Managing Director & Senior Partner
1. AI is inductive. AI systems learn from the data and feedback that they receive in response to their earlier decisions. Their predictions and actions are only as good as the data they have been trained on. This characteristic makes AI systems very different from traditional deduction-based programming. A traditional program processes data but does not learn from it.
2. AI algorithms are simple. Core learning algorithms range from a few to several hundred lines of code. Basic AI is easy to learn, which is one reason why recent progress has been so rapid. You do not need to be a computer scientist to develop an intuitive understanding of AI. The complexity comes in applying AI to real-world problems.
3. AI works at superhuman speed and scale. Electronic signals travel about a million times faster than chemical signals in the brain do, so AI can swallow lots of data and learn and act quickly. When microseconds matter, as they do in some electronic markets, AI can be the only realistic option for participants and regulators.
4. Language and vision are accessible to AI. Among the most important recent breakthroughs in AI are machines’ abilities to interact with humans, access human knowledge, and physically navigate the real world. Although these skills remain imperfect, they are already useful in many situations—and they continue to improve quickly.
5. AI overcomes traditional complexity barriers. AI can handle both linear problems (essentially generalizations of straight lines) and nonlinear problems (everything else). This twofold ability opens up a multitude of optimization opportunities in fields such as logistics, production, and energy efficiency.
6. Submarines don’t swim. Although they rely on similar heuristics, such as trial and error, machines and humans solve tasks in different ways. The goal in business is to solve a problem, not to create robots that mimic the way humans would perform a particular job. Just as engineers do not design cars to move the way a horse does, autonomous driving should not emulate the actions of human drivers.
7. AI is hard to interrogate. In order to understand why a machine made a particular decision, you must design the system so that the machine’s decision-making process is traceable. You may also need to avoid leading-edge algorithms, such as those used in deep-learning applications. Such applications are increasingly capable of intuitive or creative answers whose origins are difficult to track.
8. Action is decentralized, but learning is centralized. AI architecture combines centralization and decentralization. For instance, self-driving cars drive autonomously, but they transmit their data to a central data center. The system then uses aggregated data from each car in the fleet to foster central-system learning, and the cars receive periodic software updates based on such learning from the central system.
9. Business value comes from data and training. Many companies do not understand the importance of data and training to AI success. Frequently, better data is more crucial to building an intelligent system than better naked algorithms are, much as nurture often outweighs nature in human beings.
10. Human-machine interactions change. Efforts to optimize interactions between humans and machines have evolved well beyond training humans to use static computer programs. Augmenting human performance with AI and vice versa—introducing a human into the loop of algorithmic problem solving—are increasingly common and complementary challenges.
The BCG Henderson Institute is Boston Consulting Group’s strategy think tank, dedicated to exploring and developing valuable new insights from business, technology, and science by embracing the powerful technology of ideas. The Institute engages leaders in provocative discussion and experimentation to expand the boundaries of business theory and practice and to translate innovative ideas from within and beyond business. For more ideas and inspiration from the Institute, please visit Featured Insights.