Artificial intelligence: From expert-only to everywhere

• Read original article here

View  TMT Predictions 2019 ,  download  the report, or create a custom PDF
AI consists of multiple technologies. At its foundation are machine learning and its more complex offspring, deep-learning neural networks. These technologies animate AI applications such as computer vision, natural language processing, and the ability to harness huge troves of data to make accurate predictions and to unearth hidden insights (see sidebar, “The parlance of AI technologies”). The recent excitement around AI stems from advances in machine learning and deep-learning neural networks—and the myriad ways these technologies can help companies improve their operations, develop new offerings, and provide better customer service at a lower cost.
The trouble with AI, however, is that to date, many companies have lacked the expertise and resources to take full advantage of it. Machine learning and deep learning typically require teams of AI experts, access to large data sets, and specialized infrastructure and processing power. Companies that can bring these assets to bear then need to find the right use cases for applying AI, create customized solutions, and scale them throughout the company. All of this requires a level of investment and sophistication that takes time to develop, and is out of reach for many.
For this reason, AI’s initial benefits have accrued mainly to pioneers with the required technical expertise, strong IT infrastructure, and deep pockets to acquire scarce and costly data science skills—most notably the global “tech giants.” 4 They have the resources to engage in bidding wars for increasingly expensive AI talent. 5 They have also invested billions in infrastructure, including massive data centers and specialized processors. For example:
Google has designed its own AI-specific chips to accelerate machine learning in its data centers and on IoT devices. 6 The company has been exploring deep learning since the launch of Google Brain in 2011, 7 and uses it extensively for everything from performing video analytics to cooling data centers. 8
Amazon has used machine learning to drive recommendations for many years. The company is using deep learning to redesign business processes and to develop new product categories, such as its Alexa virtual assistant. 9
China’s BATs—Baidu, Alibaba, and Tencent—are investing heavily in AI while expanding into areas previously dominated by US companies: chip design, virtual assistants, and autonomous vehicles. 10
The parlance of AI technologies
Below are short definitions of several AI technologies. 11 While no definition can capture every nuance of these technologies, here are the basics:
Machine learning. With machine learning technologies, computers can be taught to analyze data, identify hidden patterns, make classifications, and predict future outcomes. The “learning” comes from these systems’ ability to improve their accuracy over time without explicitly programmed instructions. Machine learning typically requires technical experts who can prepare data sets, select the right algorithms, and interpret the output. Most AI technologies, including advanced and specialized applications such as natural language processing and computer vision, are based on machine learning and its more complex progeny, deep learning.
Deep learning. Deep learning is a subset of machine learning based on a conceptual model of the human brain called neural networks. It’s called “deep” learning because the neural networks have multiple layers that interconnect: an input layer that receives data, hidden layers that compute the data, and an output layer that delivers the analysis. The greater the number of hidden layers (each of which processes progressively more complex information), the “deeper” the system. Deep learning is especially useful for analyzing complex, rich, and multidimensional data, such as speech, images, and video. It works best when used to analyze large data sets. New technologies are making it easier for companies to launch deep learning projects, and adoption is increasing.
Natural language processing (NLP). NLP is the ability to extract or generate meaning and intent from text in a readable, stylistically natural, and grammatically correct form. NLP powers the voice-based interface for virtual assistants and chatbots. The technology is increasingly being used to query data sets as well. 12
Computer vision. Computer vision is the ability to extract meaning and intent from visual elements, whether characters (in the case of document digitization) or images such as faces, objects, scenes, and activities. Computer vision is the technology behind facial recognition, which is now part of consumers’ everyday lives. For example, iPhone X owners log in to their devices simply by looking at them, 13 and computer vision technology “drives” driverless cars and animates cashier-less Amazon Go stores. 14
The few are bringing AI to the many
These tech giants are using AI to create billion-dollar services and to transform their operations. To develop their AI services, they’re following a familiar playbook: (1) find a solution to an internal challenge or opportunity; (2) perfect the solution at scale within the company; and (3) launch a service that quickly attracts mass adoption. Hence, we see Amazon, Google, Microsoft, and China’s BATs launching AI development platforms and stand-alone applications to the wider market based on their own experience using them.
Joining them are big enterprise software companies that are integrating AI capabilities into cloud-based enterprise software and bringing them to the mass market. Salesforce, for instance, integrated its AI-enabled business intelligence tool, Einstein, into its CRM software in September 2016; the company claims to deliver 1 billion predictions per day to users. 15 SAP integrated AI into its cloud-based ERP system, S4/HANA, to support specific business processes such as sales, finance, procurement, and the supply chain. S4/HANA has around 8,000 enterprise users, and SAP is driving its adoption by announcing that the company will not support legacy SAP ERP systems past 2025. 16
A host of startups is also sprinting into this market with cloud-based development tools and applications. These startups include at least six AI “unicorns,” two of which are based in China. Some of these companies target a specific industry or use case. For example, Crowdstrike, a US-based AI unicorn, focuses on cybersecurity, while Benevolent.ai uses AI to improve drug discovery.
