Artificial intelligence (AI) is a fast-moving technology with a long road ahead of it. For all its surrounding hype, AI is still in its infancy. Intelligent systems have moved from simple, rule-based tasks such as spell-check into narrow machine-learning activities such as predictive maintenance and robotic process automation (RPA), but they remain far from their ultimate destiny as fully realized facsimiles of human intelligence capable of complex, contextual reasoning and strategic decision-making.
Successful companies will let business strategy lead AI deployment, not the other way around. They will focus on what their business needs, not what the technology can do. Most importantly, they will understand the extent of organizational change required as automation supplants human employees across the enterprise. Companies need new capabilities, structures, and processes to meet a new set of challenges brought on by widespread AI adoption.
What Is AI?
Even as AI generates breathless media coverage and sharply conflicting opinions about its potential impact on humankind, the technology itself is not widely understood. Artificial intelligence is a collection of digital tools that enable machines to perceive, learn, and make decisions like humans.
AI is the science of getting computers to act intelligently without being explicitly programmed. Machine learning (ML) is the subdiscipline of AI focused on using math-based algorithms and software to mimic smart actions, whose performance improves as a function of training data. There are numerous types of machine-learning algorithms available today. Deep learning, based on neural networks that mimic how brain neurons learn, is one type of algorithm that has made rapid breakthroughs recently (see figure 1).
Machine learning has accelerated in the past decade. Supervised, unsupervised, and hybrid- learning paradigms are flourishing, as neural networks based on architectures resembling the wiring of human brains enhance the ability of AI systems to compile information and apply knowledge in varying scenarios.
A combination of forces is fueling the recent surge in AI development, especially in the area of machine learning:
- Computing power. Dramatic increases in computing speeds enable systems to collect and process vast amounts of data very rapidly. Computers now process information 10 to 100 times faster, powering growth in neural network computational models.
- Costs. Costs are falling as computing speeds accelerate, improving the economics of intensive machine learning. With the cost of 1 million transistors dropping 33 percent annually, capital requirements are down and potential returns are up.
- Talent availability. Growing interest in AI among technically minded students has led universities to create programs focused on the technology. Graduates of these programs expand the talent pool for companies building automated systems.
- Cultural acceptance. Cultural barriers to AI are crumbling as consumers grow more comfortable with “smart” features in a range of everyday items. In a relatively short span of time, people have started communicating with their televisions and taking advice from Alexa, Siri, and other digital personal assistants.
The Continuing Evolution of AI
Despite its long-term promise, AI probably won’t expand beyond a relatively narrow range of business uses for the next few years. Commercial applications are emerging in five primary cognitive systems: natural language processing systems that recognize voice and text expressions; computer vision, which identifies objects, scenes, and activities; pattern recognition systems that find recurring themes in large quantities of data; reasoning and optimization technology capable of complex inferences and efficient evaluation of various options; and robotics, which integrates cognitive technologies to perform end-to-end physical and cognitive processes.
In each area, AI has moved along a spectrum of intelligence from rule-based decision-making through supervised learning and into limited applications of unsupervised learning (see figure 2). For example, natural language processing has evolved from spell-checking programs to personal assistant technologies that respond to questions. Computer vision, originally used to spot defects in fruit, now performs complex classification tasks by searching video segments or scanning invoices. Pattern recognition systems have moved from rule-based industrial inspection to narrow unsupervised learning activities such as product recommendations based on demonstrated consumer preferences. Reasoning and optimization tools have evolved from diagnosing problems in malfunctioning equipment to predicting and preventing breakdowns. Robotics, once limited to following preprogrammed instructions under rule-based RPA, is now developing the ability to predict and resolve problems in automated processes.
These advances foreshadow the potential of AI as cognitive systems develop essential characteristics of human intelligence—the ability to learn on their own in a context-aware and self- aware fashion. Such systems will gather information autonomously, evaluate it, and make decisions. They’ll provide financial advice, anticipate and avert cyberattacks, and conduct scientific research. They’ll even understand slang, sarcasm, and tone of voice.
Far-fetched? Not to the investors pouring billions into AI technologies (see figures 3 and 4). Some 5,000 AI start-ups launched since 2014 have attracted a total of $40 billion from venture capital firms and the investment arms of leading technology companies. Capital is flowing across the sector, with autonomous vehicle systems landing the most money, followed by RPA technologies.
Even as money pours in, AI faces challenges and risks that will affect the pace of commercialization and investment returns. Legal and regulatory concerns ranging from antitrust and data security issues to liability issues surrounding autonomous vehicles may slow deployment. Patent litigation among innovators could tie up key technologies, while talent shortages delay development of new applications. Though there is increasing adoption of certain solutions, there’s no guarantee that customers will embrace AI technologies as quickly as innovators bring them to market.
