Introduction
IBM Watson started as a question-answering system named after the first CEO and founder of IBM Thomas J. Watson. The system was designed to answer natural language questions and developed in the DeepQA project. The initial goal of the project was to develop an ‘intelligent’ system that can recognize questions asked in a natural language, process the information and compare it to what’s in the massive database, and answer in a human-understandable language.
History of IBM Watson
Charles Lickel, the research manager at IBM was intrigued by the TV show Jeopardy! and decided to take the challenge of putting an IBM system against humans in the same show. He passed on the idea to Paul Horn (Research Executive) and managed to get it approved in 2005. The biggest challenge was to develop a system that can respond within a few seconds (compared to a few minutes a similar system Piquant used to take with 35% success rate).
The complexity of problems posed by Jeopardy! made it look like an impossible task. The initial tests proved to be a failure with 15% success rate compared to 95% of humans. It took IBM researchers three years to develop a system that could compete against humans. In 2010, Watson started regularly beating human competitors, thanks to its access to over 200 million pages of content. Watson’s development also involved graduate students and faculty members of various universities.
Jeopardy!
Representatives from IBM communicated with Jeopardy’s management team in 2008 to discuss the possibility of competing IBM Watson against the current champions. Although both the teams had some disagreements over how to arrange the competition, a practice match was conducted on 13 January 2011, while the first competition was broadcast on 14 February 2011.
The researchers behind IBM Watson initially designed the system to compete in the quiz show Jeopardy!. In 2011, IBM Watson won the first prize ($1 mn) in the competition against Jeopardy! champions Ken Jennings and Brad Rutter. IBM donated the entire amount to World Community Grid and World Vision (50/50), while Rutter and Jennings also donated 50% of their winnings.
The system had to store all the data in its massive 16 terabytes ram otherwise accessing it from the hard drives would have bogged down performance. Different means were used by Jeopardy! staff to notify human players and IBM Watson about when to buzz. An electronic signal was used to notify Watson while a light was used to notify humans.
Commercial Application
IBM announced the first commercial application in 2013 for lung cancer treatment by utilizing management decisions at New York’s Memorial Sloan Kettering Cancer Center. The business chief of IBM Watson in 2013 announced that 90 percent of nurses who used IBM Watson were able to follow its guidance.
Components of IBM Watson
Like most other computer systems, IBM Watson also comprised of hardware, software and data. The hardware was built on IBM’s own DeepQA technology and powered by massively parallel and work-optimized POWER7 processors. DeepQA technology along with powerful processors was designed to gather and analyze huge amount of data and generate hypothesis.
Hardware
The hardware included a cluster of 90 powerful IBM Power 750 servers with each server having a 3.5 GHz, 8-core POWER7 processor. Each processor had 4 threads/core, which translates into a total of 2,880 threads. Massive computing power along with 16-terabytes of RAM enabled the system to process over 500 GBs of data per second, which is roughly equivalent of more than a million books each second. The estimated cost of the hardware was somewhere around $3 million.
Software
The software system included Apache UIMA and IBM’s own DeepQA technology that ran on top of SUSE Linux Enterprise Server 11 + Apache Hadoop framework. The software was written in different languages, including C++, Java and Prolog. Compared to regular document search software, DeepQA tech can understand natural language and search queries in greater detail, allowing it to come up with precise answers.
Data
IBM Watson used a variety of sources to process information, including dictionaries, encyclopedias, news wire articles, thesauri and literary works. The databases used include ontologies and taxonomies databases such as WordNet, Yago and DPPedia. The system was fed millions of documents to build knowledge, including encyclopedias, dictionaries and reference materials.
Capabilities
Watson’s capabilities have been enhanced in recent years and the system has evolved to take advantage of modern deployment models, including IBM Cloud and machine learning. IBM Watson is not just a question-answering system anymore and although it can still do that, its functionality has been greatly enhanced. In addition to listening’ and answering questions, the system can now also see, read, taste, interpret and learn.
How IBM Watson Operates?
The system was designed to segregate questions into smaller chunks, sentence fragments and keywords to find phrases that are statistically related. Watson’s strength was its ability to simultaneously execute hundreds of language analysis algorithms quickly. The system is more likely to pick answers that are selected independently by more algorithms.
The system checks all the potential answers against the database to establish if the answers make sense or not. Watson’s ‘voice’ was a synthesized version of Jeff Woodman’s recordings made in 2004 for IBM’s text-to-speech.
IBM Watson vs. Humans
Humans and IBM Watson process information differently so the results can also differ. The working principle of Watson is to parse different keywords in a cue and search for terms that are related. Compared to humans, Watson had its own advantages and disadvantages. The biggest disadvantage was that IBM struggled to understand and take context into account.
Humans are better at generating fast responses, especially in the case of short cues. IBM also lacks the ability to buzz before it’s sure about an answer. That’s the reason why humans competitors were able to use 6-7 seconds that Watson needed to decide whether or not to signal for responding. During that time, IBM Watson had to make sure that enough evidence is present to answer the question correctly.
When competing against humans in Jeopardy, Watson included circuitry that could receive the ready-signal and examine if there is enough evidence to activate the buzzer. Due to its powerful internals and optimized software, Watson was able to react more quickly than humans.
Applications of IB Watson
IBM Watson’s ultimate goal is to interact with us in human terms across a wide range of processes and applications in a way that we can justify and easily understand. Current and future IBM Watson applications include:
- IT operations
- Customer services
- Risk and compliance management
- Research and discovery acceleration
- Disruption anticipation
- Legal research
- Financial services
- Government institutes
- Healthcare
- Chatterbot
- Code building
- Teaching assistant
- Fashion
- Weather forecasting
- Advertising
- Tax perpetration
Conclusion
IBM Watson is marketed as an enterprise-ready AI suite for apps, services and tooling that are AI-compatible. The computing system has come a long way since its inception and has evolved into a powerful system that can help enterprises unlock the true potential of their data.
It provides deep and accurate insights that help enterprises predict with confidence and integrate Artificial Intelligence into their workflows. What started as an inspiration to compete against humans in a TV show in 2004 has turned into a powerful enterprise solution and a major technological breakthrough in the field of computing.