The Context in Artificial Intelligence

Source — Internet

Abstract

Context forms the backbone of development in an Intelligent Computing setup. The concept of understanding and inferring has been pivotal for the rapid advancements in the field of Artificial Intelligence. The majority of it can be attributed to context comprehension. This article through its course covers the notion behind context and numerous aspects of context-driven intelligence. Additionally, a discussion on the positive and negative outcomes of its enablement is done for assessing the present state of context-aware applications while commentating on the future developments.

Understanding Context

Context can be defined as “Any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and application themselves” (A.K.Dey, 2000). The fundamental intuition carried by Context is the enablement of a human-centric view on content understanding and gauging environmental changes. These adaptations can include adjustments to the display characteristics, or a selection of relevant information and services that are presented to the receiver (Pichler, Bodenhofer and Schwinger, 2004). A system that comprehends context forms an important narrative of Intelligent Autonomous Systems (Muhissen, Mudar & Salah, Zaki & Shaikh and Nihal, 2020). Contextual AI has empowered countless solutions and services, working towards the concept of AI For All through Personal Assistants and Autonomous Cars (Danaher., 2018). Contextual AI is an improvement on the Rule-Based engineering approach followed by intelligent agents. For instance, Deep Blue (Tomayko, 2003), the chess computer that beat Garry Kasparov, or Expert Systems for Healthcare do not necessarily exhibit context understanding and are narrow in their spectrum.

The Context in Artificial Intelligence

Context-awareness presents totally new requests for the knowledge of utilizations. Context-aware applications need to have a refined thought of the environment encompassing them and to make fitting moves to change contexts. From this perspective, context awareness normally turns into a prompt application field for AI standards and strategies (Pichler, Bodenhofer and Schwinger, 2004). A significant trademark that is normal to the Contextual AI is the way that inclinations and profiles are for the most part hard to catch with solitary yes-or-no ideas. Hence, modern style intelligence system builds, for example, modular, temporal, many-esteemed or fuzzy logic, which should give strong aid to context portrayal (Fodor, Roubens, 1994). The increase in the address of Context in AI in recent years can be attributed to the giant strides taken in the sub-domain of Deep Learning, solving problems in the Natural Language Processing and Computer space. More recently DARPA has undertaken the task of creating AI for the future, imparting common sense to machines (Gunning, 2018).

Pillars of Contextual Artificial Intelligence

Contextual AI needs to be intelligible, adaptive, customizable and controllable, as well as context-aware. Here’s what that looks like in the real world: When discussing Contextual AI, we necessarily do not talk about a specific set of algorithms or a fixed approach. It can however be achieved through mapping human interactions with their environment through logical reasoning and factual discourse into a sequence (Brigui, 2021). A typical Contextually Intelligent system should be Intelligible, Adaptive, Customizable and Controllable (Brdiczka, 2021). For instance,

1. Intelligible (Comprehensible) refers to the feature of being self explainable. Such a system could simplify its being. To believe their conduct, we should make AI intelligible, either by utilizing innately interpretable models or by growing new techniques for explaining and controlling in any case predominantly complex choices utilizing approximation, vocabulary or explanations. (Weld and Bansal, 2019)
2. Adaptivity is the ability to successfully process change in the environment and alter accordingly so as not to hinder progress in the process iteration. The work in the field of transparent Artificial Intelligence talks about the machine objective to correlate the internal data models with predictions to develop self-adaptive algorithms that perform generalized actions. (Rogerio and Grzes, 2019)
3. Being Customizable is an important behavior that a Contextual AI agent should exhibit. AI Assistants like Alexa, Cortana, Siri (López, Quesada and Guerrero, 2018) are beaming examples of improvements made toward Customizability. One of the salient features of these solutions is their callback ability that changes with changes in user profiles.
4. Context-Aware AI people are confronted with an expanding number of figuring devices they associate with different circumstances in daily existence. Along these lines, the significance of thinking about this different circumstance is firmly expanding. By giving applications information about the individual inclinations of their client, limitations of the gadget utilized, the client’s present area, and so on (Pichler, Bodenhofer and Schwinger, 2004).

Success in Context-Aware Machine Intelligence

Context-aware applications provide added value to their users. In recent years, with the advent of AI and its engagement in the context awareness space, we have seen rampant growth in user engagement and solution assertion. Such systems have provided us with important benefits, some of the most dynamic ones being Adaptation, Personalization, and Proactivity, (Pichler, Bodenhofer and Schwinger, 2004).

