Did Eliza do little?

By Brian Clegg

When I first worked professionally with computers, programming Dec-10s in the 1970s, we had two particularly popular pieces of non-business software. One was a multiplayer Star Trek game, and the other was a version of ELIZA. This was a proto-chatbot, dating back to the mid-1960s.

Although the code of ELIZA was relatively simple and restricted, in part because of the open question approach of Rogerian psychoanalysis, it could do what at the time seemed incredibly impressive attempt at human-style interaction. So novel was its ability that it was not uncommon for users to think they were dealing with a real person (demonstrating how useless the Turing test was).

There’s now a book available to preorder where multiple academics look at aspects of the original ELIZA software, portraying it as an ancestor of modern LLM chatbots like ChatGPT and Grok. I’m not reviewing the book as, to be honest, I find such multi-authored academic books where each writes a separate chapter desperately dull to read, and often extremely repetitive. Also apparently it provides ‘comprehensive critical analysis of Joseph Weizenbaum’s groundbreaking chatbot system through the lens of critical code studies’, and I have a deep-rooted aversion to critical anything studies. But I have flicked through the ebook version.

One thing I wasn’t aware of, as I’d only ever seen the ‘doctor’ incarnation of ELIZA, was that there were different versions attempting to do a variety of things. For example different scripts could attempt to discuss the physics of an elevator, or New England states – though to be honest, you can see why it was the psychotherapist DOCTOR script that got all the attention.

Inevitably, given the currency of AI and chatbots it’s tempting to put ELIZA as a direct ancestor of this type of software – but I would strongly reject this. For me, ELIZA suffers the same problem that occurs when people attempt to label Babbage’s difference and analytical engines direct ancestors of modern computers, making him the ‘father of computing’. 

The mechanical calculator and computer that Babbage designed, but was unable to construct, were only a side branch of the family tree that had no children and came to a stop while the main branch continued in a different direction. Babbage’s engines did not provide a starting point for future computing developments – those started anew from a different source.

The same can surely be said about ELIZA and modern chatbots. ELIZA had a totally different architecture to the LLMs, using a series of canned answers that would be selected (with some randomness incorporated) based on a small number of recognised words in the input. Just like the computing comparison with Babbage, LLMs are attempting to solve the same problem as ELIZA, but do so in a totally different way, with no similarity in approach.

This doesn’t stop ELIZA being an interesting relic – a dinosaur, if you like, compared to the current mammalian line. Those responsible for the book have put together a demonstrator of what the ELIZA experience was like (including green on black text as we had on the Dec-10, though I first used it on a teletype). You can see a snippet from this above, and you too can have a psychoanalysis session with ELIZA on their website.

Image from https://inventingeliza.com - conversation by the author

You can preorder Inventing Eliza from Amazon.co.uk, Amazon.com or Bookshop.org

Using these links earns us commission at no cost to you

These articles will always be free – but if you’d like to support my online work, consider buying a virtual coffee or taking out a membership: 

See all Brian's online articles or subscribe to a weekly email free here

Now Appearing is the blog of science writer Brian Clegg (www.brianclegg.net), author of Inflight Science, Before the Big Bang and The God Effect.