Brought to you by Keysight Technologies
By Jeff Harris
Some people see artificial intelligence (AI) as a panacea: a solution to every problem. This view can generate confusion about how to best deploy AI technology.
One area where AI is proving its value is the transformation of product development across industries. With the pressure to accelerate the pace of design, AI can play a transformational role. However, integrating AI is not enough; engineers must also understand how to optimise the potential.
Every AI algorithm possesses three essential elements: the ability to measure, knowledge regarding how much of what you measure needs to be processed, and the ability to process more than one input at a time. In order to optimise your AI algorithm, you need to orchestrate the data. So, let’s delve into this and explore what this requires.
The data decision
In our complex and interconnected world, there is no shortage of information; however, it’s not practice to assume you can analyse everything.
Outside of data centres where exchanges are unconstrained, decisions about what data is of high value become very important.
To deliver on the promise of the AI algorithm, there is a balancing decision to be made around the best data set. This requires homing in on the following areas:
- What data is most relevant to the decision the AI algorithm is making. This requires understanding the data that is additive and stripping out information that doesn’t directly impact the decision the AI is trying to make.
- How much data to send: it’s crucial to understand how much information to send. A good rule to follow is focusing on the minimum data required to maximise the output of the AI algorithm.
- The communications channel. It’s important to then check that there is enough capacity to transmit the data from the sensor back to the algorithm in a timely manner. You can cherry-pick the relevant data, but if you can’t communicate it in the correct timeframe, the AI will fail to reach its potential.
- Orchestrating the data is vital to maximise the impact. This requires sifting through the information to identify the critical elements and ensuring that the communication channels can relay the information in the required time period. Before integrating AI, engineers must measure the communications channel’s bandwidth, latency, and reliability to ensure it matches what the AI algorithm needs.
Examples of this in action:
1. IoT Device.
Think of a smart light switch that is tasked with determining when to turn lights off and on. To achieve this, the sensor must be able to measure solar levels to understand when sunlight occurs and make a local decision to turn the light switch off or on as a result.
This requires the IoT device to have the processor capability to know a measurement came in, understand that the value was above zero and – because the value was above zero from the sensor input – automatically make the decision to turn the switch off
2. Autonomous vehicle.
An exponentially more complex example of delivering on the promise of AI is an autonomous vehicle (AV). Every AV is a web of inputs, including radars, sensors, LIDAR, and cameras, all focused on delivering a safe driving experience. However, the importance of each input changes as the environment evolves. Think about a car driving in mountainous terrain versus the streets of Manhattan—the importance of the various inputs would differ significantly in these two locations.
For example, visual object recognition is critical to ensuring safe driving until it snows, at which point another input is more important. It’s the role of the AI algorithm to determine at any given moment which inputs to emphasise when deciding whether the vehicle should continue to go forward at its current speed or make an adjustment.
There isn’t enough processing capability to evaluate every element; therefore, the algorithm’s unique power is the ability to look at the inputs and decide how much emphasis should be put on any of them at a given time to make a decision.
These two examples, while very different, reflect the need to understand how a system works before you can measure it. You need to know what to expect from the inputs and outputs and how to get from the latter to the former. In addition, the data needs to be presented in a way that makes it easy to determine whether it’s correct or incorrect.
The new development framework
The shift to AI-driven development is transforming how we build products. To optimise the algorithm, it’s vital to understand how the system works and then fine-tune the measurement.
To do this, you must know what data to emphasise under which circumstances and ensure that the mechanism can transmit it back. Only then will AI deliver on its potential.
AI is changing the way products are developed. It mentions the challenges of incorporating AI and the importance of making sure AI algorithms work well. There are three important things that need to happen in order to get the most out of an AI algorithm: you need to be able to measure how well it’s doing, you need to know how much data to feed it, and you need to be able to give it a lot of different inputs. Picking the right data is important because there’s just too much out there to look at it all.