In the field of Big Data and Analytics (BDA), Artificial Intelligence (AI) is running hot with research, industry, venture capital and consumers consciously (or unconsciously) investing in its use, but it remains one of the most commonly misunderstood capabilities of the Analytics field. It stirs a range of emotions for those not well versed in the technical detail of what AI really means. What makes AI confusing is the breadth and complexity of its methods. It is both pervasive in areas such as search, and in its infancy in many suitable commercial domains.
It is becoming common for Analytics companies to apply the moniker of AI to their capabilities, as do we at Ailytic, so it is worth explaining the differences in AI approaches to real world problems to help cut through the confusion. All business sizes can benefit today from some form of sophisticated data modelling. Although AI is technically complex, it can be understood simply as the growing ability of computers, using advanced algorithms, to recognise patterns and to make logical inferences. Think about that, by taking a particular problem in your business that has sufficient data, you can simply create a model of the problem and use an algorithm that fits into the classification of AI to help create possible answers to the problem and hey presto, you are using AI! But is it really AI?
Traditionally AI has been split into two broad categories of application and associated technique:
Artificial General Intelligence
Applied Artificial Intelligence
Artificial General Intelligence (AGI)
Simplified to describe the strain of AI focussed on human like reasoning. This historical pure AI research goal tries to mimic human behaviour and aligns most closely with the “robot taking over the world" belief. While certain researchers define this as Pure AI, it remains an elusive and very difficult goal with existing technologies available to achieve true human-like reasoning, interpretation and emotion. Interest has waxed and waned over the last 5 decades, at times blossoming into commercial realms, though mostly staying confined to dedicated researchers in confined environments. Today their has been a renaissance where the ecosystem of research and application, and supporting computational power has grown to the point where there is a real chance momentum will deliver something close to the original goal in the coming decade or two.
This is the application of AI techniques and computer algorithms to model and solve particular problems, or to discover hidden insights in data. Approaches such as machine learning, expert systems, inference engines, and pattern recognition (speech and voice recognition/text and image recognition) expressed through interfaces such as Microsoft Cortana, Apple Siri and Google Now are all examples of applied AI. These most recognised BDA techniques and AI approaches can:
Observe past user interactions and other user behaviour
Gather sensory and semi-sensory inputs, such as the user’s geo-location and weather conditions, and the user’s body temperature, heart rate, etc.
Ask the user for additional information, feedback, or clarification
Access additional data, and apply additional algorithms
Store all the above as an always up-to-date user profile
Enormous investment is going into these areas of AI research and the commercial application for consumer and business needs.
Applied AI also includes the utilisation of computer algorithms to solve real world business needs such as a manufacturing challenge of determining what product to bring to market next based on consumer demands. The specification and application of a range of algorithmic approaches to well defined problems has absorbed the majority of effort in using AI in the past 30 years. Examples include Meta-heuristic and Hyper-heuristic techniques combining optimisation approaches such as, evolutionary computation (genetic algorithms), particle swarm intelligence and simulated annealing - applied to logistics, scheduling, and sequencing type problems in business. The application of these optimisation techniques will only increase as the capability moves from limited research to real world use by technology providers specialising in AI.
All planning and scheduling tools used in business today are simplified representations of the real world environment they are supposed to represent. This is acceptable, as there is always a trade-off between model accuracy (resolution) and the processing time they take to perform a function (e.g change the order of a series of production orders, or optimise a particular time period). With advances in computational power, the models we can create to simulate real world possibilities are improving greatly, and big data is a key enabler of superior business outcomes.
Why local, regional and global companies should be thinking about AI.
As our understanding of Applied AI improves and traditional computational processing power constraints relax, more businesses are looking to AI to gain a competitive advantage. The reality is that most businesses today have not assessed or adopted these approaches to continuously improve efficiency. There is often more algorithmic power in the mobile phones carried by each employee than the organisation utilises to assist in making better business decisions. Spreadsheets are often still the defacto decision making tool set as they are cheap, flexible, empower tribal knowledge retention and are difficult to remove/replace. A spreadsheet may represent an acceptable framework on top of which AI techniques can be applied but maintaining and scaling those solutions is difficult. Frost and Sullivan summarise why businesses should be taking AI very seriously. Consider these three scenarios:
AI’s powerful inference and pattern recognition capabilities could improve the speed, efficiency, and effectiveness of almost any legacy automated system. What other improvements might an intelligent system in the middle of a legacy environment suggest, or require? How will organisations sort out the many new ways in which these augmented and autonomous systems might interact, and choose the best configuration?
Text- and image-based AI solutions could put pertinent information at the fingertips of low-level personnel, as IBM’s Watson is doing already in some retail applications. If these employees have well-guided access to complete product and service information, they will not need to escalate customer requests, or ask their managers for assistance. This will speed customer service and save companies money; but how will this affect the now unnecessary product specialists and managers, and the rest of the organisation?
AI-based applications could optimise the value chains of industry, eliminating waste, and speeding up product and service deliveries. Whose priorities will drive these improvements? How will multiple entities cooperate to achieve overall optimisation objectives? AI research and application development are advancing rapidly; and business decision makers need to be paying attention to these developments, engaging AI-enabling software companies who can offer a highly differentiated platform for decision making.
Ailytic and BDA
Ailytic is focussed on deriving the maximum benefits of Big Data Analytics to drive the decision model – the plans, schedule or system that assists in determining how we achieve our goals. Big Data from manufacturing and industrial automation systems, combined with predictive analytics can expose powerful trends, relationships and metrics (manufacturing performance analytics) to then feed constrained dynamic models for production optimisation purposes. This approach can make the underlying model used as close to the real world as possible before invoking AI, or other prescriptive analytics to run modelled simulations on what is best for the given business objective or KPI. The combination of big data analytics, accurately constrained simulation and advanced optimisation represents a next generation approach to supply chain efficiency.
 &  Frost and Sullivan - Artificial Intelligence - A Practical Assessment. 2015.,
 Frost and Sullivan - Predictive Analytics in the Real World. 2015.