We have all watched the movies and read the books – evil machines with irresponsibly well developed algorithms for reasoning and planning take over the world and oppress the human race. The reality of Artificial Intelligence research and its applications are not quite as dramatic, but can be just as interesting. The current Artificial Intelligence In Industry (or AIII for short) series aims to introduce the latest and most notable developments in this field, which have left the academic enclosure and have stormed into the intense world of industrial applications.
The premiere article in the series presents IBM’s Watson, a journey that starts from IBM Research manager, Charles Lickel’s crazy idea of beating human champions at Jeopardy!, a follow-up of Deep Blue’s 1997 success over Garry Kasparov. The story goes through years of rigorous software development, incorporating the latest results from academic research, and the finale is yet to come. Nowadays Watson is being prepared to serve as a comprehensive analytic advisor in a number of industries and fields, such as healthcare, oil and gas production, tax revenue, online shopping, customer support, insurance, vine growing, fashion design and aircraft navigation.
So, what is Watson and how does it work?
IBM’s most intelligent pet is a software system that answers questions expressed in natural language, either in writing or verbally. Thus, Watson is able to perform various intelligent tasks, like participating in a TV quiz show, helping doctors with patient diagnosis, or providing financial analysis. In general, Watson listens to your questions and uses its accumulated knowledge and advanced inference algorithms to formulate potential answers. These answers are then ranked and presented to the human with a corresponding degree of certainty. The answer with the highest probability measure is declared as the output of Watson’s work, but the user has the opportunity to review the second-best, third-best, and so on, options, as well. Thus, Watson has the ability to serve as a chatbot, consultant, teaching or technical assistant, diagnostician, and many more jobs that require intelligent, yet somehow standardized, expert conversation.
Watson incorporates some of the latest and best scientific techniques from Artificial Intelligence fields, like natural language processing, machine learning, speech recognition and synthesis, visual recognition, statistical and formal logic inferences and artificial neural networks.
The catch is that Watson cannot be used just as an out-of-the-box tool. You have to train and prepare the system for the specific domain. This is done by feeding it vast amounts of data – usually, these are thousands of documents and/or web sources. Watson reads, understands, learns, and then provides insight, just as any human would do in a technical field. The advantage is that Watson doesn’t require years of school and university education to achieve the same level of expertise.
Of course, Watson is not quite there yet. The system is not stable enough to provide all-round reliable service, and the analysis capabilities are still very limited. IBM is working hard to bring Watson up to speed, and even though the system has been pulled out of service in some companies where it was piloted, the general feeling is that Watson’s goals are obtainable. It certainly looks like the current AI technologies are capable of the analytic tasks at hand. Maybe all that is needed is time (and a whole lotta data), integration with enough expert knowledge, and some good old fine tuning.
Where this amazing tale of technological progress will end – nobody can tell, not even Watson itself. One thing is for sure: technologies like this will transform the industry. Automating so many of the day-to-day tedious technical tasks, expanding the range of complex analysis, and processing incomprehensible volumes of data. Fortunately, things that still remain beyond AI’s limits are imagination and emotions. Everything else, however, seems to be within its reach.