• 531 words

Just to get it out of my system, here is yet another idea, up for grabs for anybody who has more generous time budget available than I do (you lucky bastard, whoever you are...).

In the early 1970es, when the internet was still the arpanet, someone connected two chatbots over the network. The protocol of the conversation was captured and published as RFC 439. It is a hilarious read.

One of the programs is quite well-known and studied. It was an incarnation of Joseph Weizenbaum's ELIZA from MIT, acting as a psychotherapist of the Rogerian school. This program (or at least conceptually a very similar one) is still available on contemporary systems as Emacs Doctor. One of my CS professors derived a toy version of it in a handful lines of Prolog during a 90 minute lecture, which might tell you that it is not too complicated (an important intellectual milestone of computing nonetheless).

The other part to the conversation is the lesser known PARRY. It was created by Kenneth Colby at the Stanford Artificial Intelligence Laboratory. PARRY simulates the conversation patterns of a paranoid schizophrenic person. It is the more sophisticated program of the two and you cannot run it easily today, because it seems to have not been ported to any other machine than the original PDP-10 it ran on in 1972. Martin Frost has published what was left of the source code in 1991. The source is available at this server hosted by CMU. The readme says:

The code here compiles and runs using the MLISP language (meta-lisp) on the WAITS operating system running on a DEC PDP-10. Don't expect to be able to run it yourself, because you probably don't have the original environment. The code is definitely non-portable. Parts of it are written in PDP-10 assembly code (e.g., *.fai and *.lap files). Other parts are written in MLISP (e.g., pmem* files). There may be other parts that require other language translators.

So, that was 1991. 35 years later, you should not actually need physical access to a PDP-10. There are emulators/simulators like OpenSIMH available. I don't know what it would take to get the WAITS operating system running on a simulated machine. The SAILDART Archive has (some?) of its source available here.

When you manage to get the operating system running, does it include a meta-lisp environment? It will certainly not be a modern day Linux distribution, with batteries charged and all. Even if it had one, would it be the correct version? And when there were a compatible MLISP, how much effort will it be to run PARRY? Would it even start if the source were incomplete? It is a lisp program, so it might get up, but you'd probably run into runtime errors. So, one would need the time and appetite to wade through tons of obscure documentation (if there is documentation at all) and fill in the gaps, of which I'd suspect there will be many, with lots of tinkering.

I find it a bit sad that the early history of computing is so prone to be lost to the times. Much more than any LLM based conversation, I would love to see PARRY talking to the doctor once again.