How AI Is Shaping Business—and Changing the World

By Scott G. Stephenson

summer_2019-how-ai-is-shaping-business-and-changing-the-world.jpgMicrosoft CEO Satya Nadella has been quoted as saying, “Ultimately, it is not going to be about man versus machine. It is going to be about man with machines.”

Humans are now learning how to work with machines, and the opposite is true as well. We’re just at the start of a technology that ultimately may be as transformative as MS-DOS or the cloud—with far-reaching implications that will impact our human way of life.

The value of artificial intelligence (AI) is its ability to think like humans do—but at a larger scale and much faster without getting bored or tired. Some fear AI to be the realization of man versus machine resulting in man’s displacement: Who can forget HAL 9000 in 2001: A Space Odyssey or so many other iconic science fiction tales? Other skeptics assert AI is flawed with biases that negatively influence algorithms that skew results. But the technological horses are out of the stable—AI is already being used to tackle some of the world’s greatest economic and social challenges and many more mundane functions.

In agriculture, new AI advancements show improvements in the growth of crops. In aviation, NASA’s Aircraft Operations Division (AOD) uses AI as surrogate operators for combat and training simulators. Robots in many industries are often assigned jobs considered too dangerous for humans. At the USC Center for Artificial Intelligence in Society, AI is being tapped to address socially relevant problems such as homelessness.

While those may be the types of initiatives that earn headlines, innumerable more routine applications are the ones that will likely change the business landscape as we know it and, in turn, how we do our jobs.

One of the areas where AI can and will have a significant impact is any business associated with data. We need look no further for examples than the insurance, financial services, and energy industries. Ever since the invention of the Internet, the world has seen vast amounts of unstructured data. AI has the potential to create structured data and insights from this enormous pool of unstructured data. Another possible use for AI is to help detect fraud from online transactions. These use cases have the potential to deliver significant cost savings and new revenue opportunities.

These days, the use of data analytics has become widespread. No matter the industry, country, or type or scale of business, data seems to be everywhere. And before long, we may be saying the same of AI.

AI in insurance

After years of promise, and some hype, AI applications are finally impacting the insurance industry. These changes have been fueled by an array of technologies, including image analytics and natural language understanding of text and speech. However, technologies and algorithms are not sufficient to create useful AI—relevant data and domain expertise are equally important. And with most in the industry more willing to capture and leverage unstructured data (text, image, speech, etc.), the stars are truly aligning for AI to leave its mark on insurance in many ways.

For example, consider advances in image analysis—a key driver of this shift. Technology developed for facial recognition in photographs is being tailored for insurance-specific applications. Image analysis algorithms can now reliably recognize car models, detect and quantify existing damage, identify vehicle parts, read license plates, parse documents, and more—all from photos and videos. Similar capabilities in the property realm can automatically characterize structural damage from photos and drone aerial footage. For property losses, AI is being developed to help detect and recognize objects in a room, identify damages, and automatically find replacements on the Internet.

And yet the rise of user-supplied imagery brings a collateral risk. Image-altering software is now widely available, which increases the risk of certain users manipulating photos to possibly hide preexisting damage (underwriting) or exaggerate damage (claims). Here again, AI has risen to the challenge. Using algorithms for metadata verification and error/noise-level analysis, such modifications can often be detected. AI algorithms for image matching now can spot pictures downloaded from the Internet that are being passed off as genuine. These analytical capabilities have matured to the extent that photos that easily fool the human eye can be identified as fakes. With such controls in place, insurers are increasingly willing to leverage automated image analysis in the insurance process. 

On the customer front, chatbots have been used for policy quoting and inquiry, but claim applications have been limited owing to the level of understanding, sensitivity, and empathy that chatbots must possess when interacting with policyholders who have suffered a loss. In recent years, substantial progress has been made in natural language understanding (NLU). The annual Loebner Prize in AI—awarded to the computer program judged to be most human-like—has seen increasingly empathetic chatbots. These chatbots go beyond just initiating a claim. They can ask open-ended questions and use AI to comprehend the answers. They can detect feelings and sentiments and tailor their responses to show empathy when someone is sad and appear calm when someone is upset. Coupled with AI advances in speech analysis and synthesis, these chatbots can also have voice conversations. This is a crucial development because complex situations can be communicated more easily through conversations rather than asking policyholders to type answers or explanations.

