Cognitive Assistance in Government Papers from the AAAI 2015 Fall Symposium Cognitive Assistance at Work Hamid R. Motahari Nezhad IBM Almaden Research Center San Jose, CA 95120 United Stated motahari@us.ibm.com their main communication paradigm and information sharing, 5) Vast amount of information that are produced, communicated, processed and needed to be managed by a worker in the work context in order to perform work effectively; 6) And, the fast pace of the work that has led to shaping new interaction and communication patterns and habit, and in particular the increasing use of instant and real-time communication as integrated part of work productivity tools. There are already startups and enterprise applications that innovate by bringing messaging tools and integrating it into enterprise collaboration platforms. As the result of these trends, we witness an enormous shift in the collaboration platforms which is being replaced with integrated, social and stream-driven collaboration environment where all information such as messages, files apps are reporting into an integrated view, and more importantly (cognitive) agents (automated bots) are participating in conversations among human to facilitate work. In particular, the growing trend in bringing conversational virtual agents into work context presents unique opportunities and challenges. The opportunities are enormous in terms of capitalizing on these agents to assist humans in be more productive, and have them to perform jobs that a machine would be superior and better than a human, but in a way that complements human worker. The challenges include characterizing the type and abilities of a human and machine, and investigating the art of possible in technologies to have the two (human and machine) to effectively communicate and collaborate. One other typical challenges of AI, and for building cognitive assistant [EDWARD A. FEIGENBAUM 2003] has been characterized as the ability of machines to build a model of world, their human subject and themselves in a scalable, adaptive and dynamic manner. The huge amount of data available in the work context, generated as the results of human interactions, workflow, enterprise applications, databases and knowledge bases and the advances in methods for processing and building knowledge models out of unstructured information in systems such as IBM Abstract Today’s businesses, government and society work and services are centered around interactions, collaborations and knowledge work. The pace, amount and veracity of data generated and processed by a worker has accelerated significantly to the level that challenged human cognitive load and productivity. On the other hand, big data has provided an unprecedented opportunity for AI to tackle one of the main challenges hindering the AI progress: building models of world in a scalable, adaptive and dynamic manner. In this paper, we describe the technology requirements of building cognitive assistance technologies that assists human workers, and present a cognitive work assistant framework that aims at offering intelligence assistance to workers to improve their productivity and agility. We then describe the design and development of a set of cognitive services offered by the framework, based on advanced NLP and machine learning methods. The cognitive services help workers in processing and linking information and identifying and tracking work items over interactions in communication channels such as email, social conversations and media, chats and messaging and calendar applications. These cognitive services are designed to be adaptive, online and personalized so that over time adapt to changing environment and knowledge, and the models become personalized through learning preferences and working language and style of the subject worker. Introduction The nature of work is undergoing tremendous transformation. The key drivers of this transformation are as follows: 1) Globalization and geographical distribution of workforce (remote and global workforce); 2) Faster business cycles and business agility due to faster product and innovation cycles, and shorter time to market and time to service; 3) Mobility and the wide adoption of mobile technology among workforce with bring your own device becoming the mainstream; 4) Millennials getting into the workforce, who have grown up with mobile and social networking as Copyright © 2015, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. 37 Watson [AAA 2010] has opened up a new era for AI, and for enabling technologies for cognitive assistants, towards building such models of world, human subject and cognitive agents. Nevertheless, this is not a solved problem, and the problem of developing cognitive assistants requires taking incremental steps in understanding the technology need, limits and problems that are enabler of building cognitive assistant. In this paper, we investigate the problem of building cognitive assistants that help and complement human workers. A human worker spends on average of 28 hours on collaboration, coordination and communication activities (email –reading and responding-, calendar and meetings, communicating on social media and other communication tools). This amounts, on average, for 70% of a normal 40-hrs work week for a knowledge worker in private enterprise and also in government sectors. To improve with productivity in this space, we focus on the problem of offering cognitive assistance over conversations and work organization for a knowledge worker. We start by the generic problem of offering cognitive assistance, characterizing the cognitive abilities of a human and machine (and the ones that each excel at), and then continue by introducing innovative technologies methods for the design and development of a cognitive productivity assistant for human workers. current AI technologies. Examples of this category include engaging in creative conversations, emotional intelligence, etc. Symbol manipulation also happens in the lowest level of hierarchical structure of brain function. The higher levels of hierarchical structure of brain function involve emergent concepts where higher level concepts/ideas combine, and form complex organisms (take an analogy with ‘cloud’, a whole, relation to air and water