The upshot is that these innovators are making it easier for more companies to benefit from AI technology even if they lack top technical talent, access to huge data sets, and their own massive computing power. Through the cloud, they can access services that address these shortfalls—without having to make big upfront investments. In short, the cloud is democratizing access to AI by giving companies the ability to use it now.
Cloud-based AI helps companies surmount barriers to adoption
Deloitte recently surveyed 1,900 “cognitive-aware” executives whose companies have begun to use AI for pilots and implementations. All of these companies—representing 10 industries and seven countries—can be considered “early adopters” compared with average organizations, though they are not in the same league as AI pioneers such as Amazon, Google, and the BATs. The survey found that data issues, such as accessing quality data, cleaning data, and training AI systems, were one of the two top obstacles to AI adoption, ranked as a top-three challenge by 38 percent of the surveyed companies. Integrating AI into existing processes and workflows also ranked as a top-three challenge for 38 percent of the respondents, while difficulties implementing AI—a serious problem when companies try to scale proofs of concept to full production—followed close behind at 37 percent.
Separately, we asked these early adopters about whether a “skills gap” inhibited their AI initiatives. Forty-one percent said they had a “moderate” skills gap, with an additional 27 percent calling their skills gap “major” or “extreme.” The skills gap was most acute for technical roles such as AI researchers, data scientists, and software developers.
Cloud-based software and platforms help companies benefit from AI, even if they lack the expertise to build and train systems, or to manage data on their own. And according to our survey, AI early adopters are taking advantage. Many companies that aren’t using these technologies today plan to do so in the future.
The easy path: Enterprise software with an AI infusion
Our survey of AI early adopters showed that the most popular path to acquiring AI capabilities is also the easiest: enterprise software with integrated AI. Overwhelmingly, this software is cloud-based, either through public or private cloud deployments. Fifty-eight percent of our survey respondents globally are currently using this approach. Deloitte Global estimates that by 2020, about 87 percent of AI users will get some of their AI capabilities from enterprise software with integrated AI (figure 1). 17
This method of adopting AI can have big advantages:
Companies do not need to develop their own AI applications. AI simply runs in the background, making the software more valuable to the end user.
End users do not need any specialized knowledge to use AI embedded in enterprise applications.
Companies do not need to develop intuitive, new user interfaces. This can be a challenge for AI applications developed from scratch, especially since 21 percent of respondents with skills gaps cited a shortfall of user experience designers. In fact, software firms are using AI technologies such as natural language processing to make their solutions easier to use. Salesforce, for example, recently released a voice assistant for Einstein. 18
The addition of new features such as voice assistance underscores another benefit of all cloud-based AI services: continual upgrades. Competition among AI providers is fierce; they are rapidly improving their services, and cloud-based delivery allows customers to take immediate advantage.
Companies will have an expanding range of enterprise AI services from which to choose in 2019. New cloud AI service providers are entering the market. For example, Google recently announced three AI services aimed at specific business functions such as HR and marketing, and plans to launch more. Soon, we expect that nearly all enterprise software will incorporate at least some elements of AI. 19
Companies hoping to add AI capabilities can also tap into an array of single-purpose applications, such as chatbots, that can be deployed quickly and serve as the foundation for a digital business. Lemonade, a disruptor in the insurance industry, uses chatbots to sell its policies as well as to settle claims faster and more efficiently than humans. 20
Industry-specific AI apps are also emerging—often from startups. Reflektion uses deep learning to help e-commerce sites increase sales by presenting products that match individual customers’ preferences. 21 Ayasdi develops cloud-based AI software that helps hospitals determine why insurers reject claims, suggest fixes, and identify which denied claims are worth resubmitting. 22 Such applications, though modest in scope, can help companies address thorny—and costly—problems.
However, perhaps the biggest advantage of this “easy” path is also its biggest limitation: The use cases are strictly defined by the software. On the one hand, companies don’t need to worry about whether a use case exists; the AI they buy has been developed specifically to address specific—often critical—business functions. On the other hand, these solutions offer limited customization, and the same capabilities are available to any company that uses the software. Companies that hope to gain a competitive advantage from AI will need to develop their own solutions.
AI development services: A faster track to customized solutions
That’s where cloud-based AI development services come in. 23 These include services for creating new AI applications, selecting the right models, and for getting a head start on higher-order AI technologies such as natural language processing and computer vision.
Unlike enterprise software that has AI “baked in,” AI development services require companies to have in-house technical talent, such as AI programmers and data scientists. These services can help companies get the most out of their technical talent by providing access to tried-and-true models and by accelerating key processes. In other words, they allow companies with some technical AI expertise—but not enough to develop their own AI services, or to develop them fast enough—to create a higher volume of AI services, and at scale.
