What can AI do?

AI can’t do everything we humans can, but in some areas of vision and language it can match or surpass us.

Understanding what AI really is

Enterprises that want to seize AI’s strategic opportunities need a clear-eyed understanding of what it can and cannot do. AI algorithms today are very good at tasks such as recognizing patterns they have been previously trained on or predicting the items most likely to come next in a sequence governed by known rules.

But today’s AI doesn’t yet “think” the way we humans do. It does not build up a rich model of how the world works based on years of lived experience.

AI is very good at labeling objects in a visual scene, recognizing faces, transcribing or generating speech, understanding simple natural language commands, or predicting what sequence of French or Japanese words roughly corresponds to a given sequence of Chinese or English words. AI is somewhat proficient—though not truly expert—at answering simple questions about specific subjects and serving as a speech-driven front-end to advanced back-end functions like search engines.

It can also carry out certain domain-specific actions on behalf of humans—updating an appointment calendar, sending a templated email reply to a simple query, or making travel reservations. Finally, AI is learning how to perform certain more complex pattern recognitions tasks such as diagnosing diseases or understanding when an airliner engine needs maintenance.

diagnostic AI
two people planning

AI is not yet very good at solving problems it has never seen before or learning new concepts from only a few examples. It still lacks common sense. It doesn’t know how to build and use rich models of the social and psychological worlds that humans inhabit. It is not yet able to store vast quantities of contextual knowledge and apply it when relevant. Last but not least, AI is not good at human empathy or bedside manner.

Finally, there are toss-up areas where AI is making impressive progress but still falls well short of human performance. AI can beat humans at difficult games like chess and Go, but it can’t write a good novel or love song (of course most humans can’t either, but that’s another story). While AI is taking its first halting steps at autonomous locomotion, it is not yet able to cope with complex environments that humans master routinely, such as driving safely in urban traffic where human drivers, pedestrians, cyclists, and stray dogs are liable to behave unpredictably.

car and bike

The combination of clever AI algorithms and powerful computer hardware means that AI can scale its specialized talents very cheaply to levels of productivity that leave humans far behind. What is the bottom line for enterprise leaders wondering what problems AI can solve today? It is this: simple, repetitive perceptual or cognitive tasks like those described are the best candidates for automation or scaling-up with the help of AI.

Today AI is being embedded in countless business processes large and small throughout organizations of all sizes and in every industry. Many of these AI applications will be mundane, some will be revolutionary. Taken together, they will usher in what promises to be a new and lasting era of business transformation, one that may well surpass the PC and web revolutions of the last century.

Three examples of what AI is doing today

Uber logo

Uber uses cloud-based face recognition to verify the identity of tens of thousands of drivers from selfies they take in their cars.

Watch YouTube Video

Bing logo

Bing Translator produces instantaneous translations between hundreds of language pairs for millions of users.

Watch YouTube Video

DA logo

FDA-approved machine learning algorithms can now diagnose diabetic retinopathy in seconds from images that medical assistants upload to the cloud during routine clinical visits.

Read FDA News Release

AI understanding the world

Humans can perceive and understand the world and now, thanks to AI, computers can too. They can see objects and understand visual scenes. They can recognize speech, read text, understand simple questions, speak answers, and even translate between languages. Increasingly, they can leverage these basic abilities to perform more complex tasks such as diagnosing a disease by looking at a patient’s x-rays or understanding when the components in a jet engine need maintenance.

AI and perception
AI and cognition