Make AI Boring: The Road from Experimental to Practical
In the same way the cell network enabled our modern mobile lifestyle, AI will drive the data age, but only if we make it boring and commonplace.
When you see headlines about artificial intelligence and machine learning they too often focus on the excitement of the technology, fueling hype and mystery versus understanding of the technical reality of what’s actually possible. On the other hand, imagine if every time we talked about a mobile app, we mentioned that it runs on a cellular network! Of course you can’t - that would be boring because the cellular network is so embedded and indispensable, we take it for granted.
It’s time to make AI boring. We need to stop spending so much effort talking about algorithms and instead focus on the outcomes of the AI systems that we’re building. Let’s make AI boring --practical, repeatable and scalable -- to drive real business results.
We’re witnessing the industrialization of AI. As these capabilities move from labs and prototypes to scaled production systems, and as organizations become capable of rapidly experimenting and iterating, we’re beginning to see tremendous value being driven. The practical applications of this technology have the potential to drive business benefits both internally and externally by saving money and creating new business opportunities.
Industrializing AI inside an enterprise requires us to look beyond projects that require a massive investment of time, people and computing power, only to create a little “wow-factor.” Instead, business leaders need to see it as an imminently practical tool to solve business problems and drive new opportunities.
One international accounting firm has already started to adopt this approach. By using inference algorithms to fill in GAAP-required accounting questionnaires automatically, the firm has been able to save thousands of accounting hours a year. After this initial success, they were able to invest in using natural language processing to highlight opportunities to engage with clients around additional guidance on tax strategies, growing their revenue.
These types of AI applications are not the “change-the-world” technologies that make headlines. They aren’t even fundamentally new capabilities. But an organization with a thousand projects like this is much more productive than one without.
Building a functional organization around AI isn’t easy, and there’s not yet one standard way to do it. You can’t simply copy what everyone else is doing. Today, each successful company requires their own plan, adapted to their specific data and AI opportunity and current structures. However, there are a few consistent ingredients.
Organizations must identify and recruit the right skills and talents, and make sure they are organizationally empowered to succeed. This is impossible without executive support. They must create a strategic investment framework that guides which projects to work on, what the exploration process should look like, and how a project moves from successful experiment to production. Processes for governance and security must be developed and adapted to be agile enough to maintain compliance without inhibiting innovation. Finally, technical tools and platforms must be selected and - at least somewhat - standardized to enable consistent technology investments across an organization.
Success will be built on bringing more fields of expertise into the AI discussion. Bringing people to the table and helping them understand AI from the perspective of the practical improvements and innovations it can deliver will move us closer to making the power of AI available to every department and functional area.
Often there is a learning curve from early experiments to large-scale applications for customers. In the case of the accounting firm, the initial success of the automated form-filling program proved the business benefits of investing in AI, but also demonstrated the firm could derive business value from the work. Without that experience under their belts, the journey to rolling out new, AI-driven client-facing services would have been impossible.
In the same way the cellular network enabled our modern mobile lifestyle, AI will drive the data age. We need to make AI boring so we can unlock its full potential.
Hilary Mason is general manager of machine learning at Cloudera.
The InformationWeek community brings together IT practitioners and industry experts with IT advice, education, and opinions. We strive to highlight technology executives and subject matter experts and use their knowledge and experiences to help our audience of IT ... View Full Bio
We welcome your comments on this topic on our social media channels, or [contact us directly] with questions about the site.