The Next Step in AI Is Training Machines to Think Like We Do

Deep Learning Right Now

When you think of “amazing” tasks a computer can manage, you probably think of impossibly complex calculations in rapid time, or parsing huge amounts of data—things your own mind could never manage on its own. Or maybe you think of the recent defeat of Lee Sedol at Go, a classic game of strategy, or IBM’s Watson taking on Jeopardy and winning. These more recent wins for AI were made possible in large part by deep learning, which is now opening up all kinds of possibilities for AI and the people who use it.

The simple, day to day tasks of common sense that even the human mind of a toddler manages seems to be what easily stumps AI systems: things like recognizing what kind of food is on your plate, or identifying which emotions are clouding the face of someone looking at you. These effortless tasks for the human brain were impossible challenges for machines — until now.

Deep learning techniques are gifting machines with what feels like common sense to human users. In the past, programmers would write complex algorithms that detailed everything down to the most minute possibility. This kind of explicit, deterministic algorithm is achievable when large, unwieldy calculations are the task at hand. Deep learning frees AI from these kinds of constraints, allowing the system to learn from its mistakes, remember what it has learned, and interact with users for more information.

This deep learning revolution is happening in large part because now, there is so much big data available for teaching. The human toddler can typically figure out what it needs to know after a few tries, but it takes AI many, many trials to learn the same lessons. Deep learning hinges upon access to huge amounts of data, because machines powered by AI need to base their choices on probabilities and statistical significance. As yet, there is no mechanical substitute for intuition.

Types of AI: From Reactive to Self-Aware [INFOGRAPHIC]
Click to View Full Infographic

Deep Possibilities

Advances in deep learning have already improved voice search capabilities significantly: Google replaced its Android speech system with a new deep learning based system and errors dropped by up to 25 percent overnight. Cameras using deep neural networks can now read aloud to people and understand sign language. Facebook now boasts that its deep learning based capabilities make the platform more accessible for blind users by describing photographs.

In the coming years, both big tech and a slew of startups will be using deep learning to create new products and services, as well as upgrade their existing applications. New markets and businesses will germinate and grow, fostering more innovation, services, and products. Deep learning systems will improve and become more widely available and simpler to use. The easier it is to use, the more our interactions with the technology will change.

Aditya Singh, a partner at Foundation Capital, believes that the development of a deep learning operating system will democratize deep learning and prompt the widespread adoption of practical AI. The result will be that everyday people will be able to solve real-life problems of significant magnitude using deep learning. In this sense, AI has the very real potential to be an equalizing tool, allowing people from all walks of life to engage in innovative work that can change the world.

The post The Next Step in AI Is Training Machines to Think Like We Do appeared first on Futurism.

Leave a Reply