or somewhere in between—at it.

You might try “Write a poem about traveling to a narwhal” or “Write an Ernest Hemingway-style essay about sampling fish and chips in England.” I promise you’ll be amazed at the sophistication of ChatGPT’s response, and the ebb and flow of its language.

How can ChatGPT generate all this? Well, this is an example of what is known as a “large language model” (LLM). The “large” is a reference to the vast store of online text on which ChatGPT is trained. So it “knows” how to string words and phrases together, but that’s just the beginning. What makes such LLM powerful is that it can, in fact, learn from and use data. Therefore, you may not get an answer to a question that you direct today. But it is possible that as time goes on, further use of the system will add knowledge to its store and you will eventually get a response.

ChatGPT is about the first software system to operate in this way that can actually have a proper, free-flowing conversation with a human on a variety of topics. What it says may not always be completely relevant, it can sometimes raise your eyebrows. Yet, you would find it hard to believe that everything it produces was not written by human hands, not thought out by human intelligence.

Thus a source of concern. Will ChatGPT and other LLMs eventually write student papers, design scientific experiments, reject or accept book manuscripts? If LLMs make things more convenient, if they can speed up innovations, will they also reduce the overall quality of research? ChatGPT is definitely a genie that’s out of the bottle and won’t go back in. It will shake up academia, publishing, and many other areas of human endeavor as we know them. Shouldn’t we be debating the implications?

Until then, though: here’s a paragraph that chatgpt gave me in response to the Hemingway question above.

“I finished my meal and went back to my abode, thinking of the flavors lingering on my tongue. Fish and chips in England are unlike anything I’ve tasted. The simplicity of the dish, combined with the fresh ingredients, make it a staple of British cuisine.”

Think of all the things ChatGPT has accomplished. Think about how far we’ve come from Eliza.

Dilip D’Souza, once a computer scientist, now lives in Mumbai and writes for his dinner. His Twitter handle is @DeathEndsFun,