In 2011, IBM’s Watson computer beat two of the most successful human contestants on the long-running US game show Jeopardy!, which requires participants to provide a question in response to general knowledge clues. In the event, Watson marked a breakthrough in artificial intelligence with its understanding of natural language...
and ability to make sense of vast amounts of written human knowledge.
Since then, IBM has been preparing Watson for work in business, research and medicine, aiming to help organisations find answers to the questions they often ask, faster and at lower cost.
Businesses can select from a set of 28 application programming interfaces (APIs) with which they can build Watson applications, or integrate Watson’s capabilities within systems they are developing. The APIs can help analyse the tone of text, build a list of contextually related terms, script conversations and classify natural language, and are all available from IBM’s cloud platform Bluemix.
Application of these technologies is spreading. In May 2016, IBM announced a new breakthrough macromolecule that could help prevent deadly virus infections, such as Zika or Ebola, with the aid of Watson technologies. Meanwhile, global law firm Baker & Hostetler has built a ‘robot lawyer’ on Watson.
But businesses cannot simply plug in and go. Any application must first learn the ontology – the language and definitions – particular to a domain in which it operates, a process IBM will help with. From there, developers train Watson in the knowledge that makes up a particular domain, with the help of human experts in the field.
Once experts are confident in their Watson application’s ability, they can let users loose to ask it questions in natural language.
Volume, a UK-based marketing, training and technology company, has been using Watson to develop applications to help its clients in technology sales.
Chris Sykes, chief executive officer, says: “We developed bespoke software applications for enterprise clients. The idea is to create ‘cognitive consultants’ who provide accurate answers to questions from the sales teams. They are able to query in natural language in real time, making a sales person ready from day one.
“During the normal sales process, a sales person can only go so far before they need to bring in a technical expert. But if that expert is not available, it extends the sales cycle.
“With our application, the sales team have the technical knowledge they need at their fingertips. They can query the system before a meeting or while they are with the customer. Information comes back to them in natural, accurate language.
“The net benefits are higher revenue per sales person, a shorter sales cycle and higher conversion rates.”
Vast volumes of material
Applications that help businesses make sense of vast volumes of written material could benefit from using Watson, says Surya Mukherjee, senior analyst with research firm Ovum.
For example, consultancy Deloitte is working with IBM’s Watson team to offer a service that absorbs greater volumes of legal information than would be humanly possible, helping businesses save on regulatory compliance, says Mukherjee.
“Some businesses might have 20,000 pages of regulations to sift through every month to keep on top of compliance,” he says. “To understand what is relevant to them, it takes an army of lawyers. The Watson application can parse the documents, and because it knows what to look for, flag up the relevant parts.”
Crucially, Watson learns from its errors, he adds. “There are false positives and false negatives, but with heuristic algorithms and human feedback, the software learns from its mistakes over time.”
Businesses that invest in these types of application could save time and money on employing experts to analyse large volumes of text or other unstructured data – but Watson does not come cheap, says Mukherjee.
“It is not commodity technology, so it will not be commodity priced,” he says. “There will be cheques to sign.”
While users will be able to select the APIs for their applications from the cloud on a pay-as-you-go basis, they will also need to spend money to ‘train’ Watson in a particular ontology and employ human expertise to check that the applications’ output makes sense.
“You have to ask: do you have the talent to use Watson for your purpose?” says Mukherjee. “Those people are expensive, not a commodity.”
Read more about IBM Watson
- What is the Watson IBM supercomputer?
- When IBM’s Watson analytics system bested Jeopardy! champs in 2011, the world cheered. Now IBM must up its game to make Watson a commercial success.
- The Hilton Worldwide hospitality chain is trialling a robot concierge named Connie, backed by IBM’s cognitive computing programme Watson.
IBM is heavily promoting Watson with the term “cognitive computing”, in an attempt to move its core business beyond the technologies that it pioneered, but which have become commoditised and less profitable.
“Cognitive computing, cloud and big data are the areas where IBM is investing billions, and cognitive might just be the priority,” says Mukherjee. “In technology markets such as databases, analytics and business applications, IBM has lots of competition.
“You could say that what IBM offers, Oracle and SAP also offer. But the last frontier is cognitive, and that is IBM’s story. Tactically, it is betting the farm on Watson.”
IBM has not released pricing for Watson per se, because it will depend on the particular combination of APIs and add-on services that customers consume. It has also not announced how much it is investing in its Watson venture or discussed the computing capacity it has created to support Watson worldwide.
However, it does make strong claims about Watson’s abilities in cognitive computing.
Phil Westcott, European ecosystem leader at IBM Watson Group, says: “Watson is based on systems that learn at scale, reason with purpose and interact with humans naturally. It understands the world in the way that humans do: through senses, learning and experience.”
Elsewhere in IBM’s promotion of Watson, the company claims: “Watson and its cognitive capabilities mirror some of the key cognitive elements of human expertise: systems that reason about problems like a human does.”
Watson lacks common sense
But John Carroll, professor of computational linguistics at the University of Sussex, says that despite Watson’s impressive performance in natural language processing and question answering in Jeopardy! and elsewhere, he is sceptical about the claim that it can reason or understand the world the way humans do.
“It is different and complementary,” he says. “Humans don’t have the ability to read millions of documents an hour, so it goes beyond human ability. But, at the same time, it does not have common sense. It does not have the ability to reason inductively or understand how humans act, move and behave in the real world.
“It can do something that a human can do in same way IBM’s Deep Blue can play chess and Google AlphGo can play Go, but it is still not the answer to replicating human intelligence. It can only do the types of things it was set up to do: to get information from text and from databases and integrate them. It cannot act like a human in the real world and does not have any notion of what the real world is.”
Watson bases its responses on the expert human texts it processes, but is not able to reason much beyond this evidence, says Carroll.
“Ontologies are inconsistent and incomplete, and once you make two or three inferences, it is quite possible to go badly wrong,” he says. “The computer won’t know what is an inconsistency and what an incorrect inference is because it does not have any common sense. It can be led astray very easily. It is safer to work from documents that people have written or individual facts that people have input.”
Carroll says artificial intelligence is being applied to a range of business problems (see below) and some problems may be applicable to more specialist technologies other than Watson.
IBM has impressed businesses, academics and analysts with Watson’s performance in answering natural language questions based on vast amounts text and other unstructured data. Experts agree it has many applications that could benefit businesses and other organisations, but whether its capacity for human-like reasoning stands up to IBM’s claims remains an open question.
Other artificial intelligence systems are also available
Arria extracts information from complex data sources to create output in natural language, typically in the form of a report. It partners with IBM Watson and, until April 2015, counted Shell among its customers.
Brandwatch produces insight into public opinion based on social media and the web. Its customers include Ikea, Marks & Spencer and British Airways.
iLexIR has developed text processing tools in collaboration with the Universities of Cambridge and Sussex and specialises in text analytics, mining, classification and search applications.
DeepMind was founded in the UK in 2010 and was acquired by Google in 2014. Its AlphaGo program beat a human professional Go player for the first time, prompting widespread publicity. Its Health unit is working with the Royal Free Hospital NHS Foundation Trust to present timely information to help nurses and doctors detect cases of acute kidney injury.
AlchemyAPI analyses text for sentiment, category and keywords, and recognises objects and faces in images. IBM Watson Group acquired the firm in 2015 to complement its ability to draw connections in textual data.
Seldon is an open-source and platform-agnostic machine learning system that provides real-time recommendations. It combines behavioural, social, contextual and first- and third-party data to increase the relevance of content and product recommendations. Its customers include lastminute.com.