With invaluable support of the MIT Archivists, especially Allison Schmitt, we have discovered a truly amazing draft (more precisely, outline) of a paper, hand-signed by Joseph Weizenbaum himself (!), intended to accompany the original ELIZA CACM publication, but which never made it to final form. The amazing thing about this proto-paper is that it lays out in great deatil the next steps that Weizenbaum himself envisioned for ELIZA, including extensive discussion of experiments regarding mis-communications and (what we would now call) conversational repairs, applications of ELIZA in education (which were, indeed, followed up by others at MIT, esp. by Paul Hayward and colleagues), and the mysterious closing sentence: “Build a mathematical model.” 🙂 There are many fascinating aspects of this draft/outline, and we encourage you to read it for yourself to really get into Joseph Weizenbaum’s head. 

20240306: Real Conversations with the Original ELIZA Discovered in the Garfinkel Archive

Anne Warfield Rawls and Clemens Eisenmann curate the archives of Harold Garfinkel, the father of ethnomethodology. They recently provided Team ELIZA with access to the archive where we discovered a whole series of printouts of original conversations between the original ELIZA and subjects in an experiment conducted at Harvard. As far as we know these are the only examples of conversations between real people and ELIZA!  More detail and a link to the conversations can be found on The Original ELIZA page.

20240224: ELIZA on an ASR33

We’re working on a simulation of what is was like to use ELIZA in the mid 1960s, by embedding Antony Hay’s ELIZA in an ASR33 simulator. So far the effort is pretty simplistic, but has basic sanity. Check it out one our Try an ELIZA page.

20240214: Ken Pitman (KMP)’s Lisp ELIZA from MIT-AI’s ITS history project

We’ve added Kent Pitman‘s (aka. KMP, https://github.com/netsettler) MACLisp ELIZA to the elizagen repo. This was discovered by Lars Brinkhoff, and can also be found on his PDP-10 ITS historical sources repo.

KMP has been a fixture of the Lisp community for decades. Among many other lisp-related activities, he worked on Macsyma, Maclisp, Emacs, and authored the Common Lisp HyperSpec. He also chaired several of the Common Lisp standards subcommittees. He has written several interesting essays on the history of lisp and lisp-adjacent artifacts and activities, available here: https://nhplace.com/kent/Papers/.

In the late 1970s, when he started at MIT, KMP took Joseph Weizenbaum’s Intro to Programming (6.030), where he was introduced to ELIZA. Probably having seen Bernie Cosell’s BBN-Lisp ELIZA, KMP wrote a version of his own on ITS, a locally-developed OS on PDP-10s, that were the main computers in the MIT-AI lab. ITS had a slightly different version of Lisp (MACLisp vs. BBNLisp), so Cosell’s code couldn’t be run directly, also KMP wanted to hack the program a bit. He writes: 

“While [this code has] definitely got my work in it, it’s pretty clearly built on someone else’s work. Back then there was not a sense of publication so much as just making existing things better. And everything was a proto variant of free software because there was no place one realistically expected one would sell anything. It was just hack upon hack in that environment. Both intellectual property law itself, and my naive understanding of it, was quite different then.”

KMP provided some additional thoughts on his ELIZA:

“The thing about the DOCTOR program, while it was quite elaborate, was that it sort of missed the point of what Weizenbaum was trying to say. It’s not that I didn’t understand him, but it took me a long time to really come to viscerally understand the seriousness of his point. He said that (paraphrasing, maybe badly, from memory) long before we got to real AI, we would get to things that seemed smart but weren’t, and we would overly trust systems that were not really smart. He was, for example, very worried about computerized launch of nuclear missiles using what was then thought to be AI. But the point of the original ELIZA was not to BE elaborate, it was to be simple, and to show how even a tiny, tiny program could seem smart. It was hard not to think it could be smarter if it were not so tiny, but that misses the point, which was about how easily we were confused, not about how much more effectively we could be confused if more energy were put in. So whether my DOCTOR or any other was making the right point was hard to say. My guess is not.”

