• Dheeren Vélu

Characteristics that shape an Enterprise AI Application

A few days ago I (once again) got into a bit of a deep discussion with an ex-IBM Watson colleague on whether a certain application that were talking about and which is being marketed as "AI Powered" was indeed AI Powered? We were trying to figure out what AI capabilities/technologies were being used and slowly the conversation went more into - actually, what should be the characteristics of an AI application or system, particularly in the Enterprise landscape. In this article, I have attempted to summarise, what I think are the main characteristics that will shape an Enterprise AI Application.

Firstly, let's define an enterprise AI application

AI (Artificial Intelligence) itself falls into two broad categories. The first is Artificial General Intelligence (AGI) a.k.a Strong AI. This is a hypothetical machine that exhibits behaviour at least as skilful and flexible as humans do, basically the intelligence of a machine that could successfully perform any intellectual task that a human being can. The second is 'Applied AI' (AAI) a.k.a 'Narrow' or 'Weak AI' and the focus here is on the use of natural language processing, machine learning, knowledge representation and other advanced computing techniques to build a machine that is focused on a specific or narrow task.

Whilst General Intelligence is a very important topic and numerous amounts of researches are being carried out in that space, there is a long way to go before such machines can be used. For now, such an AI lives only in the imagination of science fiction writers, Hollywood movies and in the hopes and dreams of the research scientist.

Applied AI or the other hand, has made tremendous progress in the last couple of years and whether knowingly or unknowingly it has already become a part of our lives. Example, recognising faces in facebook, autonomous driving cars by Google, voice-powered personal assistants like Siri by Apple and Alexa by Amazon and the list goes on.

 For the purpose of this article, I would like to focus on Applied AI and more specifically, Applied AI in the Enterprise scenario - Enterprise AI Applications. Such an application will usually be part of a large software system platform designed to operate in a corporate environment such as business or government and usually complex, scalable, component-based, distributed and many a time, mission-critical. This includes AI enabled process applications, industry applications and B2B products.

Based on my of experience in working with large complex enterprise applications and with my more recent experience in implementing AI solutions, I believe an Enterprise AI Application will be shaped by these 5 characterises (or capabilities):

1. Natural Interactivity

The entire history of human-computer interfaces has been about applying ever-increasing amounts of computing power to making the machines work harder to interact in a way that's easier for people. Recent advances in Natural language processing and understanding (NLP/NLU) has already brought in conversational and dialog oriented interfaces to many day-to-day applications, both in the consumer and enterprise domains. This is by far the most mature AI capability and is being used in the form of chatbots, virtual agents and assistants. An AI application should at a minimum have some natural interaction capability and some form if linguist intelligence, be it speech (voice) or text.

2. Knowledge Representation & Reasoning

Knowledge is the information about a domain that can be used to solve problems in that domain. To solve problems this knowledge must be represented in the computing system which is known as Knowledge Representation or simply KR. An AI application must have the ability to ingest large amounts enterprise data from both new and legacy sources, and efficiently encode and represent that into the system using appropriate knowledge models and schemas. The knowledge representation component must also be able to dynamically extend links to other internal and external knowledge sources as and when needed.

Knowledge representation goes hand in hand with automated reasoning because one of the main purposes of explicitly representing knowledge is to be able to reason about that knowledge, to make inferences, assert new knowledge, etc. An AI application must have capabilities to leverage language structure, probability, fuzzy logic, semantics and relationships to draw inferences

 3. Algorithmic Intelligence & Hypothesis

The enterprise AI system must be able to perform computations and pattern recognition leveraging historical data. It must effectively use statistics machine learning, classification, optimisation, ranking & scoring among others techniques to generate evidence-based hypothesis based on confidence scores. 

4. Continuous learning

There must be a mechanism for the AI application to continuously learn and evolve without being explicitly programmed but with new information/inputs, new analysis, new users and new interactions.

5. Variety Data Handling

And finally, an enterprise AI must be capable of integrating multiple heterogeneous data sources, structured and unstructured, static and streaming and facilitate synthesising ideas or answers from them. Whilst a lot of the legacy enterprise data will be available in defined structured formats (relational databases with defined schemas), a large portion of today's data is unstructured in the form of text-heavy and files, emails, PDFs etc;

These form the basic characteristics or capabilities of an Enterprise AI application system. However, some other special purpose or more advanced AI applications or platforms could have more computational intelligence and features like - Pattern Recognition,  Goal-Orientation, Perceiving Relationships, Advanced Autonomy, etc;

Hope this helps and this by no means should be used a 'Litmus Test' for Enterprise AI applications. These are purely my opinion on what basic components or features an AI applications/system/platform must have, considering the maturity, acceptance and realistic usage of AI techniques today. This will, however, change dramatically given the pace at which this technology is going.


7 views0 comments