Xiaoxi Du ’14, Dennis Chua ’14, and Joy Chua ’13 won Cornell's IBM Two Worlds Case Competition with their theoretical program “Hello Watson,” which would use the game show star's intellectual abilities to enhance tech support for consumer electronics.
The days of reluctantly calling overseas technical support centers to help fix broken electronics may soon be over thanks to a new computer program idea called “Hello Watson,” developed by a team of Cornell students as part of the IBM Two Worlds Case Competition on Oct. 14.
The competition, a collaborative effort between the Delta Sigma Pi business fraternity and the Association of Computer Science Undergraduates, challenged students to apply the problem solving computer system Watson, from the popular game show Jeopardy!, to an industry, like healthcare or financial services. Xiaoxi Du ’14, computer science, Dennis Chua ’14, chemical engineering and Joy Chua ’13, chemical engineering, won the competition with their theoretical program “Hello Watson,” which would use the game show star's intellectual abilities to enhance tech support for consumer electronics.
Jeopardy!’s Watson is an artificial intelligence with a compilation of 47 different kinds of programming software capabilities. It has a deep understanding of native languages like English. In 2011, on national television, Watson used this ability to beat out two of the most successful Jeopardy! contestants of all time in a human-versus-machine quiz show battle.
“The question at hand was how to adapt Watson, which has a knowledge base for Jeopardy!, into ‘Hello Watson’, which has a knowledge base for consumer electronics.” said Du.
Du, Chua and Chua theorized that Watson’s ability to understand English in a game show setting could enable it to improve tech support for various consumer electronics, like toasters, iPhones and computers. In order to create their Hello Watson business plan, the team needed to study how IBM’s Watson functioned in the Jeopardy! game show.
According to Du, Watson’s abilities can be broken up into three main areas. First, Watson is able to understand a question grammatically and semantically. Second, Watson can draw on a huge knowledge base. For Jeopardy!, Watson was programmed with dictionaries, thesauri, news articles, books, various databases and encyclopedias, including Wikipedia. Third, Watson is capable of decision making and analysis.
Using its large knowledge base, Watson can generate several possible hypotheses and assign them with various confidence intervals. In the game show, Watson produced the answer with the highest probability, assuming that that answer had a high enough confidence value.
The team was determined to apply Watson's skills to aiding tech support call centers for their business plan idea.
“Imagine if your computer malfunctions and you try to contact the support center by phone. Usually, you get this random person who doesn’t really understand you or the problem and you really don’t understand them, either," Joy Chua said. "This is a communication problem. The second problem is that the tech support guy is only one person and can’t always give you the right answer.”
According to Joy Chua, “Hello Watson” would provide a plausible solution to these problems because it would have the ability to understand the words that a person is saying and then quickly search its database to produce various solutions for the problems along with confidence intervals and troubleshooting tips.
The team followed a framework designed by IBM to create their computer program. The framework has four phases: inception, elaboration, construction and transition.
During the inception phase, the team explored the problems and then identified ways they could make their “Hello Watson” idea solve them in the way that it would solve questions on Jeopardy!. According to Du, having a quick response time is necessary for consumer satisfaction and cost efficiency.
In the second phase, elaboration, the team decided what sources to put into “Hello Watson”. According to Du, things like manuals, data sheets, dictionaries, internal troubleshooting logs and real online support forums would all be all crucial.
“‘Hello Watson’ would also be given information from real life experts of the companies that have electronics represented within Watson, like Sony or Apple,” said Du said. "IBM would collaborate with the companies, deciding the breadth and depth of the data, which in the end would form the ‘grand truth’ that Watson would regard as baseline facts.”
During the third stage, called construction, Hello Watson would be improved through modification and retraining. The team said that in its initial startup, Hello Watson is like a child learning to talk. It needs to be taught and continually corrected for steady improvement. With time Hello Watson would become more and more proficient, just as a child does as it grows.
The fourth and final stage is the transition stage. According to Du, this stage is the “hard core testing phase” for the program's speed and robustness. She said that Hello Watson needs to be able to handle several calls at once while still providing answers quickly and correctly.
After designing “Hello Watson” around these four phases, the team considered other issues, like preserving privacy and what would happen if “Hello Watson” did not have the correct answer.
“One problem that ‘Hello Watson’ faces is the issue of privacy,” said Dennis Chua. “Companies don’t want to be giving out potentially sensitive information that anyone, including IBM, could potentially discover,” he said.
To solve this, the team created distinct “clouds” of information. Each company would have its own private “cloud” that other companies cannot gain access to. The sensitive information put in the clouds would only be retrievable by the company that owns it. Each private “cloud” would be connected to one larger public “cloud.” The public cloud, which anyone could access, would hold the general information of all the companies. While Hello Watson would have access to both the private and public “clouds”, IBM would only have access to the public “cloud”.
In the case where Hello Watson does not know the correct answer to a problem, the team said that the program will prompt the consumer for more information, in the hopes of learning a new crucial tip that will help it provide the answer. But if it still cannot answer, Hello Watson would simply refer the user to the most relevant person.
“There would still be a team of highly trained experts who would be responsible for answering questions that slip through Watson,” said Dennis Chua.