The Future of AI Is Here: Reflections on IBM Think
The IBM Think Conference in San Francisco this month was a massive event, taking over the SoMa district, interrupting traffic, annoying the locals, and driving up hotel prices to stratospheric levels. Despite a fairly decent rainstorm that pounded the city throughout the weeklong conference, thousands of attendees swarmed through the Moscone Center and other nearby venues, intent to learn, share, sell, and network. Among hundreds of sessions, exhibits, and demos, one theme came through clearly: for large enterprises especially, the AI-driven future for which we’ve been told to prepare is already here.
IBM is of course the home of Watson, which most of us probably know for its associations with Jeopardy, or the healthcare industry, or the Weather Channel. But there’s far more to the Watson machine learning platform than can be easily surmised from its better known public appearances. In fact, enterprise companies are using Watson technologies today to address a myriad of challenges inherent in the scale of those businesses.
To take a few examples: Petco is using Watson’s Natural Language Understanding, Assistant, and other tools to provide instant access to supply chain information for numerous company stakeholders. Carhartt uses a tool called Commerce Insights to provide real-time data on business performance to marketers and merchandisers, who are able to restock products or adjust supply to capitalize on current trends. Box lets users automatically tag documents with relevant concepts and topics, making them more easily searchable. Smarttech uses Watson tools to enhance its cybersecurity efforts and increase timeliness of response to threats. AutoGlass Body Repair used Watson Visual Recognition to build a tool that allows customers to upload photos of vehicle damage for instant analysis.
These are just a few of the applications showcased at Think. The major takeaway is that any company that relies on large amounts of data stored in digital documents, spreadsheets, databases, images, or other media, or who must manage complex workflows that involve numerous smaller decisions adding up to significant business impact, can potentially benefit from technology whereby machines are trained to extract and organize meaningful content. In other words, the potential applications of AI are endless.
Let me illustrate with another example that shows the multifaceted role AI is playing in an age-old industry. Gaurav Rao, executive director of engineering at Watson OpenScale, described an insurance company that uses AI to automate the review of claims. In this example, nine out of 10 claims can be processed automatically using a heuristic that takes into account numerous factors including the substance of the claim itself as well as the applicant’s policy information, number of prior claims, age, and other components. One purpose of providing oversight through OpenScale is to improve the automation process and further minimize the need for human intervention in claims processing.
Even more intriguing, OpenScale was able to highlight potential bias in claims decisions. Rao walked through an example where the OpenScale dashboard uncovered potential causes and suggested fixes for a bias that negatively impacted older applicants. In some cases, tweaking the heuristic involves tradeoffs which must be carefully considered. For example, Rao explained that as fairness improves, accuracy may slightly decline. Humans must be present at the controls to determine the right balance of priorities.
Gartner has claimed that according to its research, AI will generate $2.9 trillion in business value by 2021 and will recover 6.2 billion hours of productivity. Enterprises are recognizing this potential, with 85% viewing AI as a strategic opportunity, according to IBM. For those already using AI, 72% see increased revenue and 28% see cost savings as the primary benefit.
Though many forward-thinking enterprises are already employing AI solutions, those who have extensively incorporated AI in their offerings or processes represent only a twentieth of all companies, according to a study from MIT Sloan Management Review. Barriers to entry include lack of required technical staff, cited as a concern by 63% of respondents in an IBM survey, as well as regulatory concerns, cited by 60 percent. For many industries such as insurance and healthcare, AI solutions will only make sense if they are trained to work within stringent regulatory requirements.
Despite those concerns, companies that have already embraced AI are beginning to realize its strategic benefits. One such company, Daimler AG, demonstrated its Mercedes-Benz virtual assistant at Think. The assistant, aptly named Mercedes, appears to end users in a few different guises. One is an Alexa-like in-car experience for the company’s E- and S-Class sedans. Another completely different deployment, created by Mercedes-Benz Financial Services, makes use of a chatbot interface to help borrowers pay off their current car loans and finance new Mercedes purchases.
To create the assistant, the Mercedes team used Watson along with a text analysis visualization tool called Quid. The team examined more than 150,000 chat messages between consumers and human agents, attempting to cluster linked topics and to capture the many variations by which consumers communicate their needs. They found, for example, that consumers ask for loan payoff information in as many as 200 different word and phrase combinations. Armed with these variants, the Mercedes team was able to train the Watson Assistant tool to respond accurately to the majority of queries.
Gartner has claimed that 80% of the world’s data is unstructured. I’ve tended to think of this statement as meaning that data is hidden away in documents, PDFs, Excel spreadsheets, and other forms that are impossible to collate and normalize using traditional methods. That’s certainly part of the story. But the discussions at Think collectively painted broader and more nuanced definitions of “structure” and “data,” suggesting that business processes themselves, from supply chain management to customer service workflows to pricing fluctuations based on market conditions, are all subject to analysis, automation, and optimization by means of AI.
As I’m completing this piece, news is breaking on CNN about the nonprofit research firm OpenAI, which has announced that it is withholding the latest release of its text-composing technology from the public due to concerns that it so convincingly mimics prose written by humans that it could be used maliciously. Public perception about the future of AI has turned somewhat sour in light of such dystopian stories. That’s one reason I found it refreshing to learn at IBM Think about the many practical uses of AI for business and industry. One company may choose to withhold its technology for now, but in large part the tools to build our AI-driven future have already arrived.
If you’re interested in some of the potential impacts of AI on local search and reputation management, I’ve shared some thoughts on those subjects over at the Brandify blog.