The artificial intelligence (AI) hype is everywhere, and it seems like the world’s biggest tech companies are all embracing it. It’s also fact that several companies have been investing in AI heavily to get the desired return on investment (RoI). But the current speed of AI innovation is much slower than what analysts originally expected.
Old thinking to new AI problems
One obvious reason could be applying old thinking to new problems in AI that’s damaging the prospects of the technology in terms of its real-world application. A new study by Lux Research outlines the challenges companies face when it comes to AI innovations and how they can change their innovation processes to improve their success rates with AI.
Lux Research Director and author of the report, Dr. Shriram Ramanathan said, “Companies are hampered because they have been using traditional processes to manage AI innovations. It is imperative to not apply old thinking to new problems, especially when it comes to AI.”
With stakeholders so narrowly focused on ROI, project selection, and execution, few pay attention to the underlying innovation processes used to manage the AI project.
“The fundamental challenge lies in that the underlying logic in an AI solution is intricately tied to the raw data that it provides insights on. While this allows AI solutions to adapt easily to continuously changing environments, it is also AI’s Achilles’ heel,” explained Ramanathan.
AI solutions quickly start deviating from their original purpose the minute they are deployed in the real world. Companies need to implement a continuous and ongoing developmental effort to keep their AI models current.”
There is a lack of use cases and proper understanding of the application of modern AI, as a result of which the technology is confined to research laboratories, with little effort to demonstrate business value.
“The problem is, while researchers try to outdo one another on contrived benchmarks, one in every nine people in the world is starving,” Hannah Kerner is an assistant research professor at the University of Maryland, writes in MIT Technology Review.
According to Kerner, most AI research is marginalized at major conferences. Their authors’ only real hope is to have their papers accepted in workshops, which rarely get the same attention from the community.
“This is a problem because machine learning holds great promise for advancing health, agriculture, scientific discovery, and more. Only if others in the field had prioritized real-world applications, groundbreaking discoveries would we have made by now,” she said.
In his paper titled “Machine Learning that Matters”, NASA computer scientist Kiri Wagstaff said, “Much of current machine learning research has lost its connection to problems of import to the larger world of science and society.”
The paper said, when studies on real-world applications of machine learning are excluded from the mainstream, it’s difficult for researchers to see the impact of their biased models, making it far less likely that they will work to solve any problems.
Bringing AI innovations to the marketplace
According to Ramanathan, in order to improve chances of success in bringing AI innovations to the marketplace, stakeholders should, first of all, reduce the level of ambiguity around AI projects Begin by defining the AI use case as narrowly as possible and plan for a broad range of potential scenarios, and secondly, take the time to clearly define KPIs that bridge the digital and physical worlds.
More importantly, he believes that executives should plan for real-world deployment early in the game. This could mean incorporating a wide range of real-world data while building an AI product, ensuring that the AI solutions are easy to track and update, and be ready to update existing traditional models and processes so as to make them more agile and efficient.
Awais Bajwa, Data Architect and AI Enthusiast, too mentioned in a recent article that the future wave of AI is to break the “AI Blackbox” and to understand the reasoning of the decisions and predictions made by the Machine Learning model.
In other words, those applying the old school thought in machine learning think this work consists of simply applying methods that already exist. In reality, though, adapting machine-learning tools to specific real-world problems takes significant algorithmic and engineering work.
Machine-learning researchers who fail to realize this and expect tools to work “off the shelf” often wind up creating ineffective models. Either they evaluate a model’s performance using metrics that don’t translate to real-world impact, or they choose the wrong target altogether. And that’s where a massive transformation is needed and soon!