The mid-1980s were the first golden age for the user-interface community. The first Human Factors in Computer Systems Conference had just been completed, launching the CHI conference series. The Media Lab was being built at MIT. Groundbreaking research and books were emerging from Xerox PARC, IBM Watson, and top universities on both sides of the Atlantic. To top it off, the Macintosh would win the 1984 Super Bowl, announcing to the world that computing was coming to ordinary folks.
In 1983, 20 American companies joined to form the Microelectronics and Computer Technology Corporation (MCC), a research consortium in Austin, Texas, headed by retired Admiral Bobby Ray Inman. Their motive was a fear that they lacked the resources to compete with foreign governmentfunded computing-research initiatives, most specifically the Fifth Generation Computer program in Japan. MCC was the first in what would become a series of pre-competitive industrial-research consortiums that were subsequently protected from anti-trust penalties by an act of Congress.
MCC launched four research areas, including semi-conductor manufacturing technology, software design technology, VLSI design technology, and the Advanced Computer Technology (ACT) program, which focused on creating a prototype Fifth Generation Computer. ACT included programs in artificial intelligence, parallel processing, advanced database technology, and human interface.
MCC pursued Gordon Bell, the senior guru at DEC, to head ACT and coordinate the four programs in producing the American entry into the Fifth Generation Computer race. When neither Bell nor any comparable luminary could be attracted to head ACT, MCC allowed the four programs, each under the management of a different vice president, to conduct their research independently of the other programs. I was recruited to be the founding technical director of MCC's Human Interface Laboratory.
Without a powerful visionary to force the four ACT programs to coordinate their research toward a common objective, the programs each pursued their own agendas, ensuring that the semblance of a Fifth Generation Computer would never appear. By unconscious default, the ACT strategy was to advance the component Fifth Generation technologies in the ACT programs and let the member companies assemble the output in whatever products they wished to market. After several years ACT created a position to coordinate the research of the programs. However, the power lay with the program vice presidents rather than the coordinator, and integration of ACT technologies was spotty. The HI Lab was originally placed under the direction of a vice president from one of MCC's member companies with no experience in user interface technology or research, assuming his business focus would accelerate technology transfer.
The founding documents for the HI Lab described ambiguous objectives, mentioning speech and image recognition and natural language interaction, but never specifying what was intended as Fifth Generation user interface. Consequently, in launching the HI Lab, we attempted to attract leading figures in speech recognition, image recognition, and natural language processing. We were, however, unable to entice elite researchers away from their established and well-funded laboratories at top universities and corporate research centers. Although we did attract some talented researchers in intelligent tutoring, graphical interaction, and natural language, it was clear from the start that we were well behind the top laboratories and would not likely close the gap during the span allowed in MCC's aggressive goals.
We crafted our original research objective to accelerate the pace of developing user interfaces that integrated knowledge-based capabilities (natural language, intelligent advising, interface design assistance, etc.) with mediabased capabilities (graphic metaphors, interactive work surfaces, handwriting recognition, etc.). We realized we would add little value to the field by competing on component technologies with laboratories that were well ahead of us. However, the challenge of creating a system that could integrate these capabilities into an effective user interface was an unsolved problem. Our strategy became to advance the software that integrated intelligent and multi-media interaction, allowing wide flexibility in choosing techniques for input or output.
Since intelligent computing was at the core of the ACT program, MCC ordered truckloads of Symbolics machines (they can now be viewed in archaeology departments), and Lisp began pouring forth. The decision to commit to Lisp in most programs prioritized the speed of prototyping advanced concepts over the realities of technology transfer. These were the priorities set by the member companies who agreed to set up technology-transfer organizations to translate MCC's prototypes into their favored languages and platforms.
We organized the HI Lab around four research areaslanguage-based dialogues, intelligent user assistance, graphic-based dialogues, and multi-media integration. Although we made early attempts to coordinate and integrate the research of the four groups into a cohesive interface program, the groups argued to work independently to master and advance their separate technologies. An executive decision to allow the individual research teams to pursue their separate research agendas reduced the need for a technical director, so in 1985 I left the HI Lab to help Laszlo Belady launch MCC's Software Technology Program.
From 1984 to 1986, the four research groups worked toward their own prototypes, with little effort to integrate them into one common interface architecture. In part, this represented a bottom-up strategy where the component technologies had to be prototyped before determining what functionality would need to be supported in an integrated architecture. In truth, however, these schisms represented the difficulty of building a multidisciplinary research program. "Paradigm wars" between the groups characterized much of the effort during these years. The research groups focused on building prototypes that were acceptable within their discipline, rather than creating new paradigms that integrated capabilities across several disciplines to produce more intriguing interfaces.
By 1986 it was apparent that providing autonomy to the HI Lab's various groups was not producing the level of results that was expected from an MCC program. To reset the program, in 1987 MCC brought in Jim Hollan and gave him executive control over the HI Lab. Under Hollan's leadership the research agendas of the four groups were integrated into producing a powerful environment called the Human Interface Tool Suite (HITS) for designing intelligent multimedia interfaces.
In Hollan's words at CHI'91, "We were led to build HITS by a concern with the role of tools in supporting the complete interface design cycle, the role knowledge plays in the development of such tools and in their integration, the importance of a flexible run-time environment to support multimodal interaction, and the need to develop a new metaphor to mediate the way to think about collaborative interfaces and the tools used to construct them."
One of the most popular prototypes was a large interactive work surfacethink iPad but four times larger with multiple options for interactionthat served as the platform for demonstrating the power of HITS. Unfortunately, the work surface was so popular that research had to be stopped several days a week at the request of senior management to set up demonstrations for visitors who ranged from corporate dignitaries to the University of Texas women's basketball team.
Underpinning the interactive work surface was the golden nugget of the HI Lab, a blackboard architecture that allowed HITS to integrate many different types of input and output media, as well as different ways to analyze the content (knowledge bases, neural nets, etc.). In essence the research behind HITS was attempting to accept input in any form and develop a response appropriate in both media and content. This architecture was the culmination of the HI Lab's strategy to innovate integrating component interface technologies into powerful interactive environments rather than focusing on advancing the capability of the component technologies.
The HI Lab was an important contributor to the growth of the global HI community. I was the program chair for CHI'85 and the general chair (head honcho) for CHI'89 in Austin, "the wildest CHI ever." MCC hosted the first CSCW Conference and teamed with Anderson Consulting (now Accenture) and the University of Michigan to win one of NSF's first awards for CSCW research. MCC also produced a steady flow of conference papers and journal articles.
By the end of 1988, AI winter had set in. The AI hype of the early 1980s had given way to the realization that we were decades from developing expert systems that could reason outside their narrow domains or support full natural language interaction. In fact, as the group working on a large, commonsense knowledge base constantly restructured the links among knowledge elements, the language-based dialogue systems that depended on them broke almost daily. The challenges of intelligent systems were practical as much as theoretical, and they were not going to be solved within MCC's lifespan. In fact, the most successful spinouts from MCC, such as Evolutionary Technologies, solved specific problems blocking progress in a research program rather than offering solutions to grand technological challenges.
By 1989 the member companies were under greater competitive pressure and needed technology they could move to market quickly. The technology-transfer groups they had agreed to establish had not materialized, and Lisp became the popular whipping boy for their failure to adopt MCC technologies. The era of long-term research producing powerful Lisp-based prototypes at MCC was over.
Jim Hollan resigned from the HI Lab in mid-1989 out of frustration with MCC's change from long-term visionary research to short-term marketable technologies. I transferred back to the HI Lab nine months later to try and reestablish the support base in the member companies. With great gnashing of teeth, the HI Lab began porting the most promising prototypes to C++. However, the most talented researchers began to look elsewhere for the next step in their careers.
Ultimately four intersecting problems caused the demise of the HI Lab. First, as corporations decentralized in the late 1980s, research funds were dispersed to executives responsible for near-term profit objectives and uninterested in long-term research. Second, this decentralization also meant that no single executive possessed sufficient funds to support their company's membership in the HI Lab, so we had to coordinate several executives in each company to jointly agree to aggregate their funds. Third, since user interface was a young discipline, there was no executive who had corporate HI responsibility and felt ownership of the relationship with MCCwe were corporate orphans. Finally, we simply could not port the core of HITS fast enough to the many different platforms supported by all the member companies.
On July 27, 1990, MCC closed the HI Lab. As the HI Lab dispersed, Apple, Microsoft, and Bellcore were among the winners in capturing the departing talent. As a consequence some of the work generated at MCC has found a home in interfaces from other companies. For instance, the neural net-based character-recognition work initiated at MCC is now incorporated into Microsoft Windows. However, the promise and potential of MCC was lost in the early years for lack of integration both in ACT and the HI Lab. The productive years of the HI Lab were cut short by the realities of a competitive market in which many of MCC's member companies did not survive. Even with these problems, life on Silicon Ranch was exhilarating while it lasted.
Bill Curtis is senior vice president and chief scientist with CAST, a leader in providing technology for measuring and evaluating the internal quality of business application software. He is also the founding director of the Consortium for IT Software Quality. He co-authored the Capability Maturity Model (CMM), the People CMM, and the Business Process MM. He was co-founder and chief scientist of TeraQuest. He is a former director of the Software Process Program at Carnegie Mellon University's Software Engineering Institute. From 1983 to 1991, Curtis was a technical director of the Human Interface Laboratory and the Software Technology Program at MCC. He has also worked at ITT's Programming Technology Center, at GE Space Division, and taught statistics at the University of Washington. He has published five books, more than 150 articles, and was recently elected a fellow of the IEEE for his contributions to software process improvement and measurement.
©2010 ACM 1072-5220/10/1100 $10.00
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2010 ACM, Inc.