IBM’s stalled ambitions with Watson show AI isn’t elementary, but can work when focused 

Contextual variability can be a roadblock to successful use of artificial intelligence, say Carlos Cordon and Amin Kaboli

When IBM’s Watson supercomputer won US TV show “Jeopardy!” in 2011, it was heralded as a sign of how artificial intelligence would revolutionize industries from healthcare to finance and agriculture and retail. 

In 2015, the company founded Watson Health with the goal of using its artificial intelligence platform to help solve healthcare problems, including improving the way doctors prescribe treatments for cancer and diabetes. But a series of setbacks dashed those expectations and culminated in the sale this week of part of the Watson Health business to private equity firm Francisco Partners. 

While some people might see IBM’s scaling back of its lofty ambitions as a failure, we believe this shows how artificial intelligence succeeds best in business when it is deployed in a very focused context-dependent way. 

Instead of closing Watson Health, IBM sold the business to Francisco Partners, a private equity company that has a long history of investing in healthcare technology companies, including Landmark, QGenda, Trellis, and Zocdoc, suggesting the company believes the technology could still succeed if run by people with focussed healthcare expertise. 

Indeed, if you look at the companies that have used AI most successfully, they have done so in a very specialized space. US company Aera technology has developed algorithms to improve supply chain performance; Swiss firm Netguardians makes software that can detect fraud in banks; people trust Spotify’s algorithms to identify new music that matches their preferences. 

IBM stumbled by believing that Watson’s technology could evolve into a technology that could be applied across all industries, in the same way that Microsoft’s Excel has become the standard spreadsheet software.  

Other tech giants have also tried, as yet unsuccessfully, to develop AI-driven technologies that can be applied generically. When Amazon and Apple launched their voice-activated software assistants Alexa and Siri, many predicted a revolution in the way we shop and manage our home devices. But this hasn’t yet come to pass, with the majority of people using voice technology mainly to stream music and podcasts or find out the weather forecast. 

Another learning from IBM’s experience with Watson is the importance of decentralizing artificial intelligence across your organization. There has long been the belief that AI should be organized within a centralized IT department under the remit of a Chief Data Officer. But the struggles of some of the big tech giants highlight the importance of having people who not only understand the technology but also the specific context across functions. 

When seeking out talent, companies should look to not only hire staff with machine learning expertise but put them in teams with people who understand the industry-specific challenges. 

By deploying their vast databases of medical images on which to train their algorithms, healthcare companies Siemens, Philips and GE Healthcare have hoped to keep an edge over big tech. But, arguably, it is also their understanding of the specific workflows of doctors and nurses who use their devices that enables them to fine-tune their software to meet their needs. 

In the case of self-driving cars, there have been a number of high-profile accidents caused by unidentified objects, such as plastic bags, flying across the windshield. The AI slams on the brakes and causes a crash.  For software engineers developing self-driving cars, they will need to develop algorithms that can accurately predict whether a pedestrian is about to step out into the road, a complex area known as social forecasting.  

Again and again, a huge variability in the context is proving a roadblock to a successful use of AI for business. The key takeaways are: 

1. AI success seems to happen when we use it a focused context 
2. You should organize you data analysis competencies in a decentralized way in your organization to allow for such focus 
3. If you need external partners or providers for AI solutions, you might do better to look for those with experience in your specific context.