40% of European “AI” startups don’t actually use any artificial intelligence, according to a recent study. Taking a technology-centered approach to AI and selecting a vendor simply because they claim to use the latest AI techniques has led to many failed companies and projects. Companies and decision makers need to be wary about the hype surrounding AI and learn how to best approach artificial intelligence.
Coming in as the top technology in our Lux 19 for 2019 report, all innovation indicators point towards a massive amount of interest in AI and related technologies like machine learning and deep learning. Last year alone, more than $25 billion in global VC funding went into AI startups threatening to displace incumbents in industries as diverse as automotive to pharmaceuticals. Moreover, major tech companies like Google have shifted from a mobile-first strategy to an AI-first strategy, prioritizing AI above all other technologies and platforms.
Shifting definitions and lack of clarity around what the term “AI” really means have only further fueled hype and confusion. Many players in the ecosystem are inaccurately rebranding themselves as AI companies simply to attract funding and attention. Thus, as the hype around AI continues to grow, it has become increasingly difficult for executives to make critical innovation and investment decisions and those interested in deploying AI must be vigilant about sifting through the noise.
Due to the advancements in the underlying technologies like deep learning, it’s easy to understand why AI has become such a hyped technology. However, what is less clear is how these advancements translate into useful business objectives and what the limitations of today’s AI are.
Some companies have overpromised their AI capabilities. In cases like IBM’s Watson for oncology care, the somewhat embryonic state of AI caused expensive, and in some cases, dangerous failures. Additionally, many early adopters of AI face unintended consequences surrounding issues like bias, the lack of interpretability of AI systems, and cybersecurity threats.
To avoid major failures with AI, companies should strive to understand what AI can actually do for them rather than taking a technology-centered approach. Companies, therefore, need an outcome-focused framework in which they can determine which applications to focus on with today’s tools and how to mitigate challenges preventing successful implementations.
At Lux, we have developed an outcome-focused framework that companies can use increase their overall chances of success for their AI projects. Once an outcome is selected, one must determine if it is even possible to solve the problem being presented with AI and then determine the level of AI required for the application. As shown in the figure below, AI tasks fall onto a two-axis spectrum depending on the complexity of the task’s environment and the amount of reasoning required which can range from quick, recognition tasks to long-term planning and reasoning tasks. Upon segmenting the framework into quadrants, it becomes clear that different AI applications have vastly different properties.
Depending on the level of AI required, AI applications fall all over the framework. Here, we mapped out examples of AI applications onto the framework and color them by associated maturity level. Companies can use this framework to determine which applications to select based on their experience with AI projects, risk tolerance, and desired impact for both short term and long term.
Opportunities exist to leverage the capabilities of this framework. However, after an outcome is selected, challenges still exist in each phase of AI implementation. In our next blog in this series, we will outline the emerging solutions to the many challenges facing AI implementations.
-Artificial Intelligence: A Framework to Identify Challenges and Guide Successful Outcomes (Members Only)
-Recent Webinar: Sensing for Insight: Getting the Most From Commercially Available Sensors (Available for Everyone)
-A Taxonomy for Machine Learning Algorithms (Members Only)
-Five Challenges for Deep Learning (Members Only)