Start with measurement to determine the potential of AI algorithms

Jeff Harris, Vice President of Global Corporate and Product Marketing, Keysight Technologies

Artificial intelligence (AI) algorithms have three basic core elements: 1) the ability to make measurements; 2) knowing how many of those measurements require further processing; and 3) the ability to process multiple inputs in parallel.

The potential of a system refers to its measurability and the depth of measurement that can be achieved, while the realization of potential refers to deciding which aspects of the system's measurements must be sent to the processor for further processing. Finally, sensor fusion refers to understanding how to combine the measurements of different sensors in the right proportions, how high the algorithm's IQ is, and how great the potential for reasoning is, which is the key to our exploration. By enhancing sensor fusion through feedback loops, the algorithm will be able to verify and correct its own logic, which is an essential component of machine learning.

These three properties are critical to understanding the depth of AI, especially its unique capabilities. The more foundational elements we can uncover and calibrate, the better AI algorithms will perform in the long run. Now that we have introduced the three areas we will explore, let’s take a closer look at the first aspect - measuring depth, and its importance in building a robust foundation for high-performance AI algorithms.

Measuring depth

Metrology is the science of measurement. Measurement depth plays a critical role in building robust algorithms. The Gagemaker Rule (10:1 Rule) states that the measurement instrument or device must be 10 times more accurate than the object being measured. Measurement depth is so important because it determines the level of accuracy that can be achieved, limiting the maximum potential of the algorithm. Therefore, the more accurate you can be in making any given measurement, the greater the potential of the AI ​​algorithm.

Metrology focuses on gaining a deep understanding of a specific measurement. This measurement may be very simple and straightforward, such as voltage, ground, temperature, or involve multiple modes like implementing an aircraft control surface, or it may be very complex, such as maximizing throughput on a production assembly line. Whether measuring a single parameter or multiple parameters, the depth of measurement determines the degree of programmability that can be achieved. For example, if the measurement accuracy is only 1/10 V in a 3 V system, it will not provide the same insight as if the measurement accuracy is 1/1000 V. Depending on what system is powered, the extra accuracy may be critical to battery life or it may be unnecessary. To fully realize the potential of the algorithm, the entire end-to-end measurement needs must be matched to the required depth. This is true regardless of the object being measured, even for data systems that may not be so intuitive. Let's look at an example.

How to optimize measurements

The enterprise IT stack is a complex web of interconnected data systems, each of which needs to exchange information to coordinate the organization's operations. These technology stacks contain a range of software, such as CRM, ERP, databases, order fulfillment, and more, each with its own unique data formats and custom application programming interfaces (APIs). According to Salesforce, the average company's technology stack has more than 900 applications, many of which are cloud applications, and their software updates may have a ripple effect. Finding and isolating problems is like finding a needle in a haystack, and optimizing the performance of multiple cross-applications is even more difficult.

Each application in the technology stack of an enterprise will have a different responsible department, such as finance, HR, sales, marketing, and supply chain. IT will put the needs of the main organization first. Each enterprise has a specially customized workflow and integrates many applications and back-end systems. The user's journey or journey using the software will involve various paths. A single linear journey is very rare. Therefore, even if the same application is used in the technology stack of two enterprises, their mapping of all exchange points and end-to-end operation verification methods will be completely different. Applications that require artificial intelligence have therefore emerged. In this case, the measurement location may be the data input point between systems, or it may be the data exchange point and data display point within the system.

To understand how an AI algorithm works in such a system, we first need to understand how it measures data at various points in three key areas:

1. Evaluate how users interact with the application software, regardless of the operating system used. In some cases, when keystrokes are required, this also involves the use of Robotic Process Automation (RPA)

2. Evaluate the data exchange between various systems in a complex technology stack and the API commands that connect these systems to ensure they operate correctly

3. Measure screen information such as images, text, logos on all platforms (including desktop and mobile) to understand how they are presented

Regardless of the operating system, software version, device, or interface mechanism used, evaluating measurement efficacy starts with measurement capabilities. The more AI cannot measure, the less impact it will have in operations.

in conclusion

When assessing the potential of something, we need to start with the basics. The foundation of an AI system is its ability to measure. The more conditions it can measure, the greater its potential impact. We need to understand what it can measure, and more importantly, what it cannot measure. The potential of an AI algorithm is limited by its ability to sense. Lord Kelvin once said it is still true today - "If you can't measure it, you can't improve it." To understand the true power of AI, it is important to start by analyzing the breadth and depth of its measurements.