For instance, it takes multiple steps to build a solution using machine learning and deep learning: building models, training the models using large data sets, evaluating the models’ performance, and “tuning” the models for optimal results. Each of these steps can be labor-intensive and require data scientists to make multiple decisions. AI development services reduce the time needed to build and test models, and to “wrangle” with data. Automated machine learning can select the most robust model out of a given set and “auto-tune” it 100 times faster than a human data scientist, allowing data science teams to produce more models with fewer steps. 24 This helps companies “test and learn” rapidly, even with only a small staff of specialists.
Some AI development services are getting so intuitive that developers don’t even need much specialized knowledge. For example, Baidu recently released an AI training platform called EZDL that requires no coding experience and works even with small data training sets. 25
Even for companies with significant resources, AI development platforms can help deliver industry-changing innovation. For example, Samsung Heavy Industries is using AWS to develop autonomous cargo vessels and the services needed to manage them. 26
Of course, as with enterprise software, there’s no need to reinvent the wheel. Cloud providers have developed pre-built machine-learning APIs for technologies such as natural language processing that customers can access instead of building their own.
The many are starting to benefit
Our survey of AI early adopters suggests that the democratization of AI is increasing AI usage. While our respondents are taking varied paths to AI, 27 enterprise software with integrated AI and cloud-based development platforms represent two major avenues for companies to access AI technology.
Among the US-based respondents to our AI survey, AI early adopters’ use of deep learning increased from 34 percent in 2017 to 50 percent in 2018. Cloud-based AI services, many of which specialize in making deep learning more accessible, contributed to this growth. In a separate survey on cloud services, Deloitte found that companies are 2.6 times more likely to prefer obtaining advanced innovation capabilities, such as AI and advanced analytics, as a service versus traditional IT. 28
As cloud technologies become more pervasive, and early adopters gain experience with them, they’re producing results:
From 2017 to 2018, US-based survey respondents, on average, increased their number of full-scale AI implementations from 6 to 9—a 50 percent jump that validated Deloitte Global’s prediction in 2018. 29
Across all countries, AI early adopters are seeing positive financial returns, reporting an average ROI of 16 percent. This is a promising start for companies that are gaining experience with a rapidly evolving set of technologies.
ROI is helping build momentum for AI, but that’s only part of why companies are adopting it. Our respondents also believe AI will have major ramifications for their competitiveness in the next two years (figure 2).
Encouraged by their successes, and betting that AI will play a critical role in enhancing their competitiveness, companies are increasing their AI investments. The companies responding to our survey spent an average of US$3.9 million on AI in 2017, a level projected to increase to US$4.8 million in 2019.
Judging by our survey results, the use of cloud-based enterprise software + AI will accelerate. Interestingly, this “easy path” is not just for beginners. The most successful AI early adopters in our study—the ones with the most internal resources—are also the biggest users of enterprise software + AI.
However, we also see that as companies mature in their AI usage, they tend to rely more heavily on AI development platforms to leverage their AI talent. AI development platform providers are seeing growth, too: According to Amazon, the number of developers using AWS for machine learning increased by 250 percent over the last year. 30
What’s clear is that AI adoption will accelerate as more services come into the market—from prepackaged enterprise AI solutions to development tools that can transform ordinary programmers into AI model builders.
Bottom line
What can companies do to emulate the success of AI early adopters?
Follow AI trends closely. The market is changing rapidly, and new capabilities are emerging. Even the most advanced techniques are becoming accessible to organizations with modest in-house AI skills. Just as the competition for market share is driving advances among tech giants and startups alike, AI early adopters are experimenting with these capabilities to leapfrog their rivals.
Get what you can “off the shelf.” AI applications focus on specific business processes, whether home-grown or available from a vendor. Where software firms have created an “off-the-shelf” solution, companies should see if it suits their needs. Don’t “reinvent the chatbot” unless it’s necessary.
Make sure to hire at least some AI experts. While enterprise software and cloud-based development platforms can provide an effective gateway to AI, they are not a substitute for having at least some technical AI talent in-house. They will not provide the competitive advantage that customized solutions can, especially as AI becomes ubiquitous in enterprise software. Companies need their own AI experts to develop and customize algorithms using AI development platforms. These experts also can help ensure that companies invest in AI applications and services that will address business needs. “Reality checks” from technical experts can become increasingly important as vendors try to pass off ordinary analytics as the latest deep learning capabilities. Internal AI experts can also help companies be realistic about what AI technologies can do for them given their current levels of talent, data access, and strategies.
Focus on the business need. The answers companies get from AI are only as good as the questions they ask. Companies should understand which challenges AI can help them solve and how it can help solve them. This requires not only technical talent, but also executives who understand business needs and can “speak data science” to technical experts. These translators can help companies ensure they’re not just building models more efficiently, but also efficiently building effective models.
Author
Jeff Loucks is the executive director of Deloitte’s Center for Technology, Media & Telecommunications. He is based in Columbus, Ohio.
Acknowledgments