Grace Kao, 21, had never built a bot before.
But in less than two hours on Wednesday evening, the junior computer science major at Carnegie Mellon University created her own weather chatbot on her Macbook Pro, which could not only tell her about the climate but also recommend suitable apparel for current conditions.
“A lot of the work we do [in school] is theoretical and I thought it’s great that this is hands on,” she said.
Ms. Kao is one of about 36 coders who attended an IBM Coding Cognitive workshop Wednesday at Ace Hotel in East Liberty as a part of Pittsburgh’s Inclusive Innovation Week. Previous locations included New York, Los Angeles and London.
Since artificial intelligence is not always easily accessible, the events aim to bridge the gap between coders and technology like IBM Watson, a question-answering AI supercomputer. This process benefits both flourishing programmers and Watson, explained Jeanne Jang, director of IBM’s Digital Innovation Lab.
“Any time you have a diverse group of people, the outputs are better,” she said, explaining that an inclusive group of people can help refine Watson’s “corpus of knowledge,” essentially its library.
Due to a limited amount of space, coders pre-registered for the free event through an online portal and provided information about their coding experience. After reviewing applications, IBM selected those who could benefit most from learning with Watson. As it turned out, most of them were students.
Those selected registered for IBM Bluemix, a cloud platform that hosts Watson applications like the weather bot. Each person received a 30-day free trial to experiment with their chatbots and make optional additions, like implementing speech-to-text capability or embedding the bot into a messaging app like Slack.
The IBM team gave coders a specific set of instructions to train Watson. This is a way to leverage algorithms in order to solve problems, explained Kyle Guske, a junior computer science major at CMU.
Mr. Guske, 21, said he must teach Watson specific intents. Here, an intent is the specific task that Watson needs to understand and respond to — think of it as a stimulus that causes a response, or a question that needs an answer.
“You give it [Watson] data and it learns from that data,” Mr. Guske said. “It’s software that you can use without necessarily understanding how it works.”
And that’s because Watson, as a technology, already has an expansive set of cognitive capabilities that differentiate it from other programmed computing systems. Watson has the ability to understand, reason and learn from its library.
However, the program does not operate under a system of explicit yes-or-no responses. None of Watson’s individual algorithms is perfect, so it creates confidence intervals to determine how likely its response fulfills a given question or intent, said Armen Pischdotchian, academic tech mentor at IBM Watson.
Asking Watson about the weather in Pittsburgh by typing, “Is it raining cats and dogs in Pittsburgh?” may not turn up a result, for example, if the program has not been trained to recognize that idiom.
Michael Greenblatt of Jefferson Hills, a software engineer for IBM, initially had trouble getting his bot to report weather for Pittsburgh, although it recognized Boston and suggested a jacket for Wednesday night’s chilly, wet weather.
Troubleshooting, Mr. Greenblatt, 38, supposed that his bot may be confused by the number of Pittsburghs across the country. To narrow it down, he asked for the weather in Pittsburgh, Pennsylvania.
Success — despite the sunlight filtering in through the ballroom window, Watson told Mr. Greenblatt he would need a jacket later in the evening.
“I think it’s cool that in this amount of time you can create a fully functional app,” he said.
To experiment with an AI chatbot, try out The Weather Channel’s Facebook bot, which is powered by Watson. Visit www.messenger.com/t/TheWeatherChannel.
Courtney Linder: clinder@post-gazette.com or 412-263-1707. Twitter: @LinderPG.
First Published: April 7, 2017, 4:00 a.m.
