AI System Architecture and Health Assessment

-

AI System

AI system receive inputs (data), process them via data processing applications, which can use AI algorithms for further advanced processing of data or intermediate outputs for generating final outputs for desired outcomes. These outputs could be classifications, clustering, contents, decisions, detections, insights, predictions or recommendations. There are several examples of AI systems such as Autonomous Vehicles, Chatbots,  Digital Assistants, Face Recognition and Voice Recognition etc.   

A. AI System Architecture

AI system architecture consists of at least three interconnected layers and underpinning components: applications, data and algorithms.  



Figure 1. AI System Architecture

1. Application 

AI system has AI enabled or augmented or embedded applications that receive initial inputs such as data and apply data processing programs. Based on the processing needs, it makes the data ready for further advanced processing using AI algorithms. As such application captures and generates intermediate outputs for the algorithmic processing. Application engages the required algorithms for generating the final outputs, which are then passed to the AI system users by the application user interface.

2. Data 

Data is core to AI system. Data architecture provides the design and integration of data components. Data architecture can include operational application data systems, data warehouse, data lake, data lake house, data mesh, data fabric (semantic layers), data exchange, and data satellite etc. AI system algorithms grow over the data. 

3.  Algorithm

AI system has different types of algorithms such as supervised learning, unsupervised learning, reinforcement learning and deep learning. These algorithms process large amount of structured, semi-structured and unstructured data including raw and pre-process intermediate outputs. Algorithmic data  can be organised into four areas: Training Data, Validation Data, Test Data and Live Production Data.  We can select an algorithm that fits to the subset of an application data such as the training data or vice versa. Algorithms process the data and results are called the product or model. 

B. AI System Health Assessment Model

Overall, quality of an AI system depends on the health of integrated applications, data and algorithms. This section discusses the AI System Health Assessment Model, which can be used to assess the health  of AI system from 7 perspective: (1) criticality, (2) business fit, (3) social fit, (4) technology fit, (5) privacy, (6) security, and (7) trustworthiness. 

Figure 1. AI System Health Assessment Model

References:
  1. Gill, A.Q. 2022. Adaptive Enterprise Architecture as Information: Architecting Intelligent Enterprises. 
  2. NSW Artificial Intelligence Assessment Framework. https://www.digital.nsw.gov.au/policy/artificial-intelligence/nsw-artificial-intelligence-assessment-framework 
  3. Artificial Intelligence Ethics Policy. https://www.digital.nsw.gov.au/policy/artificial-intelligence/artificial-intelligence-ethics-policy 


Comments

Post a Comment

Popular posts from this blog

What is Information? A Theory of Information Trilogy

-
Image
A Theory of Information Trilogy for Information Ecosystem by Asif Gill Information is/of Everything (IioE) We hear many definitions of information and related concepts such as data, information itself, knowledge and intelligence. Well, all these definitions seem useful from information probability, quantification, measurement, storage, processing and management perspective. We still wonder what is "it" and how to identity "it", which we call "information". This blog defines and describes nature ecosystem inspired definition of information, which is called here a "Theory of Information Trilogy" (ToIT).  This theory is defined with a view to define and identify information in complex and heterogenous information ecosystems.  This theory defines that the information is (1) matter and (2) energy, which has different (3) states or forms. Matter and energy flow in the natural ecosystem. Further, from natural ecosystem, these 3 matter, energy and their ...
Read more

Circular Information Lifecycle Management for Enabling Circular Economy

-
Image
Circular economy is about recycling, remanufacturing, reusing  products, services and information for continuous value re-co-creation while reducing waste and enhancing positive economic, environmental and social impacts. This requires to re-conceptualize and re-design the economic ecosystems and business models for optimisation and innovation.  Information management is extremally important to provide intelligence & insights for effective decision making  and actions enabling circular economy. This article presents the following circular information lifecycle management for enabling circular economy.  In circular information lifecycle management (CILM), information can be restored and recycled like another product or service in the circular economy.  The CILM is composed of six key stages: discover & collect, classify & secure, prepare & release, access & use, archive & purge, and restore & recycle.  Circular Information Lifecyc...
Read more

AI-enabled Intelligent Enterprise: AI Co-pilots

-
Image
AI-enabled Intelligent Enterprise  There is an increasing interest in Artificial Intelligence (AI)  systems for decision making, productivity gain, better quality products and services etc. There is a fear or assumption that AI systems will replace human. We should not assume that AI systems will replace humans, rather we can look at them as our agent or  assistants.  Human is the "Principal" who delegates the tasks to the AI system that is called an "Agent". Thus, there is a principal-agent relationship (delegation) between the human and AI system.   AI Co-pilot There are several facets of AI systems, in particular the AI co-pilot system that can assist humans for carrying out tasks on their behalf. These AI co-pilots can be embedded across the enterprise value chain. This article discusses the AI-enabled intelligent enterprise and the applicability of AI co-pilots across the connected core value chain of (1) strategy, (2) architecture, (3) development and...
Read more