Gartner Blog Network


Where have all the AI flowers gone?

by Tom Austin  |  June 29, 2018  |  Comments Off on Where have all the AI flowers gone?

Have you put a real killer application that exploits AI into volume production use?

I didn’t think so. As of last year, only 4 (that’s FOUR) percent of 3,182 CIOs world-wide report they’ve put an AI-related application into production (or planned to do so within the next 12 months.) CIOs don’t always know everything going on in the enterprise, but this number most likely isn’t off by more than a factor of 2. (And maybe eight percent of enterprises have such an application in production, but 8 percent is probably an overestimate.)

Why not?

On 27 June this year, we published research for our clients on AI Technical Maturity for Enterprise Architects and Technology Innovators. (This research is focused on the maturity of AI technology, not the maturity of enterprises in the area of AI technology.)

On the face of it, there has been breathtaking progress on AI in the last decade: The science of AI continues to blaze new, highly valuable trails. We have made great progress this decade in terms of AI-related academic research, conferences, graduate programs, startups formed, venture funding raised, corporate M&A activity, postings for AI-related jobs and patent applications.

But the story is incomplete:

  • We suffer from an excess of great research findings, so much so the technology space is extremely turbulent. Too often, today’s breakthrough may be obsolete next quarter or year.
  • System engineering guidelines (and expertise) are lacking.
  • AI technology today is about as mature now as ICT was around 1960. This will not soon be fixed.
  • Worst of all, massive production adoption is nearly at a standstill because we lack new, AI-enabled killer applications to inspire business people to open their main investment spigots.

Let’s focus a bit on the killer application vacuum.

When we ask IT leaders and business executives about the applications of AI and solicit textual (verbatim) answers, results often fit into four different categories:

  1. Decision support/augmentation — helping people be smarter
  2. Virtual agents — conversant in text or speech with users
  3. Decision automation — task automation or optimization
  4. Smart products — embedded AI

These categories (save number 2) are analogous to the marketing of automobiles as “horseless carriages” — early 20th century products shrouded in 19th century thinking. People can more readily envision things in known contexts.

Thus, for example, we have:

  • Business intelligence metaphors from the 1990s and earlier (as in “decision support/augmentation”).
  • Task automation or optimization under the label of “decision automation,” which is what we’ve been doing since the beginning of the computer era.
  • Smart products — an overwhelmingly clichéd label with little real meaning.

Yes, there’s interest in virtual agents. Indeed, two thirds of our clients who report investing in AI-related projects cite “customer facing” (often chat-related) projects. But those are really difficult to do well at scale unless they’re very narrowly conceived. No one outside of the biggest of the big tech companies are fielding “know-it-all bots” that can answer everyone’s question about everything (Insight Engines are better than chatbots for these purposes) and none of those big-tech offerings are really that impressive, at least not yet. (Google’s Duplex and Amazon’s Alexa Challenge are arguably attracting the most intellectual chatter but it remains to be determined if these are really going to move enterprise-investments en-masse any time soon.)

It’s hard to envision the future! Beyond conversational agents, all we hear about from clients are but improvements in horseless carriages! Where are the breakthroughs that are big enough to get enterprises ready to create major new business initiatives that depend on AI technology?

We (the industry, the vendors, the analysts, the consultants and the enterprises of the world) have not figured out what they are!

Part of the problem is the problems to which AI is best suited may be outside the intellectual experience of people trying to find the new killer business applications.

Back in the early days of commercial computers (mid-20th century), businesses bought computers to run applications people had sorted out on paper, for centuries. We knew how to do paper-based accounting systems and it was relatively easy to relate to the idea of porting that logic to a computer.

Today, in the early days of production applications of AI-technology, we are bereft of a natural human understanding of how we do things (unconsciously.) In the research note, we say:

“We can now use DNN-powered systems to identify people from pictures. Yes, humans (and our primate ancestors, among others) have been doing facial recognition for a long time — at least 50 million years — but, in general, we have no effective, systematized ways of doing facial recognition.

We just do it as humans (not as technology developers), using methods that rely on many different “hardwired” as well as learned circuits in our nervous system. How we really do it is opaque. How does a 15-month-old distinguish between her mother’s and father’s visual images? We don’t know. Our normal, human experience gives us little insight into how to build a technology for doing it.” 

Where we lack this historical precedence, we lack the practical experience needed to drive application ideation or development. (And anthropomorphic thinking may lead us down the wrong path.)

Despite these limits:

  • The science will continue to advance at a breakneck, turbulent rate.
  • AI will be embedded in more and more products.
  • By the end of 2020, AI technology will be present, under the covers, in virtually all new software products.
  • Vendors will use these embedded technologies to enrich and expand their product features. (And the vast preponderance of business AI uses will be bought by, not developed by, enterprises.)

And while we wait for people to discover the killer applications that will drive major new business investment cycles, we’ll continue to make small and narrow investments in practical, tactical implementations where they add real value to the business.

With adequate planning, you should be able to do likewise. Now. Inside the note, we dig into the technology maturity issues and provide advice on how to succeed despite these limitations.

Additional Resources

Category: ai  business-strategies  strategic-planning  technology-and-emerging-trends  

Tom Austin
VP & Gartner Fellow
20 years at Gartner
41 years IT industry

Tom Austin, VP, has been a Gartner Fellow since 1997. He drives Gartner's research content incubator (the Maverick Program) and is leading a new research community creating research on the emerging era of smart machines. Read Full Bio




Comments are closed

Comments or opinions expressed on this blog are those of the individual contributors only, and do not necessarily represent the views of Gartner, Inc. or its management. Readers may copy and redistribute blog postings on other blogs, or otherwise for private, non-commercial or journalistic purposes, with attribution to Gartner. This content may not be used for any other purposes in any other formats or media. The content on this blog is provided on an "as-is" basis. Gartner shall not be liable for any damages whatsoever arising out of the content or use of this blog.