Do you have an AI Solution or is it just Smart Automation?

AI or artificial intelligence is the buzz word today. To stay at the top of the technology race, every business is either talking about AI or trying to find practical applications for AI to improve their bottom line.

However, many companies use AI interchangeably with “smart automation”. So, what really defines AI?

How is AI different from smart automation?

As technology marketing company working with several AI companies, we are often faced with the difficult task of articulating the value that an AI brings to the table. Customers want to know why they should pay more for an AI based solution, when in theory, automation can perform in a similar manner. This task of clearly articulating the value of AI would be much easier if most of our target audience were able to first define AI clearly.

In this article (and the next few articles that we will be sharing) I have taken an attempt to explain AI in simple terms. Read on to see if you have an actual AI solution or is it just Smart Automation?

Definition:

Artificial Intelligence is the ability of a machine or a computer to intelligently emulate human behaviour. To be classified as “artificially intelligent” the machine must be able to intake random information or unstructured data, infer it to take logical actions and self learn like a human.

In other words, an AI-powered machine/model can

1.      Recognize or establish pattern from unstructured data

2.      Define and Re-define the end goal and take appropriate actions to meet the defined goal and

3.      Improve its accuracy to make decisions based on the outcome from past actions, without constant input from a human.

Artificial Intelligence is best suited for environments where the input as well as the expected outcome are both dynamic, and the end-goal is not pre-defined but must be established with every move.

Input or Pattern Recognition:

Every activity requires the initial input of data. For machine to properly process the data it needs to be structured. A structured data is where the data fields are pre-defined or labeled and data is organized according to these fields/labels or schemes. Smart automation requires structured data to be effective.

However almost 95% of data in real world is unstructured. An AI-powered automation can work with the raw data. 

Almost 80-90% of life/nature in its basic form has an order. Over the year philosophers, mathematicians, artists, and scientists have spent a lot of time deducing the patterns in the nature and structuring the various symmetries.  

A human child when exposed to different attributes like sound, structure, colours, taste and texture, starts his cognition journey by first memorizing patterns and then storing it. Similarly, an AI model can also extract patterns from raw data whether it is to discern different structure, polarity, speech/text, or to understand the tone or mood of a situation/interaction.

Human child can naturally understand the patterns, but an AI must mathematically model recurring patterns. Hence machine can only recognize a pattern if it can be modeled mathematically. This limits a machine considerably. This is also the reason, why machines cannot paint a masterpiece or write an original sonnet.

Pattern recognition supports AI’s ability to Natural Language processing, Visual perception, and emotional response. A good example would be that of a language translator.

A basic language translator reminded passengers to “seal their passport, before abandoning ship” by replacing each word with the corresponding English word.

An AI based translator would have applied context to correctly read “Passengers are reminded that it is mandatory to stamp their passport before entering Morocco”

An AI-powered system can work with lesser and more flexible inputs, utilizing disambiguation (usually an AI-powered interactive system would have a two-way dialogue to get clarification and validation from the user but that’s a topic for another article…) and the available context to make some logical deduction. For e.g. If the AI has access to the area-code– it can deduce the city you are calling from or initiating the contact from without you specifying it.

The ability of an AI-powered system to recognize natural language and come to logical, mostly accurate conclusion is what differentiates an AI from a smart automation. The more context available to an AI makes it more perceptive and can process the data more accurately. This takes us to the next feature of an AI – the ability to process the data input.

What does “Processing Data” look like in AI world?

In this section I will talk about what an AI will do with the information provided to it. How does an AI infer/interpret the data and actions it?

For automation – basic or AI-powered, the first step is “goal setting” or defining the end objective. The more defined the end- goal – the easier it is to automate the function. For most automation a logical action path is defined by its creator for the automation to be able to achieve the end goal or desired outcome.

For example, in the game of Candy Crush– the objective is clearly defined – eliminate all squares or collect the defined amount of loot. A well-defined outcome makes it relatively simple to identify the logical actions required to achieve the set goal in the quickest and most efficient manner. A smart automation powered by simple logic can calculate all possible logical path to achieve an outcome and easily choose the best and most effective option.

A more complex scenario would be a game of Chess which requires a more strategic approach. Each player must anticipate the moves of their opponent and make counter moves to jeopardize their opponent’s position while protecting their own king. This is where some level of Artificial Intelligence is required to create, even a very basic automated player. The machine-powered player must anticipate the opponent’s moves and take appropriate counter actions/make moves with every turn. The ability to define and redefine the end goal and the set parameters at any given point of time is what differentiates AI from smart automation. Variable goal setting can improve the outcome of AI based automation exponentially.

Here is another great example of variable goal setting. In a bakery – a basic automation can be programmed to a specific quantitative goal - set off the alarm if the internal temperature exceeds certain limit.

In the same scenario, with an AI-powered automation the goal could be wider. A qualitative goal could be to prevent the cake from being burnt. In this scenario the AI-automation may take one of several paths to achieve the set gaol – including setting off an alarm to inform its human counterpart so they can take appropriate action, to powering down the oven to prevent the cake from burning. Most real-life situations have dynamic goals and multiple paths to reach a that goal. To be termed as “Artificially Intelligent”, the AI must be able to make decisions and solve problems autonomously, based on the information available to it.

Self Learning or Machine Learning

Finally, what sets an AI apart is its ability to self- learn. An AI can extract inferences (outcomes) with every possible interaction and store it as context for future use.

Understanding the context or the intent is essential for an AI to be able to process the data quickly, efficiently and accurately. An AI is as smart as the data it is fed.

However, the machine must be treated like a child and guided initially so it can learn by association. Inspired by the human brain or “neural network”, an AI’s ability to learn by association and not memorization fuels better, more accurate outcome overtime and truly differentiates and AI from smart automation. For e.g.

A smart automation registers the image of a cat in its entirety. If any attribute of the image is changed like the colour, the automation will not be able to recognize it as a cat.

However, when the AI first sees an image of a cat it notices individual factors (i.e. color, shape, size and sound a cat makes) and retains each of these factors or attributes in different layers of individual virtual neurons. When it sees another image of a different cat the AI examines each factor individually (size, shape, and color) and may conclude that there is an 80 percent probability that the image is a cat since the size, shape and other factors are like the previous image – only the colour is different. With every additional image it sees, the AI improves its accuracy by registering individual attributes or factors that constitute the image rather than treating it as a complete object.