One other role for AI stems from and potentially provides a solution for the growing number of experts retiring from claims and underwriting and the struggle to fully replace them. This has become a strong motivator for insurers to implement AI and automation technologies to continue delivering superior service. But it’s important to determine the right mix of AI technologies and human expert involvement. With AI capabilities and business rules, insurers can establish what can be fully automated, what needs a quick human review, and what requires substantial human expertise. This notion of “right touch” processing can help revolutionize policyholder experience, produce productivity gains, and even deliver improved business outcomes.

AI in banking

Tackling a different set of challenges, the credit card industry in the United States is one of the most competitive. With a base of nearly 200 million consumers, the rivalry is real, making AI a useful tool in trying to achieve competitive advantage.

For example, banks are continually introducing innovative offers and, more recently, have been launching completely new products to target specific customers. In the past, this was typically done by writing scripts and programs seeking specific information, but AI has opened avenues once deemed unimaginable.

AI also has enabled a more accurate and better-informed assessment of potential customers with the added benefits of lower costs and faster turnaround times. A request for a credit card, for instance, can now get approved online any time of day, with AI processes and systems instantly predicting the riskiness of the potential customer in question. In addition, based on consumers’ spend behavior trends, banks are using AI to issue rewards and make travel-related offers.

With e-commerce on the rise and banks and institutions competing for market share, online fraud continues to be a concern that businesses have been battling for decades. In fact, growth in online transactions and spend has led to further increase in fraud activities. AI has been extremely effective in detection and prevention by flagging transactions based on past customer behavior, transactions that are atypical of the customer’s shopping habits, or transactions occurring in two different locations on the same day. Not only that, one of the major benefits of AI has been its ability to learn over a period of time with powerful supervised learning models. Example: If a genuine transaction is tagged as fraud and approved by the consumer, the AI-built system records the consumer’s action and makes the correction for future reference.

Beyond fraud, money laundering poses another major challenge for financial institutions. The majority of traditional anti-money-laundering (AML) processes are known to be manual and resource-intensive, and yet they’ve been inefficient in making an impact. While AI’s potential in combating AML is yet to be fully realized, more and more businesses are making the transformation and using AI to enhance existing compliance and monitoring systems to better detect questionable scenarios and reduce false positives.

AI and data analytics are evolving and already in use by various financial institutions. Where might AI be headed next? AI could contribute to customizing offers to consumers in real time based on each individual’s needs while maintaining the security of individuals and assets (by preventing fraud).

AI in the oil and gas industry

The oil and gas industry has always been at the cutting edge of technological innovation and leveraged big data (seismic imaging) before we called it “big data.” But advances in analytics, including AI and machine learning, are providing new ways of interpreting this data, yielding previously unknowable insights.

There are three main prizes for the energy industry to seize. First and foremost, health and safety: Automation, robotics, and drones are among the technologies that help ensure secure environments. Reduced accident and injury rates for the workforce will be an unequivocal boon for an industry that operates on geographical, geological, and engineering frontiers.

Second, a structural reduction in costs: To set the scene, senior management in some companies sees digitalization as a means to cut the workforce by half, perhaps more. A measure of success will be growing production as headcount and costs fall.

Value chain analysis suggests that $73 billion of potential annual savings for the industry is achievable in the next five years—based purely on known technologies. That’s about 10 percent of all-in annual global upstream spend. Development and production account for around 90 percent of the opportunity.

The wins in conventional development will likely be from improvements in project design and delivery—automated platforms with smaller, simpler topsides and automated drilling. Savings could potentially total $20 billion. Another $24 billion could be recovered in the production phase by using advanced analytics on production data, smart production systems, remote working, predictive maintenance, 3D printing, and drone-based inspections.

