The train station is jam packed and the train is supposed to leave in four minutes, but there are way too many people to push through to get to the proper lane. Stress and panic takes over. A missed train means a missed flight, and a missed flight means another $1000 expense. Over the PA a voice calls out, “Toget forlader i en minut.” Good thing all those Spanish classes came in handy. This is often the experience of people traveling in a foreign country. Just when times are the most stressful, it dawns that this is another country with another language. One of the fundamental aspects of the survival of the human race is the ability to communicate. When that ability is taken away, fear, insecurity, and even panic set in. In many ways, the case is no different in artificial intelligence (AI) technology–aside from the emotional roller coaster of course. The rapid and far reaching developments in AI technology mean nothing if systems are unable to communicate. But new AI learning systems are coming into play that will enable AI to increase its communication exponentially.
Analogies and their Role in AI Technology
One of the most unique–and incredible–means of human communication is through analogies. Through analogies, the most complex ideas can be simplified and explained. Amazing inventions have originated from a simple analogical thought process. One of the big challenges facing today’s AI industry is that it is difficult for computers to understand analogies. This is because, as one report notes, “analogies are not the most straightforward idea for a computer to understand.” As a result, AI’s capability of problem solving has a de facto ceiling–there is currently a point at which machines and systems are incapable of handling certain problems. A new AI learning system is needed.
In response to this problem, a deep learning project was founded by researchers from Carnegie Mellon University and the Hebrew University of Jerusalem. The goal of this project has been to develop and AI learning system that can understand analogies. Dafna Shahaf, assistant professor of computer science at Hebrew University, comments how the team, “took advantage of recent advances in deep learning and AI, and found a lightweight way to learn, given a product description, a representation for what the product does, and how it does it.” This is in stark contrast to old methods where handcrafted databases were used, each taking thousands of hours to create. According to Shahaf, the new capabilities allow programmers to ask more specific answers–ones that don’t necessarily demand a one for one answer. Shahaf notes that programmers can now give commands like “find me another product in the dataset that solves a similar problem in a completely different way’ and ‘find me another use for this product.’” Because of AI’s newfound ability to understand analogies, the command makes sense.
What This Means for AI and Human Interaction
Legitimate questions can–and should–be raised concerning AI’s ability to process more complex thoughts and analogies. One of the most fundamental ones is whether or not AI learning systems will allow machines to completely replace the need for human problem solving and critical thinking. Yet Shahaf and the team believe that the advances in AI learning systems will be mutually beneficial. In one of their tests, Shahaf noted how “the people who were exposed to inspirations from our algorithm came up with significantly more creative ideas… in some cases the algorithm helped people explore more diverse parts of the design space, things they would not have thought of on their own.” As it stands, the project has not only helped computers learn how to interpret analogies–it has also paved the way for AI and human interaction with a view to opening up a new and exciting frontier.