Vikram Mahidhar from Rage Frameworks presented his speech “AI in the Enterprise – Here and Now” at the AI Summit in New York 1st December, revealing new and exciting results from their research conducted earlier this year.
Mahidhar started the presentation by revealing results from Rage Frameworks’ survey conducted in October, alongside MIT and Emtech 2016-attendees. The question addressed in the research was: Which reasons prevents your company from purchasing AI tools?
The data was based on a survey conducted with 132 business executives and revealed that the two main reasons why companies are prevented from purchasing AI tools, were that it takes too long to train the AI machine (29%), or that the company cannot afford it, due to the perceived higher cost of AI technology (29%).
Following close behind in the survey results with 20.83% was: “AI technology falls short on solving the complete business problem” and “others” (20.83%), as reasons for why companies are struggling with implementing this new technology. Following this research, Rage Frameworks also raised another question:
“Which one of the following AI capabilities are important to the AI solutions you would invest in?”. Here, Rage Frameworks’ research revealed that reasoning and traceability came out “on top”, with 55%, followed by natural language understanding in second, with 53%.
Natural language processing followed with 49%, machine learning with 42%, robotic process automation 36%, natural language generation 35% and computer vision with 21%.
Looking at machine intelligence in a broader perspective, Mahidhar listed three key dimensions of machine intelligence: knowledge acquisition/representation, computational statistics and computational linguistics.
However, in addition to outlining the key dimensions of machine intelligence, Mahidhar also mentioned any potential problems related to machine intelligence, breaking AI into three different genres. Mahidhar listed “Explicitly Assumed Models”, “Learn from Data Algorithmically” and “Learn to Interpret/Understand Meaning”.
When learning from data algorithmically, the desired outcome is prediction (quantitative data). However, issues that appears can be Ad Hoc search clustering, problems with extraction of the data, and/or classification (qualitative, hybrid data). When learning to interpret/understand meaning, issues could be interpretation, either of natural language or other data.