You are here:  Home  >  Data Education  >  BI / Data Science News, Articles, & Education  >  BI / Data Science Blogs  >  Current Article

Seven Things Artificial Intelligence Won’t Do

By   /  March 9, 2018  /  No Comments

Click to learn more about author Kimberly Nevala.

Despite the amazing potential of Artificial Intelligence (AI), there are some things it just won’t do. If not properly addressed, these considerations can become immovable barriers to AI adoption. As you may already suspect, they have very little to do with the technology itself or the availability of scarce experts which is – in and of itself – a formidable obstacle.

So what won’t AI do?

Justify Itself. Yes, AI is being wielded in increasingly public, visible ways. Yes, value propositions and practical proofs abound. This doesn’t mean your organization is ready to jump headlong into AI, no matter the prospective value.  Despite AI’s potentially titanic effect on business-as-usual and the fact Artificial Intelligence has been decades in the making AI is, for all practical purposes, an emerging technology.

If your company has a means to support innovation (be it a lab or R&D programs for speculative business development), you have a head start. If not, you may need to tame your enthusiasm for green field projects and wholesale redesigns. Instead, target well understood and bounded problems (hint: existing, tactical, operational) with clear, justifiable ROI as an initial proving ground.

Explain Itself. How Artificial Intelligence reaches a conclusion is often ambiguous, at best. In most cases, solutions – particularly those based on Deep Learning and other Machine Learning (ML) techniques – remain black boxes. Numerous techniques are being explored to allow AI’s internal logic to be modeled and visualized. Even, in future, for AI solutions themselves to explain how they got from A to Z. We aren’t there yet. Unambiguously assessing and addressing your organization’s willingness and ability to act in the face of ambiguity is critical.

Perform Flawlessly. Make no mistake, AI is human too. The best trained and tuned AI solution will make mistakes. Be they big or small, acknowledging the limitations of AI allows the appropriate interaction model and guardrails to be deployed. Thereby ensuring material mistakes don’t result in mission critical failures or negatively impact key customer relationships. In some cases, this means AI needs to serve as a collaborator or expert advisor, assisting the decision maker. In others, AI may function autonomously – making independent decisions and directly interacting with other systems, end users or customers.

Solve Your Data Dilemma. Even organizations with robust data practices and readily available sources of “good” data can find it difficult to capitalize on AI’s potential. Woe be to those with less honed data chops. Especially as executives attracted to AI’s shiny promise are often not inherently aware of the requisite investment in data.  Or, perhaps more accurately, ready to pony up that investment. If you’ve historically struggled to make the case for data, AI is a great platform to reignite the cause. Just don’t expect the requisite investment to be a de-facto fait accompli.

Make Change or Risk More Palatable. Artificial Intelligence is seemingly present everywhere and nowhere. Despite its popularity, most companies have limited experience with AI. Given AI’s relative newness, intrinsic complexity, and ongoing discussions of ethics and job impacts, the reality and perception of risk is high.

If your organization is risk averse, wedded to existing roles or processes or has a tenuous relationship with its internal or external constituents (i.e. customers/employees/partners) who will be directly affected by or interact with the new AI implementation, proceed cautiously. If, of course, you are allowed to proceed at all. Underestimating the effort required to address the softer side of change can undermine the entire AI effort.

Become Your Analytical Swiss Army Knife. Practical AI solutions are still, by and large, specialized. Once trained, they do one thing, in one context, very well. Don’t expect to create an algorithm that can be replicated in cookie-cutter fashion and applied to other problem domains overnight, if at all. This is true even for packaged chat bot applications which must be explicitly trained to reflect the voice, tone and content of each enterprise or department they join: just like a new employee. While experience with AI certainly makes subsequent projects easier, it doesn’t make them easy, short or cheap.

Make the (Really) Big Decisions. Just because (AI) could, doesn’t mean (AI) should. In other words, AI doesn’t absolve human decision makers of thinking about the broader implications – positive or negative – of letting Artificial Intelligence loose or acting on found insight in a given situation. Human imagination is also required to envision a future course unfettered by past patterns, pre-existing product or service definitions or engagement models. What AI can do, and do very well, is automate and make the hum-drum more effective and efficient and provide the means to make visionary ideas practical.

About the author

As the Director of Business Strategies for SAS Best Practices Kimberly balances forward-thinking with real-world perspectives on business analytics, data governance, analytic cultures and change management. Kimberly’s current focus is helping customers understand both the business potential and practical implications of artificial intelligence (AI) and machine learning (ML). Follow Kimberly and SAS Institute at: LinkedIn, Twitter, Instagram, Facebook, Google+

You might also like...

Three Tips for Successfully Adopting Machine Learning Technologies

Read More →