As I’ve been discussing in recent posts, artificial intelligence (AI) is changing the world as we speak, in part because of its ability to humanize technology. Another reason AI is having such an impact is that it’s becoming more natural and powerful.
Interacting with AI has become so natural to us that we no longer think of search algorithms or recommendation engines as AI at all, though we once did. AI systems are incredibly powerful too. They are learning fast, with Google Home improving its word recognition error rates 50 percent in just a year, and Alpha Go Zero autonomously teaching itself the board game Go—and quickly surpassing all humans and even its cousin, the Alpha Go system that originally beat a human champion.
For enterprises looking to harness this power, a good place to start is understanding past failures. Here are some of the big ones:
Not understanding what problems AI can address
We’ve long struggled with this. After all, we thought chess was a really hard thing to do, while identifying cats is something a toddler can master. But Deep Blue beat Gary Kasparav in 1997, and machine learning systems have only recently gotten got good enough at object identification to handle the many different angles and lighting scenarios that come with real life.
Overstating its potential impact
Hyping the future doesn’t necessarily lead to progress, as fans of jet packs, flying cars and robot butlers can tell us.
Mistiming its readiness
Experts have been saying “AI will surpass humans within a decade” every decade since the 50s, it seems.
Failing to measure return on AI investments
Unbelievable though it may seem for cost-conscious enterprises, 31 percent admitted to experimenting with AI but not tracking a return on investment.
Giving AI the compute power it needs
The more powerful AI grows, the more compute power it needs. Training for deep learning systems, for instance, involves mathematical vector operations that often result in the need for thousands of cores. Luckily, this is one misunderstanding we can do something about.
Powering Successful AI
Enterprises have recently had to upgrade systems to deal with big data in much the way they’ll have to upgrade systems to deal with AI. Success in solving the first challenge will help with success in solving the second: the large Linux clusters running big data solutions like Hadoop or SAP HANA are the same kind of Linux clusters that can become the foundation for your AI solutions.
By adopting to a high-performance computing (HPC) environment, you can give AI solutions the computational power they need. And building that system isn’t as hard as you might imagine.
SUSE Linux Enterprise for High Performance Computing delivers:
- Reduced pricing with a simple, one price per cluster-node model, making it even more competitive.
- The most popular HPC functionality for parallel computing environments, supported by SUSE. This includes packages like slurm for HPC cluster and workload management, performance monitoring, high performance message passing, and scientific and mathematical libraries. Plus, there are several packages that can be obtained through SUSE Package Hub – like Singularity for containers in HPC.
- Revised terms and conditions that reduce the restrictions on what qualifies as an HPC environment, defining additional types of workloads and reducing the clusters required to adopt HPC from 64 to 4.
- Extended and long-term support options are available as well.
- The product runs on two popular architectures for supercomputers – x86-64 and ARM 64 (arm-based supercomputers are becoming more and more popular due to their lower cost).
In the end, SLE HPC enables easier adoption of HPC across various industry scenarios that are invoking new waves of high performance workloads, and who are building in-house HPC environments for AI/ML and advanced analytics workloads.
Check out the High Performance Computing solutions from SUSE at https://www.suse.com/programs/high-performance-computing/ .
Jeff Reser, SUSE HPC
 “Artificial intelligence is getting more powerful, and it’s about to be everywhere,” Vox, May 18, 2017, https://www.vox.com/new-money/2017/5/18/15655274/google-io-ai-everywhere
 “Alpha Go Zero: Learning from Scratch,” DeepMind blog, Oct 18, 2017, https://deepmind.com/blog/alphago-zero-learning-scratch/
 Outlook on Artificial Intelligence in the Enterprise, Narrative Science, 2018.
 Forrester Deep Learning: An AI Revolution, Forrester, May 12, 2017.