1. Adaptation means to change a feature or data as per available contextual needs. It incorporates activities with various impacts like filtering of data, the conjuring of extra administrations, and the deactivation of administration parts. The intricacy goes from fine-grained adaptations like considering the area during the choice of data to reconfiguring the total assistance (Kappel, Retschitzegger and Schwinger, 2003). A genuine model is the context-mindful postbox demonstrator (Hofer, Schwinger, Pichler, Leonhartsberger, Altmann and Retschitzegger, 2003)
2. The thought of personalization addresses a significant test since the end client has been placed in concern while creating intelligent applications, which traces all the way back to the mid-80s. Personalization means to adjust an application to various people, with the end goal that they see the application distinctively simultaneously, as indicated by every individual’s inclinations, propensities, abilities, undertakings, and so on
3. One more advantage of context-aware applications is proactivity. Presently, most frameworks offer responsive administrations as it were. Receptive assistance requires the client to effectively pull data from the help by giving an express solicitation. A proactive help, then again, conveys or pushes data to a client without express solicitation (Fischmeister, Menkhaus and Pree, 2002).

While aforementioned points are generalized to fit a school of solutions into the umbrella of context-aware intelligent solution setups. We are now at the cusp of a major engineering revolution where the commercial usage of Contextual AI is on a rise. Here are the following instances where successful assimilation of context in AI has created an impact,

1. Contextually enriched Chatbots in Healthcare domains created using Neural Networks are used for early diagnosis and medical assistance (Kandpal, Jasnani, Raut and Bhorge, 2020).
2. Contextual AI is used in the marketing strategizing industry where advertisement allocation happens through context understanding of a user’s content subscription. Such systems derive logic, map it with context and index it for increased reach and scale (Stormon, 2018).
3. Automated cars leverage both context-aware Natural Language Processing and Computer Vision. Scaling pinnacles in the field of self-driving vehicles. Thus, creating a mammoth impact in this process (Shantanu and Phute, 2016).
4. Personal Assistants like Siri, Cortana or Alexa understand and personalize according to user profiles, engaging with adults and kids differently, adding an important dimension to the human-machine relationship. Commercializing and Familiarizing the general persona with AI in everyday life (López, Quesada and Guerrero, 2018).

Apart from the aforementioned areas of development, there stand varied solutions scattered across numerous domains, catering to a specific problem. Smart lights adjusting with respect to the time of the day or an active human in the room are in their own sense trivial examples of context-aware intelligence.

Failures in Context-Aware Machine Intelligence

Context assumes a significant part in domains where exercises infer thinking and understanding and must be gotten by experience. This interest in the utilization of context suggests that there is no unmistakable objective condition of the context. The context appears to have, as per the domain, a twofold nature: static or dynamic, discrete or continuous, knowledge or process. This obvious twofold nature emerges from the way that the idea of context is reliant in its translation on an engineered science versus a designing (or framework building) perspective (Brezillon and Abu-Hakima, 1995). This explains why there is a hypothesis versus practice hole, and why it appears to be hard to endeavor to bring together the different thoughts of context up to an agreement in a democratic manner. As a result, one considers context as an idea with complex geography, a metaphysics, a common space of information, a steady arrangement of presumptions, a semantic foundation, the climate of correspondence, a bunch of limitations that limit the admittance to parts of a framework, and so forth. Understanding the failures of Context-Aware machines help us reevaluate our comprehensiveness in the context of Intelligent AI. A few gaps in the Context-Aware Machine Intelligence ecosystem can be,

1. Context Interpretation and Modeling

We can abstract contexts but cannot isolate contexts when provided in a bunch. Often contexts are interrelated and depend on one another for contextual information for environment creation and assessment. For example, a chatbot cannot clearly understand the concept of a conversation until it learns to thread one message with another. Similarly, solutions that include rule-based approaches like logic programming, inductive learning, clustering, data mining, data-driven modeling, and statistical analysis fail to capture context (Luger and Stubblefield, 1990).

2. Context Matching

Context Matching is used to map a context with a condition. It can be simpler discussion points, like turning off the heat when the milk is boiled, it can also get further complicated too as contexts become nested. For instance, the dating app shares profiles for connections based on the similarity matrix scores they calculate based on user profiles, but that need not be the only metric. Secondly, it is a naive assumption that a reasonable number of well-fitting matches can be found under all possible conditions (Pichler, Bodenhofer and Schwinger, 2004).

3. Reasoning

The reasoning is one of the key areas where the Contextual AI hasn’t tasted much success. Contextual AI has been rapidly developing, but the core functionalities of the system still remain a mystery. The eco-space is full of black-box models that leave little scope for understanding. There seems no way of understanding/deriving the intuition behind the recommended suggestions that the AI makes. To classify it between right or wrong is again a matter of human discretion with no way to validate its results. One prime example of reasoning can be the internet search pattern, where suggestions are presumably based on past history but one cannot affirm it with complete surety. For instance, Google understands when a certain user searches for apples, it means the electronic giants and not the fruit. The quality to reason out is based on a preconceived understanding. With such a closed setup, analysis of the outcome is used to reverse engineer the solution. One thing could be to look at the Specificity and Sensitivity to understand the movement of data through the system. There is no other way of knowing whether all variations or permutations of that error have been addressed (Srinivasan, 2016).