molecules, component). Arguably, there is a subtype of synthetic cognition that relies on fast and efficient processing of a large amount of information, which is out of power of human intelligence, while machines excel at those type of cognition. It is natural to observe that this is one area where machines can complement humans. This is where we define the scope for cognitive assistance to a person, and in work context, in particular. A software program (agent) can be characterized as intelligent if it can employ computational intelligence techniques in order to define a model of the world that facilitates synthetic cognitive tasks. However, the challenge in achieving this type of intelligence is the building of the knowledge model of the world and the domain that allows understanding of the meaning of the intellectual tasks. In artificial intelligence literature, there has been considerable amount of research in knowledge representation and knowledge acquisition technologies through reading text and other data types, and methods based on semantic and symbolic representations flourished by the concept of building ontologies in the context of semantic web, and linked data. Though, most approaches have not been able to demonstrate the process of knowledge building can be done in a scalable, adaptive and dynamic manner when entering new domains or adding new information to the base model. Though, cognitive computing methods, and in particular those methods developed in the context of Watson including deep learning methods, for processing unstructured information and building models of the world (for specific tasks or domains) can be considered as the first step towards building intelligent methods that can build models of the world in a scalable manner by processing unstructured information (without the need for manual crafting of models – e.g. ontologies). Though, these models yet have not been shows to be adaptive and dynamic. In this paper, we investigate the problem of offering cognitive assistance to enterprise workers by offering analytical cognitive capabilities that supports synthetic cognitive tasks in the context of human collaboration and communication tasks, and organizing work and performing work for human workers to support higher productivity and agility. There has been a tremendous progress in the development of personal assistants in the industry. In particular,the most popular personal assistants include Apple Siri, Problem Statement and State of the Art A cognitive agent (CA) is defined as a software tool that augments human intelligence [Engelbart 1962]. To achieve this objective, a CA should offer complementary cognitive capabilities to a human by picking up those cognitive tasks that are time consuming, daunting or require high computational and cognitive power beyond human intelligence, but in which a machine is greater than human. Let us characterize the overall cognition capabilities into those that are main differentiator of a machine and human. Cognition capabilities are categorized into two major types [Eric Lord, Science, Mind and Paranormal Experience, 2009]: (i) analytical cognition, and (ii) synthetic cognition. Analytical cognitive skills include those that the machines excel at them but would take a lot of intellectual efforts from human. Examples of these capabilities are mathematical calculations, making logical decisions in complex situations requiring a series of computation, and chess all of which are recognized as computational intelligence. This type involves manipulation of symbols (symbolic processing) through algorithmic information processing. At this level, the processing units does not know or care about the “meaning” of symbol. On the other hand, the second type of cognitive capabilities are those that human performs effortlessly but are hard for machines with 38 Google Now, Microsoft Cortana, Amazon Echo, Braina, Samsung's S Voice, LG's Voice Mate, SILVIA, HTC's Hidi, Nuance’ Vlingo, AIVC, Skyvi, IRIS, Everfriend, Evi, and Alme (patient assistant). There are also productivity focused agents including Amy (x.ai), Genee for scheduling assistant, which focus on one specific domain. These applications offer conversational interfaces for human starting by voice (or some text) which is then processed by these agents to offer services. A related paradigm is the notion of offering intelligent conversational assistance as a service. A number of these platforms include Assistant.ai (human speech conversational services), ChatBots (chatbots.io), telegram platform for including bots in chat conversations and Watson’s Dialog Service in IBM BlueMix/Watson Developer’s Cloud. All these offer enabling technologies for a personal assistance however, none, yet, addresses the problem of supporting the human workers in getting work done and improving their productivity. This is the focus of this work to offer such cognitive assistance. tems, knowledge basis in forms of structured and unstructured information. In order to support a worker in above three scenarios, we define a Cognitive Work Assistant as an intelligent software for the work context that offer cognitive services to a worker following the mythology of: monitor, process, recommend and act. In particular, it would offer the following capabilities: Understands human language; Monitors collaboration channels including email, calendar, chat and enterprise information sources; Builds a model of the user and the world (work context), and is situational aware (context); Offer assistance by pre-processing information, and presenting information in human understandable format; Categorizes and filters information; Gathers and organizes related information; Schedules meetings and manages the time on behalf of its human subject, Identifies requests, and organizing to-dos of its human subject, Assists in performing tasks such as organizing events, travel assistant; And, suggests taking certain actions to its human subject that supports increasing productivity, and growth. We also recognize that there are different roles in the enterprise that may benefit from such a work assistant: managers (people who have a human personal assistant), employees (who do not have a dedicated human personal assistant), and human personal assistants themselves in