20231203: ELIZA and Ethnomethodology (specifically Garfinkel) 

20230703: Javascript ELIZA by Michael Wallace and George Dunlop added to repo

  • Paul  McQuesten (mcquesten@gmail.com) pointed out a Javascript ELIZA, originally written by Michael Wallace and enhanced by George Dunlop. I’ve added it to the repo, and at this writing, a live version is online here.

20220627: ELIZA Game by Zachtronics

I came across this interesting game based on ELIZA, by Zachtronics, a semi-defunct boutique game company: “Eliza is a visual novel about an AI counseling program, the people who develop it, and the people who use it.”

20220519: 1975 LISP ELIZA (Contrib. Lori Levin, author unkown)

Lori Levin has contributed a 1975 Lisp ELIZA from The University of Pennsylvania Moore School of Electrical Engineering. Interestingly, this ELIZA has the script programmed into the code, as opposed to most earlier ELIZAs that read a script. (Ed. note: Lori notes that she didn’t write this code. We aren’t sure who did, but this is very likely the ELIZA that I remember seeing as a Penn undergrad, as I knew Lori then, but this was well after I’d written my BASIC version, in 1973. — Jeff) 

20210329: As Close an Eliza as One Can Get (Anthony Hay)

Anthony Hay <anthony.hay.1@gmail.com> writes:

This is an implementation in C++ of ELIZA that attempts to be as close to the original as possible. It was made to closely follow Joseph Weizenbaum’s description of his program given in his January 1966 Communications of the ACM paper. It is controlled by a script identical to the one given in the appendix to that paper. When given the prompts that begin “Men are all alike” it gives back the identical responses and recreates the exact conversation from the paper.

I was interested to read that Prof. Millican thinks Weizenbaum “cheated” in order to obtain the conversation shown in his paper. The response in question was “DOES THAT HAVE ANYTHING TO DO WITH THE FACT THAT YOUR…,” which is the last entry in the MEMORY section of the script, rather than the first. But Weizenbaum’s paper says that “…one of the transformations on the MEMORY list is randomly selected…,” so I don’t think it unreasonable to think that the last entry could have been randomly selected on this occasion.

https://github.com/anthay/ELIZA

20200809: Elizabeth an Eliza with an easy-to-understand script notation (Peter Millican)

Peter Millican <peter.millican@hertford.ox.ac.uk> writes:

I’m a Philosophy professor at Oxford, who’s spent most of my working life on the boundary with Computer Science and trying to encourage students into that space.  Trying to get Humanities students interested through LISP programs is hopeless, of course, so when teaching Computing and Philosophy at Leeds, I developed a chatbot designer system called Elizabeth which is programmable using an easy-to-understand script notation.  You can find the latest version at http://www.philocomp.net/ai/elizabeth.htm, and learning materials at http://www.philocomp.net/ai/elizlearn.htm.  One of the built-in Illustrative Scripts is transcribed from Weizenbaum’s original paper of 1966.  If you look at the second video on http://www.philocomp.net/videos/turing.htm, from 18:30, you can see me going through the dialogue in an online talk (part of whose point was to prepare the students for using Elizabeth in an exercise by seeing the sorts of tricks that can be done).  As far as I know, this is the only system that allows non-specialist users to work through the operation of ELIZA, seeing a trace of all its internal processing, and I believe that it faithfully copies the behaviour of the original program as published, with just one exception …

 One interesting point that I mention in my talk, but have never seen mentioned by anyone else, is that Weizenbaum’s published dialogue doesn’t actually fit with his published program: one of the responses is out of order (though this can easily be fixed by swapping two of them round, as I do in my script transcription).  I take this as evidence that Weizenbaum “cheated” to make his dialogue appear as impressive as possible, and then probably forgot or didn’t bother to change the program accordingly.  Have you ever seen a mention of this point?  Is it actually well-known?

