Reality Ends Here, Part III
You are in a maze of twisty passages, all alike.
—Colossal Cave Aventure
Zork, the 1977 text-based game by Infocom, was more than just a text-based adventure; it was a cybernetic modeling tool capable of mining natural language and cognitive patterns. DARPA-funded pseudo-alchemists, under the guise of “computer scientists”, hinted at this through their advertising. While the game's exoteric marketing suggested that the player's imagination drove the simulation, this was true in more than way than one. "Zork" literally means "unfinished product", the final component was the user's brain.
The game's inspiration traces back to ADVENT / Colossal Cave Adventure by Will Crowther, developed on a PDP-10 at MIT. Crowther's association with Bolt Beranek and Newman (BBN), a company contracted by ARPA/DARPA and the Navy, is notable. The same year Colossal Cave Adventure was released, Crowther's colleagues at BBN published "A Tutoring and Student Modeling Paradigm For Gaming Environments" funded under an Air Force Human Research Laboratory (AFHRL), and ARPA Human Resources Research Organization (HumRRO) contract.
Richard Burton & J.S. Brown’s computer-aided instruction (CAI) research would lead to “a fully operational AI-CAI system (accessible over the ARPANET)” called SOPHIE:
“In order to study how easily users could adapt to the imposed linguistic limitations, we have collected well over a hundred hours of protocols of people using SOPHIE over the ARPA network … Any time SOPHIE encountered a sentence which it could not parse, that sentence was automatically stored on a file which was continually used to provide data for expanding our grammar.”
The approach described above was described as “learning by teaching” in the words of the cybernetician J.C.R. Licklider. In a 1979 National Science Foundation planning report—in which J.S. Brown would take part—Licklider notes the efforts of the director of the Dynamic Modeling Group and Infocom board member, Albert Vezza, who had “been exploring how to design computers to make them amenable to being ‘taught.’” Infocom was founded in the same year. Licklider, very briefly, just says out loud what they are doing:
The aim would be to convert the knowledge of each field from its initial form-- text, equation, or diagram, plus whatever the form is inside the heads of people-- to the new form selected to facilitate representation in computer memory and processing by computer.
It is very easy to let this statement wash over you without processing the esoteric connotations of such language. Arpanet acted as a distributed laboratory, experimenting on and data mining “net randoms,” people that would wander into the network, as well as relevant “domain experts”. It is a surveillance enclosure where human subjects can be observed and recorded, amalgamated, and “dynamically modeled”, revealing whatever is “inside the heads of people”. Mining the human soul— the secrets of pattern recognition, intuition, and even extrasensory perception.
Burton & Brown would add in their paper “[o]ne particular use currently under consideration involves using SOPHIE in a gaming situation.” BBN's overarching mission was to develop "speech understanding systems" for ARPA and the Navy to monitor communications, for the ultimate goal of “distributed sensor nets”. The challenge was deciphering "ill-formed inputs" like garbled text or unusual commands, much like the challenges in Morse code recognition.
In the 1960s, in Memorandum: An Air Force Role in Counter-Insurgency Warfare (1962) J.C.R. Licklider had posed the need to invent “distributed” communications systems. These systems were never meant solely for Vietnam, he considered military simulation systems necessary for the development of civilian ones (Nemenyi 2019). The term “dynamic modeling” is a smuggling of Norbert Wiener’s more esoteric writings on the “operative image,” or “a working simulacrum of a living organism.” Licklider was an active member of Wiener’s Cambridge, MA cybernetics group that convened every Tuesday night and he took part in their dinner seminars, describing himself as a “faithful adherent” (Hauben 1996). He later took part in the CIA-funded Macy Seventh Cybernetics Conference, where he presented on “psychoacoustics”.
