The Accidental AI: How ELIZA's Lisp Adaptation Derailed Its Original Research Intent

Table of Links

Abstract and 1. Introduction

2. Why ELIZA?
3. The Intelligence Engineers
4. Newell, Shaw, and Simon’s IPL Logic Theorist: The First True AIs
5. From IPL to SLIP and Lisp
6. A Critical Tangent into Gomoku
7. Interpretation is the Core of Intelligence
8. The Threads Come Together: Interpretation, Language, Lists, Graphs, and Recursion
9. Finally ELIZA: A Platform, Not a Chat Bot!
10. A Perfect Irony: A Lisp ELIZA Escapes and is Misinterpreted by the AI Community
11. Another Wave: A BASIC ELIZA turns the PC Generation on to AI
12. Conclusion: A certain danger lurks there
13. Acknowledgements and References

10 A Perfect Irony: A Lisp ELIZA Escapes and is Misinterpreted by the AI Community

Our thesis is that Joseph Weizenbaum, afraid of people misinterpreting computers and AI, developed ELIZA not as an AI, but as a platform to support research about people’s interpretive process, perhaps more specifically their interaction with AIs. Weizenbaum, of course, had to provide an example – the DOCTOR script – to demonstrate the platform, but, as above, he had no illusion that it was actually intelligent, nor even very easily interpreted as intelligent; Much more “intelligent” programs already existing in various game playing programs (even his own, from the Datamation paper!), in Colby’s Parry, and in other programs that solved math word problems[17], did database lookups from natural language queries[29], and others. Although some of the sort of research envisioned by Weizenbaum was carried out by the MGH team, a specific event thwarted this use case for ELIZA, and, in almost the deepest possible irony, led to exactly the conflation of ELIZA with AI that Weizenbaum set out to study, not to bring into being.

Weizenbaum built his ELIZA in MAD-SLIP on the IBM 7090, which was the primary machine at MIT’s Project MAC, on the 5th-through-9th floors of Tech Square. Almost immediately after ELIZA’s publication in 1966, Bernie Cosell created a Lisp knock-off of ELIZA, based on the algorithm and DOCTOR script in Weizenbaum’s paper. Cosell worked at Bolt Beranek and Newman (BBN), a large RAND-like consulting company situated nearby MIT, that commonly hired MIT graduates. Coincidentally, at almost the same time, BBN was a founding site of the ARPAnet[14], and Cosell’s Lisp ELIZA diffused rapidly through that network and across the soon-to-be Lisp-centered world of academic AI. As a result, Cosell’s Lisp ELIZA rapidly became the dominant strain, and Weizenbaum’s MAD-SLIP version, inaccessible on the ARPAnet, was almost instantly forgotten.

Through interesting and complex criss-crossing histories, described in detail in [45], Danny Bobrow, a recent MIT AI graduate who headed up BBN’s AI program, brought McCarthy, and thus Lisp, to BBN, where Cosell was exposed to it, and built an ELIZA knock-off in Lisp as a side-project.[15] Cosell reports: “When I was working on the PDP-1 time-sharing system [...] I thought I would learn Lisp. That spring, Joe Weizenbaum had written an article for Communications of the ACM on ELIZA. I thought that was way cool. [...] He described how ELIZA works and I said, “I bet I could write something to do that.” And so I started writing a Lisp program on [the] PDP-1 system at BBN.”[41, p. 540] He continues, “I wrote that program and got it up and working. Playing with it was an all-BBN project. [...] It was written, at first, in the PDP-1 Lisp. But they were building a Lisp on the PDP-6 at that point—or maybe the PDP-10. But it was the Lisp that had spread across the ARPANet. So [ELIZA] went along with it [...].”[41, p. 541]

Once Cosell’s Lisp ELIZA hit the academic world via the rapid spread of the ARPANet, Weizenbaum’s MAD-SLIP version was no longer relevant, and the name “ELIZA” (and the “DOCTOR” script), as well as the concept, was, from that point forward, associated with Cosell’s Lisp version, although its origin was still correctly attributed to Weizenbaum via the CACM paper, leading to a 50- year-long, community-wide misapprehension that ELIZA had been written in Lisp. In addition to being promulgated by Cosell’s knock-off being the one that was most easily available, via the ARPANet, it was a natural confusion because Lisp was rapidly becoming the go-to language of AI. Once Lisp came onto the scene no one thought much again about SLIP, or, for that matter, IPL-V.

[14] At the time what became the internet, and then the web, was called the ARPAnet, and BBN built its core hardware and developed the programs that ran it.[11])

[15] Coincidentally, Bobrow was also the author of STUDENT, the program mentioned above that solved math word problems.[17]