get- Cognitive Work Assistant We envision that need for cognitive assistance of a knowledge worker at three different domains in private or government sectors: (1) the assistance in on-boarding, Employee Cognitive Agents Community of cognitive agents that collaborate effectively with one another to support human activities. Assistant’s Cognitive Agents Expert Cognitive Agents Cognitive Assistant Platform Interactions types need to be supported: • Cog-to-Cog interactions, • Human-Cog interactions, and • Cog-backed human-to-human interactions Figure 1. Different types of cognitive work assistants form a community, each specializing in offering one type of service orientation and growth, which entails ability to proactively point the worker to the right information and material at the right time, (2) assistance in off-loading work entailed in communication and interactions, which account for a significant part of a worker, as it was pointed out earlier in this paper. In this context, the worker would interact using synchronous and asynchronous communication channels such as email, calendar, texting/messaging and network information diffusion and consumption. And, (3) the work and project management context in which the worker would need to interact with organization application sys- ting their job more efficiently and effectively. Therefore, we define the following three types of personal assistants: Cognitive Employee Assistant: These assistants would have access to the data space (and applications) that their human subject (employees and managers) has access to with the same level of visibility, and offer cognitive work assistance to them. Assistant’s Cognitive Assistant: An assistant to a human assistant helps them to become more productive, and focus on work that require human judgment, while more routine requests to them is handled by the cognitive assistant themselves. 39 Expert Process Assistants: This type of assistants are er of an email, participant in a chat session, or receiver of a experts in a specific domain such as travel, human resources (HR), etc. and are accessed by personal assistants to serve their human subjects. Delivery Platforms In Figure 2, we present an architecture for a Cognitive Work Assistant Platform targeting offering Cognitive Work Assistant Platform cognitive services for improving the Cognitive Services APIs Conversation Interface Bot (voice, text) productivity over communication Context-aware Calendar and To-do, Task Email Analytics, Personal and collaboration tools and platInformation Finder Scheduling and Process Auto-Response, Model Builder Assistant Assistant Classification forms in the enterprise. The key focus is on offering actionable insights integrated into collaboration Watson Apps or Services on BlueMix tools such as emails, chats, workspaces and other work management Concept and Knowledge Questions Natural Language Parsing Watson Dialogue Relationship Graph and Answer Toolkit (PoS Tags, and Service systems through integrating an inExtraction Builder Dependicy Parser) telligent agent that processes unstructured (text) and structured inEnterprise Repositories, Applications and Data Sources formation and identifies action item (or tasks) that a worker is assigned Feeds Document Repositories collections (requests), or commitments and promises that he makes to others. Figure 2. The Cognitive Work Assistant Framework and Platform Assisting a worker requires personalizing all such services, and therefore the “personal model builder” text/sms, etc.). module is responsible for scanning and preparing a model Action recommendation. This module’s functionality is of the world and the user’s work by monitoring the work to determine the type of the actionable statement, and make environment including people he is interacting with, his recommendation to the subject to take an action. Examples communication and conversation, and key entities includincluded adding to to-do list, or updating the status of an ing to-dos, follow-ups, calendar invites, and processes that action in the to-do list, scheduling a meeting, canceling or the person is involved in. rescheduling a meeting. The system has a number of action The cognitive work assistant is architected as an agent types that it recognizes, and a template language that althat can be offered as a mobile app that can interact with lows definition of new action types and corresponding logusers using a voice activated or messaging interface (a chat ic to hand the action by the agent. bot) that can participate in chat sessions. In addition, it Context-aware information finder. This module is an offers the same services via API calls. At a lower level, the advanced search module that proactively scans the inforcognitive assistant benefits from Watson services on mation space of a user and makes links and connections Bluemix, and a natural language parser that delivers part of among the entities (emails, calendar invite, files, people speech tags, word dependencies to verbs, on top of which and other supported entities). For identifying such links, an advanced natural language model builder extract key the information within the content of the entities, and text elements for actionable statements within conversations, and the role of the target users will be considered. and are delivered to the higher level modules within the cognitive work assistant. In the following, we provide an overview of key components of the cognitive work assisReferences tant. EDWARD A. FEIGENBAUM, Some Challenges and Grand Actionable Insight Analytics over Conversations. The Challenges for Computational Intelligence, Journal of the ACM, conversation among workers include exchange of work Vol. 50, No. 1, January 2003, pp. 32–40 item and action requests and promises. The module related David Ferrucci, Eric Brown, Jennifer Chu-Carroll, James Fan, to actionable insight over Email, Chat and other communiDavid Gondek, Aditya A. Kalyanpur, Adam Lally, J. William Murdock, Eric Nyberg, John Prager, Nico Schlaefer, and Chris cation channel focuses on identifying from the text, actionWelty, Building Watson: An Overview of the DeepQA Project able statements that are of type request, promise or quesAAAI Magazine, 2010. tions addresses to the audience of the conversation (receiv- 40