Within the recently emerged unconventional resource plays (also known as shale gas, tight oil, and fracking), automation is among a number of factors that can speed up drilling and potentially save $5 billion annually. In the production phase, unconventional wells decline rapidly. Smart production management can be effective in helping reduce lease operating expenses and extending the economic life of mature shale wells, adding another $5 billion a year in savings.

The third prize is within the subsurface discipline, where geoscientists apply technical data such as seismic, well logs, and geological models to find oil and gas. Cost savings of up to $7 billion annually can be achieved from faster drilling and processing of geological and geophysical work and fewer dry holes.

Potentially an even greater upside may stem from improved exploration and appraisal performance, perhaps derived through new understanding of well logs and chemical analysis or better processing of seismic data to identify and characterize subsurface structures that may contain hydrocarbons. Not only would this offer oil and gas exploration companies the bonanza of finding new resources in existing acreage, but anyone with a competitive advantage in exploration would have a strong advantage in areas of licensing or M&A.

The key technology is advanced analytics, including machine learning. Outcomes include improving the ability of geophysicists to spot anomalies, identify drilling targets, and find analogues from an unparalleled reference library. Highlighting the potential for new technologies to spot opportunities (though not a direct example of digitalization), a major British-Dutch oil and gas company cites the use of its GeoSigns software suite and other proprietary technologies to help discover more than 150 million barrels of oil beneath complex subsurface salt structures in the Deimos offshore field (U.S. Gulf of Mexico).

While the ultimate goal is for machine learning and AI to be able to process data and spot hydrocarbon-bearing reservoirs with an almost perfect success rate, secondary benefits include making better, faster decisions on where and how to drill or saving time and money by opting not to drill at all. By accessing effectively unlimited computing power via the cloud, one of Europe’s leading independent oil and gas exploration and development companies (which began its digital transformation in 2015) has identified that it now has the ability to shave months off its 3D seismic processing. For an exploration-focused company, the improved speed at which it can make drill-or-drop decisions is transformational.

AI in the future

While interest in AI has grown in recent years, many don’t realize that AI began as a discipline in the 1950s. Over the last 60-plus years, tremendous progress has been made in AI research—from game-playing programs that perform better than human champions, to diverse algorithms for discovering complex patterns in data, to the more recent buzz about self-driving cars. With advances in computing capacity and storage and cost-effective consumption models like the cloud, real-world applications of AI have moved from academic domains into the mainstream, as discussed in the examples above. With data now available at enviable scales, depth, and rates, AI applications are poised to change our world.

Verisk is solving AI’s toughest challenges

Here at Verisk, we’re working across many industries and domains to help solve a number of the toughest challenges in AI. We’ve harnessed our groundbreaking research to develop actionable AI solutions that help our customers assess and minimize risk. Here are just a few examples:

  • creating forensic algorithms to obtain and validate image-based property/casualty claims
  • using text and sensor data to diagnose and forecast oil production issues
  • building an AI system to adjust insurance claims automatically
  • using machine learning and aerial imagery to extract data points—such as siding area, number and size of solar panels, and roof condition—that insurers can use to evaluate risk
  • developing an AI solution to help financial institutions automate regulatory compliance
  • creating structured data and information by ingesting and interpreting vast amounts of unstructured data on the Internet to proactively make decisions in areas such as environmental, social, and governance (ESG) and insurance underwriting
  • extracting information from digital documents to speed up workflows
  • analyzing and interpreting insurance forms, contracts, and regulations

What’s more, we’re using AI to create human-machine collaborations—where humans work with AI systems and other machines—to extract information from structured and unstructured data, such as documents, images, videos, and speech. This intelligent information mining has applicability in the insurance and financial services industries, such as extracting relevant information from customer-submitted loan documents for more efficient and automated underwriting. Our goal is delivering machine-level accuracy and quality with the right balance of human intervention.

Scott G. Stephenson is chairman and chief executive officer of Verisk (Nasdaq:VRSK).