Peter Millican

Gilbert Ryle Fellow and Professor of Philosophy

Hertford College, University of Oxford

Faculties of Computer Science and Philosophy

A bit more about the problem that Peter discovered in the scripts (from a subsequent email exchange): 

I talk about the “cheat” from around 24:25 to 26:00 in the video.  

In the script file “Weizenbaum.txt” the matter is annotated like this:

/ ***** No-keyword responses involving “MY” memories *****

/ Order changed to conform with published dialogue:

/ 15. “Bullies. DOES THAT HAVE ANYTHING TO DO WITH THE FACT THAT YOUR BOYFRIEND MADE YOU COME HERE”

N DOES THAT HAVE ANYTHING TO DO WITH THE FACT THAT YOUR [M+1]?

  &{M+1}

N LET’S DISCUSS FURTHER WHY YOUR [M+1]

  &{M+1}

N EARLIER YOU SAID YOUR [M+1]

  &{M+1}

N BUT YOUR [M+1]?

  &{M+1}

 Here “N” signifies a “no-keyword” response, while [M+1] picks up the first stored memory (ELIZA only stores “my …” memories, so there’s no need for more sophisticated memory management here).  “&{M+1}” then specifies an action to be performed when this response has been triggered – it deletes the first stored memory (which, of course, has just at that point been used).

For comparison, the relevant snippet from Weizenbaum’s SLIP program in his article is like this:

(MEMORY MY 

  (0 YOUR 0 = LETS DISCUSS FURTHER WHY YOUR 3)

  (0 YOUR 0 = EARLIER YOU SAID YOUR 3)

  (0 YOUR 0 = BUT YOUR 3)

  (0 YOUR 0 = DOES THAT HAVE ANYTHING TO DO WITH THE FACT THAT YOUR 3))

Notice that the “DOES THAT HAVE ANYTHING TO DO…” response is last of the four in this group, whereas for the dialogue to work, it needs to be first.

In every other case, my script follows the relevant orderings from Weizenbaum’s article, so this is very much a special case!

20190215: Eliza in gnuCOBOL (Arnold J. Trembley)

This amazing Eliza in gnuCOBOL was written by Arnold J. Trembley. It’s a slightly indirect translation of Shrager’s 1973/1977 BASIC version. The most amazing part of this (to me) is that it was written in 2017!

Mr. Trembley writes:

I also have a link on my website for an executable download for Eliza in GnuCOBOL that runs on Windows 7 and higher. I think it would probably run on Windows XP too.

I probably first saw ELIZA in BASIC from Ahl’s computer games for BASIC book, but I also wrote a version in 1974 ANSI-standard COBOL for an IBM Series/1 Minicomputer way back around 1984. But I never saved that source code, so I had to start all over when I converted it into GnuCOBOL.

Getting Eliza to work was a lot of fun.

20190121: Eliza on ITS on a DEC 10 Emulator (Lars Brinkhoff)

A few days ago I had the pleasure of taking a trip in a real live time machine.

About a week ago I got an out-of-the-blue message from Lars Brinkhoff via the ElizaGen issue tracker: “I have a version of ELIZA ported to Maclisp, dated 1977-09-05. Unfortunately, it’s just a compiled binary file. Is this of any interest to you?”

Just a binary wouldn’t be too interesting, so my initial reaction was muted: “Is there any way to either run it, or extract the embedded code?”

Turns out that Lars had significantly understated the situation: “The binary file is a compiled FASL, fast load file, so it’s not easy to extract any code other than PDP-10 instructions. It might be possible, but a lot of work, to reconstruct Lisp code that compiles to approximately the same FASL code.”

However, went on: “…we can run it in Maclisp running on a PDP-10 emulator. I have an emulator online.”

Eliza on MacLisp on a PDP-10 emulator? Isn’t there a level missing here? On what OS?

The answer gave me pause: MIT’s ITS. ITS is one of the most important operating systems in CS history. (Among many other interesting historical facts, the original EMACS was developed on ITS!) I recommend reading about it here: ITS on Wikipedia. When I was an undergrad and grad student, in the late ’70s and early ’80s, we used MIT-AI’s ITS remotely through the early ARPANet. Interestingly, at one point (I don’t know for how long) the ITS at MIT-AI had an Eliza hacked into its command line driver so that unless you prefixed your commands with a “:”, the response was from Eliza!