Licklider revealed the true essence of dynamic modeling, prophesying that graphics would become integral to thinking and "command and control." He envisioned a future where computers would facilitate more effective communication than face-to-face interactions, “dynamic hieroglyphs” beamed onto the retina of the eyes, exceeding the fundamental limits of human cognition, systems reaching for seamless machine-human communication (Licklider 1976). For his dream machine, Licklider chose the biological concept of symbiosis:
“The fig tree is pollinated only by the insect Blastophaga grossorun. The larva of the insect lives in the ovary of the fig tree, and there it gets its food. The tree and the insect are thus heavily interdependent: the tree cannot reproduce wit bout the insect; the insect cannot eat without the tree; together, they constitute not only a viable but a productive and thriving partnership. This cooperative "living together in intimate association, or even close union, of two dissimilar organisms" is called symbiosis” Man-Computer Symbiosis J. C. R. Licklider
Zork's development was influenced by the research of the Dynamic Modeling Group at MIT, led by pioneers like J.C.R. Licklider, Dave Lebling, Marc Blank, and Albert Vezza. Zork was first crafted in Muddle, a language designed for the “Dynamic Modeling System”, emphasizing user-oriented interactive computing, dynamic graphics, heuristic programming (“learn by teaching”), and always online iterative development.
While Zork was being created, and a year before Infocom’s founding, Albert Vezza, Timothy Anderson, and Marc Blank presented a paper on language understanding and AI at an NSA-DARPA workshop called “Distributed Sensor Nets”. They discussed creating an "Augmented Transition Network"1 to decode Morse code and ambiguous inputs. To construct this grammar, a network environment filled with realistic chatter and errors was essential:
"Two sources of test data were used. We obtained approximately 30 thousand characters worth of data from amateur operators and others who could send and receive Morse code. This data was hand-keyed in our laboratory. Some of this data was obtained using actual transmitters connected to our cable network. This data was composed primarily of plain text with a small admixture of network chatter." "Machine Recognition and Understanding of Manual Morse" [Page: 137]
Augmented transition networks have been used for many purposes, from game design (Cavezza 2005) to multimodal brain-computer interfaces (Kirchner 2019) to ONR-funded computer-aided learning programs (Mayer 1987). Drawing parallels with "Machine Recognition and Understanding of Manual Morse," Zork's groundbreaking ability to understand and respond to “natural language” inputs parallels the challenges faced in decoding garbled Morse transmissions. Just as Morse code recognition systems had to account for human-induced inconsistencies, Zork's parser was able to decipher a myriad of player inputs, each with its unique phrasing and intent.
Zork on the ARPANET essentially served as a "distributed sensor" to gather ambiguous inputs from “net randoms”. Directionally, the development process of the Morse research was identical to Infocom’s seminal game. Dave Lebling co-writer of Zork (aka Dungeon) and co-founder of Infocom characterizes the interplay thus:
I had also worked with Marc Blank and Tim Anderson on a Trivia game which had a database of thousands of questions submitted from all over the old Arpanet. All of this stuff was only peripherally related to my research, which had to do with subjects such as Email, Morse Code decoding, Office Automation and programming tools. A lot of the game stuff was strictly after-hours fun, although it did help improve the language we used (Muddle, aka MDL), and was a test-bed for application packages that were used in the serious work. There was some interest in natural language parsing and understanding in our Morse Code project, and that may have had an effect on Zork as well. (Dave Lebling)
However, the statements “all of this stuff was only peripherally related to my research” and “the game stuff was strictly after-hours fun” are contradicted in the above quote and Infocom official history from the “New Zork Times”:
The next game of note was Trivia (who says research labs aren’t ahead of their time?), an ongoing “can you top this” contest for the truly crazed. Trivia, unlike Maze, could be played by network users, and achieved wide popularity on the ARPAnet. Marc Blank wrote the second version, and I maintained/hacked it; it was actually a legitimate test of a database system the group used for a research project.
Indeed, "Zork: A computerized fantasy simulation game," (1979) is cited in “Progress Report 16 July 1978 - June 1979“ for the Office of Naval Research and DARPA. In the same paper, MAGE is discussed: