A Very Brief History of Cognitive Assistants

Hopefully, someone will start a better discussion thread on the history of cognitive assistants, but here is a starting point or baseline discussion.   This very short history of cognitive assistants will examine instances of cognitive assistants.  By instances, we simply mean named systems, projects, or challenges.   However, first we must drive a stake in the ground concerning, “what is a cognitive assistant?”  For example, what capabilities help us distinguish a cognitive assistant from just a “good tool?”

How should we define cognitive assistant?

We will not try to summarize the extensive literature on cognitive assistants and evaluation measures for cognitive assistants in this short discussion, but see Steinfeld (2007) for more.  For our purposes in this short discussion, the difference between a “good tool” and a cognitive assistant can be a fine line.   We will use three cognitive capabilities to help distinguish “good tools” from cognitive assistants: language, learning, and levels (confidence levels in responses).   Language refers to natural language communications in words and sentences, but ultimately includes more – gestures, gaze, diagrams, and much more – all the ways people communicate with each other.   Learning refers to the ability to learn from positive and negative examples of questions and responses, but ultimately included more – direct user feedback and teaching dialogue, analogical reasoning, and more.  Levels refers to the ability to provide estimates of confidence levels in multiple possible responses, but ultimately includes more such as explanation, debating, and argumentation capabilities.  We might want to add a fourth capability “limbs” if we want to talk about embodied cognitive assistants – robots.

Paralleling a short four-stage history of Artificial Intelligence we will examine instances of potential cognitive assistants from these four eras: formative era, micro-worlds era, expert systems era, real world era.

Formative era

1945 MEMEX (Bush)
1962 AUGMENT (Engelbart)

These two early systems provided much of the vision for “cognitive assistants for knowledge workers ” by using technology to augment human intellect, especially with respect to symbolic processing of text and networks of inter-related concepts.

Micro-worlds era

1955 Logic Theorist (Newell and Simon)
1956 Checker Player (Samuel)
1966 Eliza (Weizenbaum)

There are of course many more examples of pioneering systems from this era, but these three provide a nice illustration of three categories.  The Logic Theorist can be seen on the path leading to systems such as WolframAlpha dealing with computable knowledge.    Samuel’s Checker-playing Program was the forerunner of so many game playing programs, culminating in Deep Blue, fulfilling and early AI prophesy of defeating the world champion in Chess, and more recently Watson Jeopardy!.   All these game playing programs can be used in a performance support mode to aid people learning and playing games.   Finally, Eliza illustrates that simple tricks and deception can make for entertaining aspects of cognitive assistants – linking to somewhat surface solutions or deception-oriented versions of the Turing Test grand challenge.

Expert systems era

1965-1987 DENDRAL
1974-1984 MYCIN
1987 Cognitive Tutors (Anderson)
1987 Knowledge Navigator System

The early expert systems led to corporations envisioning cognitive assistants for professionals to improve the productivity and creativity of knowledge-workers across a wide range of jobs.   Cognitive tutors are proxy for all the intelligent tutoring systems, learning support, performance support systems of this era – too numerous to mention.    The Knowledge Navigator video provided an update on MEMEX and AUGMENT in many ways with natural language dialogue and multimedia in an envisioned executive assistant and collaborative research assistant.

Real world era

2009 WolframAlpha
2011 Waton Jeopardy!
2011 SIRI
2014 Watson Solutions

WolframAlpha: Is WolframAlpha a “good tool” or a cognitive assistant?  Wolphra Alpha provides a natural language interface and computational engine for a wide range of natural language and mathematical language queries – pushing the limits of computable knowledge (WolframAlfa 2014).  It is a “good tool” evolving towards becoming a cognitive assistant, once users and other can help it learn, and once it provides levels of confidence on answers.  Currently, WolframAlpha tries to provide one “right” answer, or just give up and reply no answer found, rather than several possible and ranked answers.

Watson Jeopardy! Like all game playing systems, Watson Jeopardy! has a use case where it could be used by a person to compete against competitors to win a game.  Watson Jeopardy! clearly demonstrated some language, learning, and levels capabilities.  If it were to be packaged as an app, or in some other way to help people perform the task of playing and winning Jeopardy games then it could be considered a cognitive assistant.

SIRI: SIRI is clearly marketed as an intelligent or cognitive assistant.  It exhibits language capabilities, but not so much on learning or levels of confidence in alternative answer, but like a search engine it will often return alternatives, if asked for a list of possibilities.  The history of SIRI traces back to CALO and PAL, which were DARPA funded projects, at SRI International (Bosker 2013)

Watson Solutions: Engagement Advisor, Discovery Advisor, Watson Chef, and other Watson Solutions target specific market needs where human expertise needs to be augmented or scaled to improve productivity and quality of specific occupational tasks.   These systems use language, learn, and provide confidence levels for alternative responses.   Still, building these systems is complex and difficult.

Concluding Remarks

This discussion thread aims to collect comments about instances of cognitive assistants throughout history.  Beyond a workable definition of cognitive assistants in terms of capabilities, this discussion thread can ultimately contribute to the discussion of streamlined development and evaluation methodologies for cognitive assistants for all professions.  The Cognitive Systems Institute is motivated by the vision of augmenting and scaling human expertise – providing everyone eventually with an executive assistant, personal coach, and mentor for any and all occupations (Spohrer 2014).

References

Bosker B (2013) SIRI RISING: The Inside Story Of Siri’s Origins — And Why She Could Overshadow The iPhone
January 22, 2013 URL: http://www.huffingtonpost.com/2013/01/22/siri-do-engine-apple-iphone_n_2499165.html

Bush, V. (1945). As we may think. The atlantic monthly, 176(1), 101-108.

Colligan B (2011) How the Knowledge Navigator video came to be.
November 20, 2011 URL: http://www.dubberly.com/articles/how-the-knowledge-navigator-video-came-about.html

Engelbart, D. C. (1995). Toward augmenting the human intellect and boosting our collective IQ. Communications of the ACM, 38(8), 30-32.

Newell, A., & Simon, H. A. (1956). The logic theory machine–A complex information processing system. Information Theory, IRE Transactions on, 2(3), 61-79.

Spohrer, J. (2014) Cognitive Systems: Vision and Directions.  NUS Cognitive Colloquium.
September 12, 2014 URL: http://www.slideshare.net/spohrer/cognitive-20140912-v3

Steinfeld, A., Quinones, P. A., Zimmerman, J., Bennett, S. R., & Siewiorek, D. (2007, August). Survey measures for evaluation of cognitive assistants. In Proceedings of the 2007 Workshop on Performance Metrics for Intelligent Systems (pp. 175-179). ACM.

WolframAlpha (2014) Timeline of Systematic Data and the Development of Computatable Knowledge.
September 12, 2014 URL: http://www.wolframalpha.com/docs/timeline/computable-knowledge-history-6.html