In the pursuit of computers that can not only read but interpret documents, Dynamic Risk - North American's leading pipeline integrity management provider - launched the Cognitive Computing Challenge. With $200,000 in prizes, this challenge seeks to award the creation of software that can not only process information from different types of documents, but map it to target fields in a larger database.
Not surprisingly, this is not the only attempt to produce machines and software that are capable of performing tasks usually reserved for humans. Consider Google's recent development of a chatbot that can mimic human conversations. Given that chatbots (a.k.a. chatter robots) are programs that are specifically designed to provide users with hands-on technical guidance, and in ways that are meant to be familiar and conversational, Google has chosen to develop a chatbot that can hold its own when discussing the most important topic of all - what is the meaning of life?
In a paper recently published on ArXiv, the developers of the new chatbot - Oriol Vinyals and Quoc Le - describe how they created a system that could analyze existing conversations and teach itself to respond. Whereas traditional chatbots rely on a team of engineers programming rules about how they should respond to specific queries and comments, Vinyals and Le chose to go with a "neural network" model instead.
By relying on neural networks, Google programmers were able to create a chatbot that can answer existential questions. Credit: Shutterstock
These networks, which rely on a network of machines to mimic the functions of the human brain, are able to find patterns in large sets of data. More and more, IT giants like Google, Facebook and Microsoft are looking to these to perform increasingly complex tasks - such as managing company data, recognizing objects in photos, identifying spoken words on phones, and offering translation services.
The technology has endless applications, and ensuring that chatbots can provide a more life-like level of service is perhaps one of the most important ones. With customer service increasingly becoming the domain of machines (and all the complaints that result), it only makes sense to build machines that are capable of responding to any question, regardless of whether it is abstract or technical in nature.
For the sake of their data sets, Vinyals and Le exposed their program to massive amounts of dialogue from movie scripts and it left to work out how certain comments require corresponding reactions. Over time, the program was able to work out patterns of proper responses to specific questions or prompts, which included the big questions like, "what is the purpose of life?" and "what is the purpose of dying?"
Using dialogue from movie scripts, the Google chatbot was able to work out proper responses to tough questions. Credit: virtualagentchat.com
According to Vinyals and Le, there are numerous benefits of using neural networks over programmed rules. As they state in their paper:
"Our model converses by predicting the next sentence given the previous sentence or sentences in a conversation. The strength of our model is that it can be trained end-to-end and thus requires much fewer hand-crafted rules. We find that this straightforward model can generate simple conversations given a large conversational training dataset."
In addition to movie dialogue, Vinyals and Le also relied on specific datasets, such as domain-specific IT helpdesk info. In conjunction with what they refer to as a "large, noisy, and general domain dataset" (i.e. movie scripts) their program has shown itself to be capable of offering helpful IT advice and provide answers to tough questions about life. It was even able to offer opinions on everything from politicians to athletes to entrepreneurs.
Vinyals and Le acknowledge that there is still a noticeable lack of consistency with their program. However, this is a common complaint with chatbots who, despite our best efforts, still can't pass for human. Alas, there simply isn't a program that has been capable to beat the Turing Test just yet. But given time, who knows what machine learning and cognitive computing could accomplish?
Interested in creating machines that can think? Register in the Cognitive Computing Challenge for a chance to win the $200,000 prize.
Top Image Credit: nlinews.com
