Editor’s note: 2017, the year of artificial intelligence? Could be. Advances in technology enable organizations to treat AI less like an expensive, long-term research endeavor and more like a springboard for agile product development and business model evolution, says Technology Business Research Analyst Jennifer Hamel.

HAMPTON, N.H. – Big data, accessible algorithms and advanced computing power smooth the path to enterprise AI adoption A confluence of factors make 2017 the year artificial intelligence (AI) will reach an inflection point in enterprise adoption.

Advances in technology enable organizations to treat AI less like an expensive, long-term research endeavor and more like a springboard for agile product development and business model evolution. These advances include the availability of big, rich, trained data sets; democratized access to robust machine learning algorithms through open-source communities and proprietary APIs; and innovations in computing hardware and system architecture. Enterprise IT customers remain more skeptical about the promise of AI than vendors, but most accept AI will factor into their business futures. With a growing array of use cases such as automated, yet human-like, customer service, accelerated diagnoses of complex and deadly medical syndromes, autonomous vehicles and machine-generated investment recommendations, AI is becoming a core element of digital transformation initiatives.

As the technical feasibility and business utility of AI come more into focus, the stage is set for AI commercialization. However, a significant gap remains between customer expectations and reality that must be closed for the AI market to flourish.

TBR attended business and technology learning content company O’Reilly Media’s second O’Reilly AI Conference, a meeting of minds around ideas spanning AI research, technology innovation, business implementation and cultural impact. True to its theme, “Putting AI to Work,” the event went beyond the flashy solution showcases present at many vendor conferences (though there were plenty of impressive case studies), providing clear and practical advice for companies to start working with AI technologies and featuring candid conversations about the challenges and pitfalls of AI adoption.

Designed for an audience of data scientists, software engineers, business strategists and C-level executives, the conference included high-level keynotes by leaders from event sponsors such as Google, IBM (NYSE: IBM), Intel Nervana and NVIDIA (NasdaqGS: NVDA) and tactical sessions such as a step-by-step instructional guide to building applications using Google’s TensorFlow machine learning library. TBR also interacted one-on-one with several presenters, including founders and senior data scientists from AI-related startups such as Bonsai and H20.ai, throughout the conference. Attendance at the conference doubled compared to last year, and the second day’s keynotes culminated in the announcement of a partnership between O’Reilly Media and Intel Nervana to co-present future AI conferences, which we see as a sign the event is becoming a key industry forum for applied AI.

Implications for customers

Though technology vendors make it easy for companies to get started in AI through APIs and open-source building blocks, the most difficult aspects of adoption discussed during the conference had nothing to do with technology.

  • Use cases: Multiple speakers noted that companies considering what their AI strategy should be are approaching the problem in the wrong way. With incessant marketing around AI-enabled solutions, from IT vendors and vendors in non-IT industries such as manufacturing, it is tempting for companies to pick up the AI hammer and look for nails. However, given that current applications of AI are still very narrow — that is, single-task-oriented, as opposed to generalized AI, which is capable of tackling any intellectual task similarly to a human — companies will be better-served by gaining a clear sense of what they want their businesses to look like in the future and only then consider where AI might fit to accelerate transformation.
  • Data challenges: Training an AI system — particularly one that relies on image recognition or natural language processing — tends to require massive volumes of high-quality, labeled data to provide enough examples from which the system can learn. If an organization is struggling with analytics due to data quality, integration or accessibility issues, the leap to AI will be massive. AI-related innovation increasingly comes from lines of business, spurred by demand for speed, competitiveness and process transformation, which will further strain IT and data science resources. As a result, new vendors such as Mighty AI and CrowdFlower have emerged to help companies outsource the more monotonous, yet highly important, tasks to collect, annotate and organize training data sets, enabling data science teams to focus on highervalue activities such as modeling and tweaking algorithms.
  • Culture: Another theme that permeated multiple sessions and discussions was the cultural and operational shifts AI introduces into organizations. Working with AI means dealing with uncertainty and probabilities as the system learns, which can be extremely difficult for companies that expect absolute results from and causal explanations for business decisions. Enduring through this process while continuing to run a business takes commitment from top to bottom of an organization. Additionally, many companies still treat AI as a research-driven initiative rather than integrated with product engineering and development. During his keynote, Dr. Richard Socher, chief scientist at Salesforce, called on companies to “close the loop” by bringing insights generated from products back to research to continually improve AI programs. Peter Norvig, director of research at Google, reiterated the need for organizations to revamp their traditional product development processes to suit the unique needs of AI, including shifting the developer’s mindset from one of programming, debugging and patching applications to one of teaching and allowing AI systems to learn and solve problems independently.
  • Ethics: Debate over the implications of AI for public policy and corporate ethics, particularly related to fears of widespread job losses due to AI’s disruptive impact on traditional labor models, has intensified over the last six months, and these discussions came into focus throughout the conference. Proponents of AI technology (including the sellers of this technology) counter these concerns by insisting that AI systems will augment the effectiveness of human workers and create new types of jobs. TBR observed at least one new job type directly attributable to AI, that of data annotation microtasking through crowdsourcing platforms such as CrowdFlower and Mighty AI’s Spare5 application. These new jobs exemplify the ongoing shift in the labor market from full-time employment to short-term contracts (otherwise known as the “gig economy”), which TBR intends to explore in a future special report. Other ethical topics included the potential for discrimination or other unintended negative outcomes driven by poorly trained AI systems. In multiple one-on-one conversations, TBR asked AI vendor representatives how their company plans to address issues of implicit bias in data used to train AI programs. The consensus was that, at the moment, AI technology is getting better at detecting bias but not at preventing it.

Next: Implications for vendors