hfng - Fotolia

Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

What businesses need to know about cognitive computing systems

While cognitive computing tools have come far in utility, they still have some important gaps. Understanding these hitches may be key to getting value from the platforms.

Businesses wanting to cash in on the trends of artificial intelligence and cognitive computing need to give their plans a reality check before launching lengthy, and potentially expensive, projects.

"A lot of people are looking at cognitive computing as a bright, shining object that you can apply to anything, and problems will go away," said Hadley Reynolds, managing director and co-founder of the Cognitive Computing Consortium, in a presentation at the TDWI Accelerate conference in Boston. "But what we hear daily is that, if you don't have a very good idea of what you're trying to accomplish with cognitive projects, whatever efforts you make are likely to end in failure."

Since the twin concepts of cognitive computing and AI burst onto the scene a year or so ago, there has been no shortage of people arguing that the technologies will revolutionize business. Then, there followed a backlash of people saying cognitive computing systems are pure hype. But the truth is likely somewhere in the middle.

Cognitive computing systems still lacking

For Reynolds, the advances made by self-driving cars in the last couple of years are a case in point. Autonomous vehicles have undeniably come a long way in a very short period of time. They are now capable of traversing streets much more competently than ever before. They are also still not ready to replace human drivers on the road.

So, in this period where AI may not quite be ready for prime time, but is still, nonetheless, capable of delivering limited benefits, Reynolds recommended businesses be very selective about where they apply the technology, and think really hard about what they hope to get out of it.

According to Reynolds, cognitive computing systems are currently most effective as assistants; more intelligence augmentation than artificial intelligence. Things like chatbots and smart assistants have proven to be fairly effective. Point projects like these are a good place for enterprises to start, rather than with business-wide initiatives.

Be aware of cognitive computing's nature

Enterprises also need to be aware of the nature of the benefits of cognitive computing.

At the conference, Diego Klabjan, a professor at Northwestern University, whose research projects involve developing cognitive systems for enterprises, described a project he recently worked on for a large retailer. The point was to improve forecasting of product sales. Since the retailer had hundreds of thousands of products in its catalog, it was a complex problem that was a good fit for cognitive computing systems.

Klabjan and his team developed deep learning models to improve sales forecasts for each product. Ultimately, it worked: The team was able to improve predictions for 78% of the products in the catalog. But what's important to point out, Klabjan said, is that the magnitude of the forecasting improvement for each individual product was relatively minor. Deep learning, in this case, resulted in broad, but somewhat shallow, gains.

However, the benefits of the project were not limited to forecasts. Klabjan said that, compared with traditional machine learning projects, this one was less time-consuming. Typically, in machine learning, data scientists have to spend a lot of time selecting features they think are relevant to the problem. They'll test their models, and consult with business experts before further refining the data elements included in the model until they eventually get something that works.

Deep learning: Don't expect miracles

Deep learning, on the other hand, is much more automated. Once you have your data, and have selected the type of model you want to run, you basically just hit go, and come back when the machine has done its work.

"Don't expect deep learning to perform miracles," Klabjan said. "If you're already doing sophisticated machine learning, deep learning will give you a slight improvement. But the real benefit is you don't have data scientists spending hours on feature selection."

Next Steps

Cognitive computing has yet to deliver consistent results

Cognitive computing platforms a natural pairing for the internet of things

APIs broaden appeal of cognitive computing use

This was last published in April 2017

Dig Deeper on Artificial intelligence and analytics

PRO+

Content

Find more PRO+ content and other member only offers, here.

Join the conversation

2 comments

Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

What do you think are the biggest problems with today's cognitive computing systems?
Cancel
Not enough parallel processing power and lack in sensor precision in order to get over the next development threshold. Also, feature scaling is hard to get right. We'll get there, though.
Cancel

-ADS BY GOOGLE

SearchDataManagement

SearchAWS

SearchContentManagement

SearchCRM

SearchOracle

SearchSAP

SearchSQLServer

SearchSalesforce

Close