Artificial Intelligence is finally starting to match the hype of years past. So, for those just now dipping into the AI waters for evaluation of its potential, or to enhance your existing strategic capabilities, this post may just help. This post is about adoption of deep learning AI.
First, let’s talk about a typical set of components that make up deep learning. The core of AI is pattern recognition supported by very large datasets. Patterns are resolved from the supplied datasets as well as by feedback from human users of the system. The output can assume many forms such as feeding into other systems or made human consumable. The output can be in a format that best fit the requirements.
The terms automation and recommendation are broadly defined here. For example, automation could mean a set of steps that execute a desired outcome such as building an image, talking, or running a machine. Recommendations could be the next steps in solving a specific problem, targeting specific consumer groups, or even predicting the next time an earthquake occurs (early days).
The graphic above attempts to concisely represent the layers of a simple deep learning AI stack- this is where data and training would result in recommendations and potential automation of your processes.
Outlook for AI Over the Next Five Years:
As you may expect, the leaders and early adopters in AI are Transportation, Logistics, Insurance, and Consumer-related industries, while a handful of other industries are still catching up. What’s common? You guessed it: pattern recognition. Patterns directly impact the bottom line for these industries. As you can see the value proposition for these companies are hung on new offerings and internal processes.
Currently, the bulk of activity for ramping up AI regarding Internal processes amounts to standing up proof of concept applications. Even with POC’s, independent consulting firms stand to make money down the road.
Adopting AI:
The following graphic depicts a maturity model not unlike traditional software.
Use cases for AI are based around decisions - the decisions you make each day, month, and year. Be mindful of those decisions. It may be obvious to you, for example, if you’re a marketing company that any decision you make will clearly impact customer engagement and the bottom line. For example, “What demographic should I target and when?” Resolve your business cases.
As with any software, separate the need to improve internal processes from potential new product or service offerings. A proof of concept (POC) could start with a high-value internal process before committing to a larger AI project. However, just because something could benefit from automation does not mean AI is the correct solution. Training and pattern recognition is a step up from business workflows. Prioritize your business cases as candidates for your project.
The next immediate concern is about data:
Product or Platform:
On the product side, one example of AI is Amazon Echo for business: see here.
The Echo device is fully capable of controlling any number of external devices via voice commands. The Echo device is extensible through software API’s. Amazon Echo is billed as the intelligent assistance for business.
On the platform end is the idea of AI as a service. Companies such as IBM and Microsoft offer programming interfaces that cover a range of capabilities. Incorporating these API’s into your business involves a modest investment in technical capability but the tradeoff is specialization of AI to your requirements. Some of the general capabilities offered are:
IBM -
Lighthouse Computer Services is introducing customers to AI by building proof of concept applications. One such example is based around a decision support platform for higher education, called Minerva Cognitive Solutions, developed by Itzik Maoz. Read his latest blog, Applying Cognitive Computing for HIgher Education Course Selections, for more information.
It’s an interesting time we live in. I can’t wait to see where AI take us in the future.