Eliza on MacLisp on ITS on a Dec 10 emulator? Really?!

Yeah, really!!

Here’s the emulator: https://github.com/PDP-10/its

(Lars notes: “…there are more people involved than me. Foremost Eric Swenson who worked on Maclisp and Macsyma back in the day. We […] update the emulators to support more hardware, fixing bugs, figuring out how to build and run programs, adding new software, Etc.”)

If you want to play with a great piece of software history, you can log into this ITS emulator yourself; It’s free and open. No password required!

From Lars:

Telnet to “its.pdp10.se, port 10003”. Type Control-Z to log in. Then “:lisp games; eliza (init)” to start Maclisp and load ELIZA. It should look something like this:

==================

You: 

        telnet its.pdp10.se 10003

Trying 88.99.191.74…

Connected to pdp10.se.

Escape character is ‘^]’.

Connected to the KA-10 simulator MTY device, line 4

You:

     type Control-Z

TT ITS.1648. DDT.1546.

TTY 25

2. Lusers, Fair Share = 0%

THIS IS TT ITS, A HIGHTY [sic] VOLATILE SYSTEM FOR TESTING

For brief information, type ?

For a list of colon commands, type 😕 and press Enter.

For the full info system, type :INFO and Enter.

The management apologizes for the use of the slow TK10 terminal

multiplexer.  We’ll upgrade to the much faster Datapoint kludge once

it becomes available.

Then you type:

     :lisp games; eliza (init)

(Please Log In)

QUIT

SPEAK UP!

 Et voila. you’re typing back through 40+ years of computer history!

 (Use double ENTERs to terminate your sentence to Eliza.)

==================

To test it, I pasted the “standard” example conversation from Wiezenbaum’s Eliza paper. Here’s a screenshot from the paper:

And here it is, running live on Lars’s reborn ITS (the mess at the bottom is because I accidentally pasted the sentence twice, and had to backspace over it. The repeated chars are apparently how ITS displays backspacing):

There’s a bit of a mystery about where exactly this version came form. Lars and I theorize that since BBN was right across from the AI lab, and it’s known that BBN were users of the ITS systems, this may well be Bernie Cosell’s original code!

20181031: Stefan Holtgen and Marianna Baranovska have just published a volume entitled: “Hello, I’m Eliza”, with Computer Archaeologie.

Here is the abstract (rendered from the German by Google xlate):

The contributions to this volume are partly based on a conference held in October 2016 at the Institute of Musicology and Media Studies, and bring together authors from the fields of computer science, philosophy, media studies and media dramaturgy. In the individual chapters, the origins of the first chatbot ELIZA, which Joseph Weizenbaum developed at MIT’s AI Lab in 1966, are reconstructed, Weizenbaum’s programming language SLIP is presented and a first German-language translation of his publication to ELIZA is presented. As an extension of the topic, the authors ask what influence ELIZA had on computer culture, AI research, cultural criticism of the computer and art.

And here’s the interesting cover. (Note that the code in the background is a part of Shrager’s 1973/1977 BASIC Eliza)

20171211: BASIC Eliza via Speech Synthesis (Michael Wessel) 

20171211: My friend and colleague, Michael Wessel, who apparently has a soft spot for ancient speech synthesizers, recently noted this curious coincidence/historical anecdote about a German version of my BASIC Eliza. (Michael and I worked together on the AURA project at SRI for about a year around 2013.)

Michael writes:

“About 1985 I typed in a version of your BASIC Eliza from a public library copy of the ‘Creative Computing Computer Games’ book, into my Amstrad CPC 464 home computer. The book had been translated into German, and so was Eliza! I wish the editors hadn’t done so, as it makes much less sense!