Any technology attracting billions in capital carries significant financial risks. Some AI bets will pay off, but others won’t. Companies that spend heavily on the wrong technologies or overpay for acquisitions will suffer. Conflicts over AI expenditures may arise, pitting innovators and company executives focused on long-term innovation against outside investors looking for quicker returns. So far, companies such as Amazon, Apple, Microsoft, Google, and Facebook have been able to fund next-generation technologies from the ample cash flows of their core businesses. But if those cash flows were to decline, shareholders might press management to cut spending on AI projects.
In the long run, these obstacles are unlikely to prevent AI from realizing its potential. But those who overlook or underestimate possible pitfalls risk painful stumbles.
The scope of investment in various AI capabilities reflects a wide range of potential business applications. Early adopters of AI in many industries are starting to reap benefits across several dimensions, from customer service and marketing to manufacturing and regulatory compliance.
We have found that AI yields worthwhile benefits in any business process and underlying activities or tasks with six key characteristics:
- High-quality data. Machine learning requires large quantities of easily accessible, heterogeneous data as a base for accumulating knowledge, recognizing patterns, and developing a set of decision-making options. However, AI is increasingly being applied to solving data quality issues, so that requirement will eventually go away.
- High frequency. AI generates acceptable returns when it’s used to reduce labor costs in recurring activities that involve significant amounts of human effort.
- Analytical complexity. AI is well suited to analyzing complex data sets, where significant computing power is necessary to generate useful insights.
- Risk mitigation. AI works best in areas where the impact of bad decisions and the learning time required to achieve acceptable error rates are well understood.
- Clear parameters. AI improves efficiency in decision-making processes with clear, quantifiable inputs and outputs.
- Ability to learn. The underlying process and task can be mastered by current machine-learning paradigms. Typically, these are activities that take a set of inputs and derive an outcome such as classification, prediction, or forecasting. More complex activities requiring context knowledge, human idiom, or emotional sensing are less amenable.
Following the AI Leaders
Not surprisingly, major technology firms are leading the way in commercializing artificial intelligence (see figure 5). Amazon, Apple, Google, Facebook, Microsoft, IBM, and other digital heavyweights see AI as an opportunity to transform not only the tech sector, but a wide swath of the economy. For these companies, maintaining a leadership position in the next wave of technological disruption is a strategic imperative. In a sign of its determination to dominate artificial intelligence, Google recently changed its rallying cry from “mobile first” to “AI first.”
Perhaps more surprising is the enthusiasm for AI among old-line industrial organizations such as GE and Ford. These companies recognize the importance of cognitive technologies in a broader digital transformation of their core businesses. Both companies aim to create new or better products while also spawning markets for services made possible by AI technology.
The leading tech firms have launched an array of AI-based applications, and provide AI-as-a- service to other companies. Most advanced are offerings based on natural language processing, text analytics, image analysis, and speech recognition.
Digital personal assistants—Apple’s Siri, Amazon’s Alexa, Google Voice, and Microsoft Cortana—use natural language processing to understand and emulate human speech. Apple’s face recognition security interface and photo organization app capitalize on image analysis. Machine learning and other AI capabilities feature prominently in new Google products such as the Pixel 2 smartphone, Clips camera, and Daydream View virtual reality system. Watson, IBM’s digital interlocutor, is a big bet on AI in business-to-business markets. And Amazon is deploying data science and AI engines and platforms in its quest to disrupt retail and other industries.
Companies outside the tech sector take a more-focused approach to AI. Automakers including BMW, Ford, Honda, Tesla, and Toyota are pumping resources into autonomous vehicle technologies likely to alter not only their products but also their business models and basic industry structures, such as General Motors’ intent to offer ride-sharing services with self-driving cars. Other manufacturers such as Bosch, GE, and Samsung are investing in machine learning, connected devices, and the Internet of Things to create service offerings that support and expand their existing business lines.
Embracing the Future
Artificial intelligence represents a challenge and an opportunity for virtually every company. Few industries can reasonably hope to avoid disruption by a technology that endows machines with human reasoning capabilities—and boundless processing power. AI will create new products, transform organizations and industries, and level playing fields in global markets. Capabilities once concentrated in a few large organizations will become widely accessible, enabling small challengers to take on entrenched industry leaders. Companies in many industries already have automated a wide range of processes. Yet AI has reached only a small fraction of its ultimate potential. The race is on, and some contenders have moved ahead, but nobody has a commanding lead yet. Companies that make the fundamental changes needed to integrate AI holistically into their strategies and operations still have a chance to win.
Simply buying AI software and implementing new technology and data management systems will not be enough. Nothing less than a complete process and reorganization around AI will do (see figure 6). Organizations that reconfigure themselves to tackle those challenges will keep pace with evolving AI technology, and reap their full share of the benefits of this transformational technology. Even at this early stage of development, automation is generating improvements in customer service, product design, and operational efficiency. Far greater rewards await those who make the right moves today.
A long journey lies ahead. But nobody has fallen behind—yet.
You can read the full length whitepaper at www.atkearney.com
By: Michael Hu, Hugo Evans, Eric Gervet, Renata Kuchembuck