On the Path to Pragmatic AI – It’s More Realistic Than You Think

December 22, 2017

While a lot of the buzz in artificial intelligence (AI) is centered around pure or open AI – human-like machines that look, speak and react like people – the real benefit today lies in Pragmatic AI. In fact, this type of solution, designed to solve specific real-world problems, is really the only viable AI option around today. And, while it may not get the same hype or fanfare as its pure AI cousin, the business benefits that Pragmatic AI can deliver cannot be denied.

With Pragmatic AI apps, organizations can sort through huge amounts of data quickly, and benefit from increased productivity, faster decision making, improved customer service and more.

Working Hand in Hand with Humans

Rather than trying to supplant humans, Pragmatic AI augments human capabilities through machine learning apps that handle discrete tasks.

For example, developing an app to help commuters more accurately predict when their buses will arrive requires a combination of human and machine collaboration. First, humans need to sort through a massive amount of satellite images and identify which ones are buses.

A data scientist uses large quantities of this data, along with custom-designed algorithms, to “teach” deep learning apps to “see” buses. Once this trained app is connected with live satellite feeds, it can then predict the bus schedule in real time.

Sorting through vast amounts of data and images, labeling and cleaning them are critical to the development of accurate algorithms. Some firms employ a dedicated private crowd, or group of data specialists, whose sole job is to do this sorting to expedite the process of preparing the large amount of clean, extremely precise identification of specific objects and data needed to ensure the highest quality algorithms.

Pragmatic AI is being used in numerous industries — to help call centers provide a better customer experience, predict defects in medical products before they go to market, identify the latest cancer treatments to meet a patient’s needs, improve the care of diabetes patients and much more.

The Challenges

With the significant business benefits that Pragmatic AI delivers, what’s holding organizations back from embracing it right now? There are several challenges that need to be addressed:

  • Knowing where it’s needed. Before organizations can implement a Pragmatic AI solution they need to figure out the specific problems that can benefit from Pragmatic AI. What are the key performance indicators (KPIs) that they want to measure? Where would AI fit in the business workflow?
  • Making sure they have the right data. Without accurate data, companies are not going to get good outputs. Consider a company that wants to reduce some of the calls to a call center using a virtual agent or chatbot. To create an effective solution, the company needs to accurately identify the top problems callers have so they can program the chatbot to proactively address them. If they don’t capture this information accurately, it will not do much to help alleviate the volume of calls.
  • Finding the right people for the task. It’s no secret that there’s a shortage of data scientists and data engineers. Most companies don’t have these resources in-house so they need to look externally to find them.
    Despite these challenges, there are clear steps that companies can take to implement Pragmatic AI in their organizations.

Here are some initial first steps they should take:

Find the right partner. Pragmatic AI requires advanced skills, experience and the right resources. The partner should have a team of data engineers and data scientists with the advanced skills to handle the complexities of deep-learning algorithms. Some firms are offering specialized “Insight-as-a-Service” practices, leveraging their expertise in a particular area and best practices gathered from successful AI-based projects to address a company’s AI needs more quickly and accurately.

In addition to having a strong team, the right partner also must have the right resources – such as specialized GPU-based servers. The massive computing power these supercomputers offer enable users to rapidly train extremely large deep learning datasets in as little as two hours, instead of the hundreds of hours it would take on a CPU-based system.

Get the data in order. A good pragmatic AI solution is only as good as the data, so companies must ensure that they know where all their data is, clean it and centrally store it.

Also, it’s important to consider that developing an app is only the beginning. The data and subsequently, the algorithms, need to be continually refined and re-trained, requiring ongoing work.

Pragmatic AI is very do-able, and organizations can begin seeing results very quickly. It can be implemented in small, incremental steps by identifying specific problems in key areas and addressing them. By augmenting human activities with machine learning, all types of companies can work better, faster and smarter – and position themselves for significant competitive advantage.

 

Leave a Reply

  • (will not be published)