Machine Learning: Impacting Innovation Today

AI and Machine Learning (AI/ML) is not just for Google and Amazon but is beginning to be a substantial part of new products in unexpected categories, and is helping established business innovate. But even today, it is impacting how we manage teams and projects, enabling teams to be more efficient and creative. Although your job is not threatened anytime soon -- most observers believe that the first wave of AI will augment specific tasks -- it’s role will only grow from there. A great example (here today) is using Natural Speech Processing to create meeting minutes automatically the second a meeting is over. Soon we will see “agents” that excerpt action items, and “bots” that follow up (here soon). Unfortunately, this new wave of technology is surrounded by a great deal of hype. By understanding how companies are using AI/ML to innovate processes and products, teams even in mature markets, can learn how to enhance core offerings in ways that delight customers.

Unexpected Availability of Large Data Sets

Driving the AI/ML boom is the availability of large data sets, coupled with the explosion in computing power. This combination has enabled companies to gain powerful insights around consumers: their habits and behaviors, and how they will use a product or service. With the pervasive availability of smart phones, GPS, weather services, and social media, huge data sets are continually being created that can be leveraged by all organizations, not just the consumer internet giants. Data Scientists combing through the data, working with product marketing can suggest novel ways that existing products are augmented and extended to create greater differentiation.

But the overhyping of AI/ML has led to a confusion between three distinct levels of the technology:

• Data Science – Using relatively straightforward statistics such as means and standard deviations to analyze large data sets.
• Machine Learning – Systems that not only analyze data sets but also use data to inform and modify their own analytic algorithms.
• Artificial Intelligence – Systems that mimic human intelligence in a host of ways.

There are relatively few examples of “pure” AI and the examples tend to be limited to well-funded, high technology giants (and to very narrow, specific tasks). However, companies beyond Silicon Valley have made use of the more modest levels of Data Science and ML technology in novel and surprising ways.

Examples of AI/ML – Minus the Hype

Fortunately, there are emerging examples of real companies that are using Data Science and ML in ways that make a difference for customers and many of these companies are not necessarily the large, technology players. Some of the clearer contemporary examples come from the unlikely field of agriculture, one of humanity’s oldest forms of production.

For instance, John Deere acquired a start-up that developed machine learning systems to deliver a precision dose of pesticides where crops need it the most, sparing those that don’t. A report from Forbes describes how robots use artificial “sight” to identify pests in specific locations allowing farmers to spray where pests are likely to cause damage. Precision application of pesticides reduces costs for the farmer while mitigating risks for consumers and the environment. John Deere’s equipment also uses GPS technology to help in the processes of sowing and plowing, while its Farmsight System enables farmers to make better decisions based on data shared by subscribers across the globe. Data science also enables farmers to get weather data in real time. Imaging specialists have used satellite imagery to predict crop yields with great accuracy.

On a much smaller scale, Makoto Koike, a Japanese software engineer who returned home to work on his family’s cucumber farm, found an application for machine learning. Japanese consumers are choosy about the aesthetics of their vegetables. Only the most perfect, blemish-free cucumbers make it in the larger markets. Koike’s mother goes through the tedious process of sorting the cucumbers by hand, separating out the best, which go to wholesalers, from the rest, that must satisfy the local market. 

Koike taught himself about Machine Learning. He then used Google’s TensorFlow, which became open source in 2015. He adapted a feature in Google’s software that recognizes handwritten numbers. Taking photos of cucumbers from three angles, he input some seven thousand photos of cucumbers sorted by his mother. He “taught” his software to recognize the best samples and developed an automated conveyor belt that allows the system to sort the vegetables into appropriate containers. Not yet perfect, Koike continues to improve his machine learning approach to grading cucumbers.

AI/ML Improves Programs

Teams developing systems more complex and less organic than a cucumber, are using applications of AI/ML to improve the product development process. Vendors of Program Management and Portfolio Management software packages are using prior project histories to better estimate risk and to improve schedule predictability, leveraging Microsoft’s Forecasting module and other services from Google and Amazon.

Stratejos is a startup that uses clustering to monitor project discussions and identify areas that are program hotspots. From their website, “Now we're on a mission to create artificial intelligence that advances the state of managing software teams. From automating the routine work through to providing concrete, data-driven ways to improve team performance.” 

At Fujitsu, engineers used AI/ML to improve engineering speed and reduce costs in electronics design. When designing an electronic system, an engineer specifies the components and how they’re interconnected on the printed circuit board (PCB). Because of the complexity of PCBs in today’s devices, the complexity of circuits and interconnects, and the sheer number of address lines and data lines, using only the top and bottom of the board often will not do. When the complexity of the circuitry no longer fits on one layer, designers resort to creating PCBs with multiple layers.

But a side effect of increasing layers is that it adds cost. Often engineers face a tradeoff between features, quality and cost. Fujitsu used applications of ML to estimate the minimum number of layers necessary for a given PCB, using data from past designs combined with the specs of the new, proposed design. Fujitsu’s application of ML helps engineers make the tradeoff between circuit complexity/quality and cost since properly identifying the minimum number of PCB layers has been demonstrated to be a key factor in the design process.

Up from Here

Once the province of science fiction, and science fact only for the most sophisticated players, AI and ML have now come down to earth, from cucumber farming to consumer electronics. The utility and reach of AI/ML will only continue from here. It will find its way into surprising niches, transforming mature products, into exciting enhanced offerings.

The future lies in distinguishing how AI/ML a) can enhance products in ways customers can recognize and b) how AI/ML can help developers innovate and perform tasks better and faster. The future has arrived. The key to leveraging it is to separate hype from reality by looking to real-world examples, while focusing on applications that will improve products or process.