Furthermore, a few examples that support the claim of failures in Contextual Intelligence are,

1. The inability of Personal Assistants like Google Assistant, Siri, Cortana in handling humor, sarcasm, double negatives and longer conversations
2. Context understanding in Computer Vision solutions remains a weak aspect of Contextual AI. For an instance, determining if the person is in distress or is resting
3. The Contextual AI fails to answer questions pertaining to its existence, often giving preset answers and hampering interaction flow

Conclusion

Through the course of this article, an intuition about Contextual AI gets developed, along with all the benefits and gaps. From an approach perspective, there is a huge gap between theory and practice which will eventually be filled using faster computing and revolutionary innovation. Contextual AI across healthcare, education and operations could result in a massive positive impact. It also ships the idea of AI to everyone. Certainly, Contextual AI is the future; and the future is now!

Bibliography

Dey, A., 2000. Providing Architectural Support for Building Context-Aware
Applications. College of Computing, Georgia Institute of Technology.

Pichler, M., Ulrich, B., Schwinger, W. and Hagenberg, 2004. Context-Awareness and Artificial Intelligence. ÖGAI Journal, Vol. 9.

Muhissen, M., Salah, Z., Shaikh and Nihal, 2020. Autonomous Systems — Artificial Intelligence AI by PISIQ. Artificial Intelligence and Robotics Research. Vol. 6.

Danaher, J., 2018. Toward an Ethics of AI Assistants: an Initial Framework.
Philosophy & Technology. Vol. 31.

Tomayko, J., 2003. Behind Deep Blue: Building the Computer that Defeated the World Chess Champion (review). Technology and Culture. Vol. 44, pp. 634–635.

Fodor, Roubens, M., 1994. Fuzzy Preference Modeling and Multicriteria Decision Support. Kluwer Academic Publishers, Dordrecht.

Gunning, D., 2018. Machine Common Sense Concept Paper, Defense Advanced Research Projects Agency (DARPA)/ I2O.

Imène, B., 2021. Dealing With Context in AI [Online]. France: The European
Business Review. Available from:
https://www.europeanbusinessreview.com/dealing-with-context-in-ai/ [Accessed
15 Feb 2022]

Brdiczka, O., 2021. Contextual AI: The next frontier of artificial intelligence
[Online]. United States: Adobe Experience Cloud Blog. Available from:
https://business.adobe.com/blog/perspectives/contextual-ai-the-next-frontier-of-artificial-intelligence [Accessed: 15 Feb 2022]

Weld, D., Bansal, G., 2019. The challenge of crafting intelligible intelligence.
Communications of the ACM. Vol. 62. pp. 70–79.

Rogerio, L., Grzes, M., 2019. Self-Adaptive Artificial Intelligence. 14th
International Symposium on Software Engineering for Adaptive and
Self-Managing. pp. 155–156.

López, G., Quesada, L., Guerrero, L., 2018. Alexa vs. Siri vs. Cortana vs. Google Assistant: A Comparison of Speech-Based Natural User Interfaces. International Conference on Applied Human Factors and Ergonomics. pp. 241–250.

Kappel, Retschitzegger, W., Schwinger, W., 2003. Customization for ubiquitous Web applications — a comparison of approaches. Int. J. Web. Eng. Technol.

Hofer, T., Schwinger, W., Pichler, M., Leonhartsberger, Altmann, G., Retschitzegger, W., 2003. Context-awareness on mobile devices — the Hydrogen approach. Proc. 36th Annual Hawaii Int. Conf. on System Sciences. Big Island, Hawaii. pp. 292–301.

Fischmeister, S., Menkhaus, G., Pree. W., 2002. Context-awareness and adaptivity through mobile shadows. Technical Report TR-C047, Software Research Lab, University of Salzburg.

Kandpal, P., Jasnani, K., Raut R., and Bhorge, S., 2020. Contextual Chatbot for Healthcare Purposes (using Deep Learning). Fourth World Conference on Smart Trends in Systems, Security and Sustainability. pp. 625–634.

Stormon, A., 2018. Contextual Intelligence: Everything you need to know and why it matters [Online]. United States: Oracle Advertising Blog. Available from: https://blogs.oracle.com/advertising/post/contextual-intelligence-everything-you-need-to-know-and-why-it-matters [Accessed: 20 Feb 2022]

Shantanu Ingle, S., Phute, M., 2016. Tesla Autopilot : Semi Autonomous Driving, an Uptick for the Future. International Research Journal of Engineering and Technology (IRJET). Vol. 03.

Brezillon, P., Abu-Hakima, S., 1995. Using Knowledge in its context. AI
Magazine. Vol. 16. pp. 87–91.

Luger, G., Stubblefield, W., 1990. Artificial Intelligence and the Design of Expert Systems. California: Benjamin-Cummings Publishing.

Srinivasan, V., 2016. Context, Language, and Reasoning in AI: Three Key
Challenges [Online]. United States: MIT Technology Review. Available from:
https://www.technologyreview.com/2016/10/14/156968/context-language-and-reasoning-in-ai-three-key-challenges/ [Accessed: 15 Feb 2022]