Artificial intelligence is hardly a new concept, but in many ways it shines best for enterprise IT automation. AI is certainly not plug-and-play, yet it remains largely misunderstood.
About half of enterprises are using some form of artificial intelligence (AI), mostly in their off-the-shelf IT automation software—but that doesn't mean they are doing it right, analysts from IDC said.
Artificial intelligence, which some companies may call cognitive software or machine learning, goes back to research on chess-playing computers of the 1950s; now it's best known for powering virtual assistants through your smart device. AI really earns its keep by helping IT staff keep their information infrastructure running.
SEE: Quick glossary: Artificial intelligence (Tech Pro Research)
IDC analysts Peter Rutten and Ritu Jyoti talked with TechRepublic about how to do AI right.
"There's a lot of confusion and a lot of misinformation out there. AI is hype... we're trying to be as realistic and empirical as possible about it," said Rutten, who covers the hardware side. The combined market for AI servers containing technology such as application-specific chips, field-programmable chips, graphical processing units, and processors with hundreds of cores will be $22 billion by the year 2022," he said.
"What we've seen is that companies find someone within the company who is somewhat knowledgeable on the subject. They will typically ask them to create a little team that starts exploring what the opportunities are," Rutten explained. "Medium-sized and smaller companies will typically have a champion who maps out what the opportunities are, what the competition is doing... and essentially create a map of 'This is what we could or should be doing,'" which could involve developers, line-of-business experts, and data scientists, he said.
Rutten said the most common mistake he sees when companies want to test the AI water is they dive in without proper teams and use underpowered components—virtual machine partitions or small clusters won't necessarily be enough. "A lot of businesses today vs. 12 months ago realize you can't start AI on just any infrastructure," he said. Specific problems include I/O limitations, data models that are too large, and processing that's too slow. "At that point there's typically sort of a realization that we need to figure out what the infrastructure is that we need for our AI efforts," he said. "A lot of companies go through a trial-and-error culture."
On the software side, Jyoti emphasized six critical tips for enterprise AI projects in a presentation at the IDC Directions 2018 conference in March:
- use public cloud services because they're more scalable than your company's own infrastructure;
- build your team around the project requirements—don't build the project around the employees you happen to have;
- bring in experienced AI consultants;
- see if your software vendors already AI-enabled any of your applications;
- adopt AI in tiers such as prediction-level results first and full automation second; and
- establish change-management organizations.
SEE: How to implement AI and machine learning (ZDNet special report) | Download the report as a PDF (TechRepublic)
While only half of big companies are currently using AI for IT automation, that figure will increase to 75% by 2020, Jyoti said. Most of today's AI tasks are straightforward, such as alerting human overlords when a hard disk is nearing maximum capacity, or when the disk fails, or when CPU processes are coming apart at their seams—that kind of AI is not unlike having a robotic IT intern. The technology truly helps when it reports and automatically fixes complicated issues such as a performance problem on an Oracle database running in a virtual machine—humans might blame each other's department, but software has no biases and can drill down into CPU states and network/storage settings, she noted.
Not that AI doesn't have its limitations, Jyoti observed. The 'A' still stands for artificial, not actual, intelligence. Getting the most from your AI dollars requires that you spend time and effort ensuring that your data is of high quality; otherwise, you'll suffer from garbage-in, garbage-out, where your AI system concludes that everyone else jumped off a bridge and thus you should too.