Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.
A Realistic Framework for AI in the Enterprise
Amid all the marketing hype and buzz around AI, and the complexity of implementing it at scale, here's a practical high-level framework that could help enterprises with their programs.
4 Min Read
Image: WrightStudio - stock.adobe.com
The year 2019 may have been a bit of a disappointment for enterprises in terms of their efforts to deploy artificial intelligence at scale. While executives in the C-suite remain committed to the idea of AI, and many organizations have created successful pilots of some form of AI technology, getting from pilot to scale has proven a greater challenge.
How do you get from here to there?
A new report from Lux Research offers a view into the state of AI in the enterprise today, perspective on its history, and some practical advice on a programmatic framework for how to think about AI in the enterprise and tackle implementation challenges.
"Given the massive amounts of hype and promise surrounding AI and related technologies like machine learning and deep learning, it's become increasingly difficult to make critical innovation and investment decisions in the space," writes Lux analyst and lead report author Cole McCollum. "Taking a technology-first approach (e.g., selecting a vendor simply because they claim to use the latest AI techniques) has led, and will lead, to many failed companies and projects."
Depending on who you ask, we are in the midst or just past the peak of a huge amount of buzz about what AI can do. But there's still a great deal of confusion about just what AI actually is and the problems it can solve. McCollum points out that the definition can vary. For instance, some use AI to describe artificial general intelligence, which is capable of completing a wide range of tasks comparable to human intelligence. Others use AI to describe narrow AI that performs a single task or a few tasks with high competence, such as identifying cats in photographs.
AI has been the force to beat Jeopardy champion Ken Jennings and to win the game of Alpha Go against the world's best. Yet there's a breakdown between what it can do, and how the term AI is bandied about as a marketing term. The Lux report points out that 40% of European AI startups actually do not appear to use any AI.
"With AI startups attracting 15% to 50% more funding compared to their non-AI peers, many startups are inaccurately rebranding themselves as AI companies in order to attract funding," McCollum writes.
There are also cases where the current reality of what's possible hasn't measured up to the marketing. For instance, the Lux report points to the IBM Watson Healthcare partnership with MD Anderson that was ultimately cancelled after $62 million in investments.
What's more, early adopters of technology are the ones that encounter the early problems. For instance, many early adopters of AI tools have faced issues with bias, cyber security threats, and the lack of interpretability of AI systems, according to the Lux report.
To avoid these pitfalls, McCollum recommends that organizations employ an outcome-focused framework. Organizations must focus on product capabilities instead of buzzwords. They should look at a potential AI application's maturity level, and then they must also identify opportunities and threats in the fast-moving market.
At a high level, McCollum writes, any AI project has four major steps:
Problem selection - understanding the core use cases and outcomes of AI and mapping those onto business problems. This also includes an examination of whether the application is possible with today's AI technologies. For instance, some technologies are mature now, such as text sentiment analysis and facial recognition. Others are at the research stage, such as automatic software development.
Data preparation - identifying, cleaning, and wrangling data from multiple sources including sensors, databases, and APIs.
Model selection and training - selecting and testing machine learning models.
Deployment - Utilizing and monitoring those ML models where they can be used to make predictions with new data.
McCollum notes that many applications, such as autonomous vehicles capable of driving on any road in any city, will still take significant development. If your organization is facing that kind of development dilemma, McCollum recommends reducing the amount of AI needed in the solution or planning for longer development cycles.
Read more of our articles on AI and other emerging areas of technology:
About the Author
You May Also Like
More Insights
