You are here

From here to AI: Beyond the hype

19 October, 2017
Kim Storin
IBM

Lessons from the Gartner Symposium

I asked over 500 people two simple questions during our recent Gartner Symposium panel:

  1. “Who is implementing deep learning or machine learning technologies in your company?”
  2. “Who thinks these technologies are drivers of innovation, but have no idea how to get started?”

The media coverage of artificial intelligence (AI) would have you believe that most people in the room raised their hands for the first question. But from my (unscientific) count, only 20 people raised their hands for #1, while close to 500 raised their hands for #2.

It is obvious that we need to help IT leaders get from here to AI. Which is exactly why we invited thought leaders and technical realists from NVIDIA, Hortonworks, IBM Cognitive Systems and CloudPulse Strategies to cut through the AI hype and give us tips on how to get started.

Here are my top takeaways from our panel session at Gartner Symposium:

Deep learning could be the easiest place to start. It’s confusing to know when to look at machine learning, deep learning or other AI technologies. The panel of experts commented that deep learning has recently taken off due to convergence of two factors. The recent explosion of data is one, along with a new generation of powerful computational engines like GPUs and FPGAs that are ideally suited to the types of highly parallel training and inferencing workloads that deep learning requires.

Have a holistic data strategy. Deep learning allows you to extract value from all of the data that you have been collecting, which is why your overarching data strategy is critical. Sumit Gupta, VP of IBM Cognitive Systems, suggested that you first articulate the business problem that you want to solve, and then start to build a list of data that you need in order to answer the question. Once you understand what you need from a data perspective, you can begin to develop a strategy to integrate data across silos. Ultimately, you can build a deep learning approach to solve the business problem.

When it comes to deployment, everything follows the data. We’ve gotten a lot of questions around when to deploy deep learning in the cloud, and when to build out an on-premises model. At the end of the day, there is no one-size-fits-all approach – advice from the panel of experts was to “follow the data.” When you have significant data on premises and/or privacy concerns, on-premises deployment models (including private or hybrid cloud) might make more sense. Other times, it will be more effective to leverage a cloud application.

Just do it! Start small and scale. Sometimes it’s easy to get overwhelmed by the hype. All of the experts on the panel agreed that it is better to start small – from keeping your data variables manageable to picking a small project – and scale as you and your team get more comfortable with deep learning.

Eager to get started with deep learning? Start with PowerAI.

The post From here to AI: Beyond the hype appeared first on IBM Systems Blog: In the Making.

Original Post