Overall, 59% of respondents said they already have some AI currently deployed. All of those surveyed who already had the average number of projects in place expect to add six more in the next 12 months, and another 15 within the next three years. Gartner predicts that by 2022, those same organizations will have an average of 35 AI or ML projects in place.
The Gartner study was conducted via an online survey in December 2018 with 106 Gartner Research Circle Members — a Gartner-managed panel composed of IT and IT/business professionals. Participants were required to be knowledgeable about the business and technology aspects of ML or AI either currently deployed or in planning at their organizations.
Lack of Skills Curbing Artificial Intelligence Uptake
The top challenges hindering respondents adoption of AI were lack of skills (56%), understanding AI use cases (42%) and concerns with data scope or quality (34%), all of which are potentially project-hampering obstacles. So is Gartner’s confidence misplaced? Is the level of adoption and use likely to rise in the way it predicts within the next 12 months?
Allan Grosvenor, CEO of MSBAI, an equity crowdfunding platform in Los Angeles with investments in many AI startups, said that while there is consistent market research and direct feedback from the industry demonstrating a need for an increase in AI adoption, too many companies lack internal expertise to do it. It is unrealistic at a large percentage of companies to just go and hire AI experts and start an AI team.
“What they don't understand is that for a lot of companies to start an AI initiative internally they are unwittingly creating a new software company that will not only develop new software once, but will have to support the adoption by others in the organization, and provide bug fixes and upgrades, as well as software support to users,” Grosvenor said.
Because of the difficulty involved in finding experts to hire, and the growing cost of hiring, many enterprises won't do it. A relatively small percentage of industries across multiple sectors will do it and succeed. For everyone else they will either wait, try-and-fail or purchase solutions from vendors (out-of-the-box AI systems, or limited engagement consulting jobs for custom developments).
Related Article: AI Deployments Depend on Human Skills Too
AI Problems Ahead
Opinions are divided on whether Gartner’s predictions are likely to hold true or not. In fact, Chris Bergh, CEO of DataKitchen, a DataOps consultancy and platform provider that manages analytics creation and operations, said not only is the number of projects realistic, they are already happening. Even still there is a problem.
ML tools are evolving to make it faster and less costly to develop AI systems. But deploying and maintaining these systems over time is getting exponentially more complex and expensive. Data science teams are incurring enormous technical debt by deploying systems without the processes and tools to maintain, monitor and update them. Furthermore, he said, poor quality data sources create unplanned work and cause errors that invalidate results.
“Funny enough, just a couple of years ago Gartner predicted that 85% of AI projects would not deliver for CIOs. Forrester affirmed this unacceptable situation by stating that 75% of AI projects underwhelm,” he said.
“We can't claim that AI projects fail only for the reasons we listed. We can say, from our experience working with data scientists on a daily basis, that these issues are real and pervasive. Fortunately, data science teams can address these challenges by applying lessons learned in the software industry.”
Related Article: 5 Tips for Attracting Artificial Intelligence Talent
Building AI/ML Models
AI is most frequently implemented using ML techniques. Building a model is different than traditional software development. In traditional programming, data and code are input which, in turn, generates an output. This is also true of traditional modeling where a hand-coded application (the model) is input, along with data, to generate results.
In ML, the code can learn, the ML application trains the model using data and target results. An ML model developer feeds training data into the ML application, along with correct or expected answers. Errors are then fed back into the learning algorithm to boost the model's precision.
The over $50 billion AI market is divided into two segments: tools that create code and tools that run code. The point is — AI is code. The data scientist creates code and must own, embrace and manage the complexity that comes along with it.
Related Article: Statistical Modeling and Machine Learning Coexist, Not Compete
AI Strategy Is Key
lava Kurilyak, founder and CEO of Vancouver, Canada-based Produvia, said their experience with developing intelligent software for enterprises have led them to three observations that dovetail with Gartner’s research:
1. AI investments are driving AI adoption
AI projects are driven by funding and access to talent. Capital allows companies, government agencies, governments, academic institutions, nonprofit organizations, national and international research consortia and businesses from multiple industries to launch AI projects and ensure the continual investment into AI technologies.
2. AI job openings are driving the number of AI projects
AI talent allows companies to scale the number of AI projects. When analyzing AI industry trends in terms of AI job openings, machine learning is the most adopted title used to create AI solutions, while deep learning is growing at the fastest rate.
3. Number of AI projects is increasing
Companies that create an AI strategy are more likely to launch multiple AI projects than companies that are experimenting with AI technologies. Over the past 5 or so years, Kurilyak has noticed that companies with a defined AI strategy launched multiple projects in order to accelerate the adoption of AI technologies.
Understating AI Growth
Many believe Gartner’s predictions are an understatement of what is likely to happen.
Scott Clark, CEO of AI, ML and deep learning solution SigOpt, said the trends they are seeing among their customers suggest AI is accelerating even more quickly than Gartner is predicting.
“Most of the growth we see in AI projects is around teams building their own proprietary models using their own proprietary data sets to solve proprietary modeling tasks,” Clark said.
Rather than buying an AI-enabled solution off the shelf, these teams are harnessing their own domain expertise and contextual awareness to build these differentiated models. This investment in modelers and the modeling stack supporting them is the biggest trend that is driving growth in the number of AI projects among these companies.
Traditional roadblocks to AI projects — like access to clean, labeled data and sufficient compute infrastructure — have largely been solved by massive investments over the last several years. Now firms are accelerating their ability to produce projects and impact with tools that augment their experts and amplify their success and impact. Clark added that AI leaders continue to scale two processes that are critical to the success of AI projects.
"First, they scale their volume of experimentation, which enables them to explore new modeling use cases. Second, they scale the use of model optimization algorithms for hyper parameter and parameter tuning to know with confidence that the models they are building are of high quality and best configured for robust performance. This process creates a virtuous cycle, where this experimentation and optimization improves both modeling productivity and impact, which in turn leads to more projects over time," he said.
As more enterprises make these investments to become model-driven, we expect more companies to have hundreds — if not thousands — of models supporting each use case, dozens of use cases to which they apply modeling, and significant ongoing growth of AI projects that are supported by a combination of software, talent and processes.