The divide between haves and have nots in experimenting and achieving results with machine learning is growing wider. At least that’s my perception after spending two days at the O’Reilly AI Conference in New York last week and diving into the latest survey results from their AI Adoption in the Enterprise.
The haves are clearly the technology companies. Facebook, Twitter, Salesforce and others shared significant details on what the of problems they were solving with machine learning and their efforts to standardize and scale their machine learning practices.
Technology suppliers also demonstrated their latest capabilities and enterprise offerings with Intel AI, Microsoft Azure, and IBM Watson leading the charge. I reported last year that deep learning was more accessible to mainstream enterprises and at this year’s conference, small and large vendors offered a mix of data science platforms, dataops frameworks, and data management tools to help enterprises start and mature experiments in machine learning.
AI investments: many enterprises are lagging in
But from what I can tell, many CIOs appear to be playing a wait and see game when it comes to machine learning and artificial intelligence. A recent Gartner survey had only 37% of respondents saying they were investing in artificial intelligence. In the AI Adoption in the Enterprise survey, Ben Lorica and Paco Nathan share their own findings that many enterprises are still lagging in adoption.
Survey results show that AI is dominated by tech financial services, healthcare, and education representing fifty-eight of survey respondents. All other industries including telecommunications, media and entertainment, government, manufacturing, and retail were each under four percent in survey respondents.
Even in the top industries, only a small percentage of respondents reported having mature practices. Technology was highest at thirty-six percent and over fifty-percent of respondents in the top industries reported that they were in evaluation stages.
CIOs have several AI hurdles to climb
The survey points to many difficulties with experimenting in machine learning and likely give CIO a pause before making it a top priority for research and development.
- Enterprises have several prerequisites to address before investing in AI. Twenty-three percent of survey respondents state that their company culture does not yet recognize the need for AI and nineteen percent lack data or have data quality issues.
- Fifty percent of AI projects were reported to be in research and development followed by customer service, IT, or operational use cases. To conservative CIOs, investing in experimental AI may be a second or third choice in driving customer experience or operational improvements versus other more proven strategies and tactics.
- AI requires hiring a multi-disciplinary team of machine learning modelers, data scientists, business analysts, data engineers, and infrastructure specialists with survey respondents reporting skill shortages across all these skills.
- There are clear technology risks as there are no clear winners and losers across tools. Many AI practitioners are using multiple tools and while TensorFlow, scikit-learn, Keras, and PyTorch remain the top four, survey respondents listed ten other tools that that they are also using.
- Even beyond selecting technologies, artificial intelligence has a whole new set of practices that require maturing with model visualization, automated training, and model monitoring sited as the top three by respondents.
- AI still has many business risks with non-trivial mitigation strategies. Top risks sited include unexpected outcomes and predictions from models, model transparency, bias and ethics, model degradation, privacy, safety, reliability and security vulnerabilities as top risks.
Why CIOs should not delay AI experiments
Despite all of these hurdles, enterprise CIOs are taking great risks by not getting their feet wet in machine learning and artificial intelligence. AI is not like web 1.0, mobile, social, and cloud computing where laggards may have been penalized for coming late to the game but could catch up by investing in the right technology platforms, adopting best practices, and partnering with skilled service companies.
The issue is that investing in AI has three key prerequisites that require the CIO’s leadership. Organizations need a defined data strategy, the ability to execute with new technologies, and an organizational capability of change management and driving culture change. These are all foundational capabilities of digitally-native companies and remain works in progress for many enterprises investing in digital transformation.
CIOs embracing digital transformation should add AI experimentation to these programs. It’s one of the ways to deliver business benefits to the data, technology, and organizational change activities that all CIO should already be investing in. By adding AI and machine learning to the scope, CIO can start getting a better picture of the potential business benefits, competitive threats, and operational risks of applying AI in their industry. Without this research and development, organizational learning lags and CIOs may find a growing moat between their capabilities versus competitors that invested earlier.
How to drive AI experimentation
CIO have an obligation to ensure that their organizations do not fall too far behind their industry’s adoption of machine learning and artificial intelligence capabilities. That doesn’t necessarily mean making significant investments in new technologies and skills right away. Instead, CIO can start addressing some of the prerequisites by taking on these responsibilities
- Investing in organizational learning so that business and technology leaders are more aware of what’s happening with AI across industries. That means getting beyond the hype and marketing technology companies are showcasing as they are clearly leading industries in AI benefits and capabilities. CIOs should look to send business leaders to AI conferences, and have high performers with technology and data skills trained on machine learning.
- CIOs should organize lead blue sky thinking and sponsor machine learning proof of concepts around where machine learning can offer the most significant business benefits. It’s through these gatherings and tests that additional investigation and exploration can be sanctioned against the most promising opportunities.
- CIOs should lead proactive data governance efforts. The oil for all machine learning programs is a volume of well-defined data with low data quality issues. It’s a program in itself to catalog data sources, profile and cleanse data, educate analysts on data assets, and make the data infrastructure available for machine learning experiments.
These efforts all constitute low risk, humble beginnings to a machine learning program but all drive additional benefits for a growing number of organizations that need to compete on customer experience, automation, analytics, and technical capabilities.