Back then I owned a speech synthesizer (the Amstrad SSA-1 for the CPC), and I had extended your BASIC Eliza with speech capabilities. The audio/speech quality was as awful as some of the TRS 80 speech synths you can hear on YouTube.

Now, 32 years later or so I am working on a little spare time hobby project – a modern, understandable speech synthesizer for the Amstrad CPC, which is designed to be compatible with that original “retro” Amstrad SSA-1 speech synthesizer. And indeed, the old Eliza that got extended with SSA-1 speech output is working well with my modern speech synthesizer.”

Here’s a youtube video of Michael’s German-speaking Eliza.

And here is another version for the CPC, also with speech output.

Michael continues:

“With respect to the second video (the English one), the original author of this Eliza version also mentions your Eliza as the source for the program, but he says he extended it significantly, and then I added speech output, this time using the ‘native mode’ of my speech synthesizer, which gives much better speech quality than the emulated ‘retro SSA-1’ compatibility mode, as you can hear.”

As I mentioned above, Michael and I worked closely together at SRI for about a year. He concluded his note this way: “It’s really a funny and strange coincidence that our ways have crossed; What are the chances that you get to work with someone whose BASIC program you typed in from a public library book 30 years ago? Sometimes, the world is small!”

I couldn’t agree more!

Here’s the disk image of Michael’s German BASIC Eliza:

https://cpcrulez.fr/applications_list_util-elisa.htm

To run it:

  • Download extract the DSK image

  • Go to: http://cpcbox.com, which is a CPC Emulator in Javascript running in the Browser

  • Select “Choose configuration -> 464”

  • Select the Eliza DSK file for “DRIVE A:”

  • When the CPC boots up, enter “cat”; You should see “ELISA-NS.BAS”

  • Enter: RUN “ELISA-NS (Without Speech (“NS = no speech”)

      (The other version is with speech, but doesn’t work on the emulator)

Here are some PS2 Screenshots from emulator running in Firefox:

20170728: Through the efforts of Dave Walden and Bernie Cosell, the tapes and listings of Bernie’s original Lisp version of Eliza/Doctor have been accepted into the Computer History Museum archives as lot number: X8258.2017.

20170307: Jeffrey Perrone pointed out this hysterical video of Eliza talking to Eliza on two side-by-side Google Home systems!

20151122: John D. writes (email to Jeff Shrager of 2015-11-22): “Back in 1979, Radio Shack released their own commercial version of Eliza for their TRS-80 Model I, sold in their stores. It was written in machine language, not Basic, on cassette tape. In fact, the tape had two versions: a regular one, and a special version that gave Eliza’s response through the expensive Voice Synthesizer they were selling (a Votrax derivative).”

“The interesting thing about Radio Shack’s version is that it seemed a bit more sophisticated than the original Weizenbaum version. It recognized more words. (Including swear words–just so it could react with ‘There’s no need for that language.’– which is weird to see inside a mass-market product sold by a major corporation!) And when you said “Goodbye” it gave you a diagnosis: everything from “You have no psychological problems besides a normal amount of paranoia” to ‘I must conclude you are dangerously psychotic and should be locked up’!”

“It was apparently written by Device Oriented Games in Dallas (near where Radio Shack was based).”

Links to runnable code:

John also found this YouTube video of a real TRS-80 with the Voice Synthesizer, running Talking Eliza. The Votrax quality is terrible, but that makes it all the more fun!

20151013: Paco Willers has written a GO version of the classic BASIC Eliza. In the process he did a detailed, nearly line-by-line, deconstruction of the Shrager/North 1973/1977 BASIC code.

20150910: Melody Ayres-Griffiths (melody@planetirata.com) found this completely wacko version written by Dave Babcock in Integer BASIC, which uses computed gotos! It looks to my eye like someone took the Shrager/North version and stuck it in a blender. 🙂 It would be easier to generate this from a compiler, than to write it out straight!

20150105: James Markevitch has updated his 1966 version with a slightly cleaned up transcript, and his amazing PERL BBN Lisp Interepreter! I’ve uploaded everything to the github site, and it is all also available for download as a tarball here.