Understanding the Technology Behind AI6 November, 2020 / Articles
If you’ve found yourself perplexed by the question of what AI is, you are likely to be even more confounded by the question of how AI works. A very deep and complex topic, but here are the key principles for non-technical stakeholders.
From the previous article and a cursory knowledge of AI, we know that AI can be applied in a variety of different ways and used to perform different types of tasks. AI is both a set of tools and the toolbox and we must ensure we select the relevant tool for the relevant task. Whilst the context of its application may vary, as do the tools and data used to generate the desired outcome, there is a method that is relatively consistent across all applications.
Algorithms and Models
Before we begin, it is important to establish top-level clarity and alignment on a few terms that whilst you may have heard of in relation to AI, perhaps have never made a great deal of sense. Those terms are algorithms and models.
An algorithm is a set of instructions designed to perform a task. In the context of AI, the algorithm applied over time is what enables a machine to learn how to perform a task on its own, without or with minimal human intervention. The set of instructions that is the algorithm dictate what functions are performed on the data supplied to the algorithm, which is also relevant to the task the algorithm is intended to perform.
A model is what is created by the algorithm as an outcome of analysing the supplied data; a representation of the patterns and relationships that exist within the data. Think of an algorithm as an intelligent entity and the data being the entirety of information and knowledge of its world. The model is its understanding of the world – interactions, correlations and causations – which can be generalised to interpret and understand new data to which it is exposed.
Input and Output
The first involved feeding data (the ‘input) that is relevant to a specific task or objective and where the relationship between variables contained within the data are known, into an algorithm which analyses the data in order to create a model. The model, which is a mathematical and statistical representation of the patterns and relationships within the data, can then be used to generalise the identified relationships to new data, where the relationship between the variables is unknown. In doing so, a prediction is generated as the outcome (often called an ‘output’).
Put simply, AI involves taking data that you do have to generate data that you don’t have. The value being that the generated output is based on massive quantities of high-quality, clean and representative data that makes it valuable to inform decisions about some unknown or hypothetical circumstances or conditions.
The second approach is perhaps more intuitive as it’s very similar to how humans learn to perform a majority of their tasks. It’s called reinforcement learning. For tasks too complex to program the machine, the machine interacts with its environment and learns through trial and error. Those actions that progress the machine towards achieving its goal are rewarded and reinforced, while those actions that hinder or don’t advance progress are discouraged. Reinforcement learning is a popular technique in applications such as robotics, where the challenges of programming a robot to navigate through a potentially infinite number of scenarios is infeasible and therefore having the robot learn is far more efficient and effective.
This brings us to the ‘underlying principle of AI’. In order to solve a problem using AI, the task must be expressed in a form a machine can understand and the machine must be supplied with the necessary data (the input) to perform or otherwise learn to generate predictions (the output) that enable it to accomplish its objective. By reframing business challenges as prediction problems and focusing on the types of predictions that help make decisions that solve business problems, we remove much of the complexity and unknowns from determining what are key priorities to be pursued using AI.
If business leaders are successful in recognising this, AI has immense capability and can do much of the heavy lifting in helping humans achieve our goals. However, it cannot do everything. In order to leverage the power of AI, we must first define and align on the goal or task we desire AI to perform before it can be developed and implemented. AI is the tool and the toolbox; not the hand that wields it, the eye that guides it or the mind that conceives the idea and outcome. For now, this is still the responsibility and domain of humans and ultimately comes down to the core ingredients required to derive success from AI – clear objectives, quality data and viable technology, the synergies between them and the standard to which a strategy to exploit the opportunity is planned and executed.