Although artificial intelligence has triumphed over the world’s best human players at Go and Poker, and outperformed humans at imaging analysis and speech recognition, few are aware of the actual dollar value that AI techniques can bring to industries such as travel, retail, transport & logistics.
The McKinsey Global Institute this month released the report Notes From the AI Frontier Insights From Hundreds of Use Cases. The 36-page discussion paper surveys cutting-edge machine learning algorithms, and discusses how they can be integrated or transformed into practical applications across 19 selected industries.
AI can potentially create US$3.5 — 5.8 trillion in annual value
The report defines AI as deep learning techniques based on artificial neural networks, such as feed forward neural networks, recurrent neural networks (RNN), and convolutional neural networks (CNN). These algorithms have grown from fledgling research subjects to mature techniques in real world use. Advanced AI techniques such as generative-adversarial-networks (GANs) and reinforcement learning are not within the scope of the report.
In the 19 industries studied, AI’s potential annual value was between US$3.5 trillion and US$5.8 trillion. Retail is the industry expected to be most impacted by AI at US$0.4-0.8 trillion, followed by travel (US$0.3-0.5 trillion), and transport & logistics (US$0.4-0.5 trillion). Marketing & sales, and supply-chain management & manufacturing are sectors where AI can help companies grow US$1.2 — 2.6 trillion in annual revenue.
AI can increase value up to 128 percent over traditional analytic techniques
The report says AI is more likely to improve performance over other analytic tools in 69 percent of the use cases McKinsey studied.
The industries with most potential incremental value benefit from AI compared to analytical techniques are travel (128%), transport & logistics (89%), and retail (87%). Industries at the bottom of the list are insurance (38%), advanced electronics/semiconductors (36%), and aerospace & space (30%).
Sixteen percent of the report’s use cases are new applications developed by AI techniques, for example a smart customer service assistant in retail or medical imaging detection in healthcare.
Getting accurate labeled data to train AI models is challenging
Leveraging AI algorithms requires a large amount of clean and labeled data. The textbook Deep Learning, written by Google Researcher Ian Goodfellow and Head of the Montreal Institute for Learning Algorithms (MILA) Yoshua Bengio, suggests that a deep-learning algorithm can achieve acceptable performance by training with 5,000 labeled examples per category. If a model is expected to match or even exceed human level performance, it has to be trained with a dataset of at least 10 million labeled examples.
Collecting large-scale datasets is challenging, particularly in industries such as healthcare where there is not so much available or useable data. Also vexing is data processing, including data cleansing and labeling, which is a difficult and time-consuming engineering problem. While advanced techniques such as reinforcement learning and GANs can effectively simulate data for academic research, they are not mature enough for wider implementation.
Meanwhile, AI still has other limitations that need solutions. Interpretability, also referred to as “black box” problem, means that even scientists cannot explain how an AI arrives at a decision. Google researchers recently attempted to create a step by step visualization of the process involved in a computer recognizing an object.
Think twice before you embrace AI
McKinsey analysts suggest that AI is an elusive proposition for many companies as it remains unclear whether injecting huge investments in AI is worth the potential value the tech promises. There is also the concern that any careless technical executions could cause unpredictable, expensive or grave consequences, especially in sensitive fields like healthcare or legal systems.
The report offers stakeholders this advice:
- for AI technology providers, “understanding the potential value of AI across sectors and functions can help shape the portfolios of … AI technology companies.”
- for AI technology application, “before launching more pilots or testing solutions, it is useful to step back and take a holistic approach to the issue, moving to create a prioritized portfolio of initiatives across the enterprise.”
- for policymakers, “given the scale of the beneficial impact on business the economy and society, the goal should not be to constrain the adoption and application of AI, but rather to encourage its beneficial and safe use.”
Journalist: Tony Peng | Editor: Michael Sarazen