Covering Disruptive Technology Powering Business in The Digital Age

Home > DTA news > News > Would you want to debate IBM Research’s Project Debater?
Would you want to debate IBM Research’s Project Debater?
April 3, 2020 News


IBM recently announced several new IBM Watson technologies designed to help organisations begin identifying, understanding and analysing some of the most challenging aspects of the English language with greater clarity, for greater insights.

Among those was the first commercialisation of key Natural Language Processing (NLP) capabilities to come from IBM Research’s Project Debater, the only AI system capable of debating humans on complex topics. We in Disruptive Tech ASEAN are always fascinated with new technologies that disrupt. So when we discovered there was an AI that is capable of debating, we reached out to IBM Malaysia to get some comments from them on this innovation.

As the saying goes “your AI will only be as good as the data you feed it”, and IBM upped the stakes in the AI race with the announcement of AI ability to understand the language of business. IBM Malaysia’s chief technology officer Freddy Lee spoke exclusively to Muhammad Zulhusni at DTA on the subject.

According to Freddy, the technology is ready for businesses that depend on customer services such as contact centres and call centres where it can be used to better understand conversations between customers and a virtual assistant.

“Two clients I can name are ESPN Fantasy Football that uses Watson Discovery and Watson Knowledge Studio to analyse millions of football data sources daily produce insights in real-time for fantasy football fans. What it does is identify the tone and sentiment from the media – news stories, blogs, podcasts and tweets to surfaces insights to give a positive or negative summary about a player. Another client is KPMG who worked with us to co-create an engine that can identify, claim and retain potential tax credits for their clients. KPMG told us that their clients have benefited with more than a 1000% increase in the number of documents reviewed. “

For now, Freddy said English is the first language IBM Watson Natural Language Processing and Natural Language Understanding can support. And this is because English is a common language used widely for business. Other languages in the pipeline are French, Spanish and German that will be added in the second half of this year. In 2021, IBM will add Arabic, Chinese, Dutch, Italian, Korean, Japanese, and Brazilian Portuguese.


Understanding Idioms

Interestingly, Freddy said the IBM Research team are improving the ability of the IBM Watson Natural Language Processing and Natural Language Understanding by feeding it with common terms, phrases, colloquial and idioms for analysis.

“Here is an interesting story: there are over 25,000 idioms in the English language but we started out with around 9,000 idioms. The idioms were filtered based on common use. Idioms that are too rare were excluded. and filtered again to remove neutral idioms. The team ended with a core 1,600 idioms that were necessary for sentiment analysis. For example, idioms like “and so forth” or “after all” do not affect the sentiment of a sentence in which they are used. But our work does not end here, this is the foundation for our client to customise and add local idioms.”

Freddy explained that the AI systems are taught to understand subtext, nuances and emotion as IBM demonstrated in Project Debater last year in Cambridge when it faced off Harish Natarajan who is the world debate champion. Project Debater lost, but it was clear to the audience that the AI was able to argue persuasively. This indicates that the AI learning is no longer literal.

For example, Freddy said a term like ‘hardly helpful,’ or ‘hot under the collar,’ were difficult and a challenge for AI systems to understand because they are difficult for algorithms to identify. But with advanced sentiment analysis, we can now analyse these common phrases and understand the context it is spoken in to get the full picture.


Understanding more APIs

Freddy commented that one of the APIs is Advanced Topic Clustering with the capability to gathers incoming data to create meaningful clusters for specific industries such as banking, insurance, research, manufacturing or healthcare. So basically, he believes any topic is possible, and this is driven by the client’s need and requirements.

Apart from debates, IBM Watson Natural Language Understanding is able to deliver decision making, relevant conclusions and hypothesis through four APIs:

  1. Advanced Sentiment Analysis capability to identify and understand complicated words, idioms and phrases.
  2. Summarisation capability to pull textual data from different sources such as audio, text and video to give a summary of what was said about a topic.
  3. Advanced Topic Clustering capability that allows the user to automatically form data clusters from incoming sources to create meaningful information for analysis.
  4. Customisable Classification of Elements in Business Documents to allow users to build AI models for classification.

As the technology is commercially available to businesses today, IBM are open to pilots in the local market. If a learning institution is interested in investing and building an AI system using their technology, just give them a call.

Perhaps in the future, we may get a chance to have a debate with Project Debater as well.