A Competitor Rises to Watson’s Defense
Following the release of a scathing report by investment bank Jefferies, it would be tempting to jump on the bandwagon and call into question the intelligence of one of the world’s most famous artificial intelligences, IBM Watson.
Gloating is rife in the tech community right now, with Gizmodo publishing an article headlined “Why Everyone is Hating on Watson,” and Social Capital CEO Chamath Palihapitiya going even further: “to be completely honest… Watson is a joke.”
Yet, even as a competitor to Watson, we feel some of the criticism being levelled in the wake of the report is unfair.
“Not as clever as they say”
The principal criticism is Watson is not as clever as IBM says it is – and as such is struggling to generate a profit on the vast sums pumped into its development. “It seems unlikely to us under almost any scenario,” the report’s authors say, “that Watson will generate meaningful earnings results over the next few years”.
Industry insiders will tell you this mismatch in expectations is rooted in a particularly cumbersome aspect of Watson’s technology – the need for pre-tagging all the training data before the supercomputer can learn anything. Usually such training requires Subject Matter Experts (SMEs), and is the most expensive part of a solution based on Watson. Contrast this to a nimble version of cognitive computing called Tactical Cognitive Computing.
However, just as IBM in certain ways overhype Watson’s power, so are the media and tech community now overhyping the implications of this report. IBM Watson as a product might not be making any money – but that doesn’t mean the technology behind it isn’t making money for its many clients worldwide. To understand why, you must first understand what Watson is, and what it is not.
Watson, like Coseer, is best described as a ‘Cognitive Computing’ platform. It utilizes a combination of AI technologies to automate tasks that previously only human intelligence could complete. At its core Watson is a platform that can uncover insights, inform decision-making, and interact with people.
These are not new capabilities – they account for tasks billions of humans around the world complete everyday. But by automating such capability we can uncover massive time-and-cost efficiencies in almost every kind of business and organization – across finance, government, law, healthcare, manufacturing, consumer, tech, and many industries beyond.
While IBM has made unfortunate choices about business and delivery model, the underlying concept is still very powerful. The sales team sells the concept, while customers get to see the execution model. This mismatch in expectations is the basic cause for Watson’s woes.
What Watson is not is an all-seeing, super-AI that business can ‘hire’ and set loose on big data. It is not a single, finished product. IBM essentially anthropomorphized their AI knowhow into ‘Watson’ for marketing purposes, which has been very successful in making them the most well-known player in the market but perhaps not so successful in advertising what their product actually is. As the cost of cognitive computing technology drops over the next ten years, Watson will become a much more realistic prospect to business – and begin delivering substantial value for IBM shareholders.
Fixing the problem
The most noteworthy news in the Jefferies report (especially for early adopters of cognitive systems) is its confirmation that Watson is incredibly picky with the data that can be fed into it. Before an enterprise can begin getting value, they must first commit to a large amount of consultation, data preparation and human hand-holding. The report cites the case of cancer research center MD Anderson, which began working with Watson to match cancer patients to clinical trials. Though initial results were positive, the project had to be put on hold after it burned through more than $62 million without reaching its goals.
The promised transformative benefit of cognitive computing is its ability to process unstructured data, with minimal human assistance. If an enterprise must first risk racking up millions in costs gathering, curating and treating data, this core benefit collapses, and we can understand why it would make many balk at engaging with IBM.
This is especially the case when we consider that the risk associated with Watson is not a necessary side-effect of cognitive computing as a technology. By economizing the most time-intensive part of training an AI system (retraining an AI blackbox when first results are not as expected) and focusing on one form of data, it is possible to drastically increase the cost effectiveness of cognitive computing, without compromising on accuracy. Instead of Watson’s years of planning, strategic budget allocations, and high-risk costs, cognitive platforms can be prepped and returning on investment in a matter of weeks.
At Coseer, we call this Tactical Cognitive Computing: cognitive platforms that require no pre-labelling, tagging or data-cleaning, and do not need mountains of data to train themselves to accuracy.
This mode of deployment means the platform can be scaled up and down with ease and applied with precision, only to the processes in a business that need it most. It also means Tactical Cognitive Computing can only focus on one problem or workflow at a time – and only process one form of data, text (Watson pulls insights from images, audio and video, as well as text).
But even just by focusing on text, we can still achieve results IBM would be proud of: In the financial sector, we can imagine Tactical Cognitive Computing as a team of interns, thousands-strong, crunching through a million documents each day, finding the five-to-ten bullet points that will be relevant to each of a stock market investor’s decisions. In retail, as an intelligent assistant intuitively stepping-in to influence consumer buying decisions. In healthcare, as the most well-read doctor in the world, annotating pathways at life-saving speed.
Intelligence at Scale
Like Watson, the benefit in every circumstance is Intelligence at Scale: self-taught decision-making and a human-like understanding of context, deployed at a scale far greater than any human could achieve. Even if some early adopters of AI are frustrated with its high costs and the time and resources it takes to train a system, it’s easy to see why Gartner predict it will feature in almost every new software product by 2020.
Watson, meanwhile, is still a true superstar of enterprise AI, and there is no suggestion IBM’s software itself was at fault in the MD Anderson case. But that case highlights the pitfalls and costs that contracting Watson might entail. If your data doesn’t match up to Watson’s exacting standards, you’re in for a long ride.
Luckily, there are plenty of other AI for enterprise options out there offering powerful results with much less risk. You can learn more by setting up a meeting with our tech team.