When IBM's Watson defeated two past champions on TV's Jeopardy! game show last February, it awoke many people to the awesome power of computing. Watson demonstrates that computers are at last becoming learning systems-capable of consuming vast amounts of information about the world, learning from it and drawing conclusions that can help humans make better decisions.
At IBM Research, we believe that learning systems will shape the future of information science and the IT industry, and that Watson represents a very significant step on that journey.
But every innovator needs a target to aim for, so, after the Jeopardy! challenge, we're searching for the next "grand challenge" to will drive the next advances in Information Technology. To help shape our thinking, we're engaging in a conversation about the future of computing with scientists and business leaders at an IBM Research Colloquium on Friday at the lab in Yorktown Heights, N.Y. The questions we're asking are straightforward: What should the next grand challenge be? How should we design it? How should we pursue it?
We want to throw a wider net, as well. The Jeopardy! contest inspired a team of IBM and university researchers to create a system that could beat the best Jeopardy! champions. What "grand challenge" would you choose? Hopefully, the colloquium and follow-up conversations will help us set an audacious goal.
The colloquium is part of an IBM centennial program designed to convene thought leaders - including leading scientists, academics, leaders of industries, public policy makers and IBM clients -- for a series of talks and panel discussions on transformational technologies and their potential impact on the world. In addition to addressing learning systems, there will be guest lectures at the colloquium about emerging, disruptive technologies that will change the computing landscape and help enable learning systems in the future -- biologically inspired nanosystems, exascale-level processing and the analysis of massive quantities of data from multiple sources.
The decision to focus on learning systems for this particular lab event emerged out of a year-long project that was connected to the IBM centennial. The leaders of IBM Research asked a group of us to look out decades into the future and identify the most important trend in computing that we believe will be a major focus of interest over that long time span. After much deliberating, we chose learning systems.
We picked this topic, in part, because of our belief that for all that computing does for us today, it doesn't yet do nearly enough. We need new systems that can become our partners in expanding the horizon of human cognition to help us navigate the increasing complexity of our globally interconnected world. Until now, most electronic computers have been based on the "calculating" paradigm. Our expanding technology frontiers are providing us with the opportunity to build a new class of systems that can learn from both structured and unstructured data, find important correlations, create hypotheses for these correlations, and suggest and measure actions to enable better outcomes for users. Systems with these capabilities will transform our view of computers from "calculators" to "machines that learn", a shift that will radically alter our expectations of what computing is and the nature of problems it should help us solve. These systems will impact virtually every sector of the economy, enabling applications and services that will range from preventing fraud and providing better security, to helping launch new products, to improving medical diagnosis.
Achieving this level of performance will require advances (and sometimes breakthroughs) in learning algorithms and architectures, expanded data input and output modalities (e.g. the ability to process text, graphs, images, video, sound, and other sensory information) and novel device technologies that will exploit the latest semiconductor and nanotechnology advances (as an example, researchers at IBM are actively working on employing phase-change-memory crossbar arrays to mimic neuronal synapses, paving the way for a new class of biologically inspired neuromorphic computation).
We believe that there will be three phases in the learning systems revolution.
The first phase will be driven by "static" learning systems. The Watson system that was built to play Jeopardy! is a good illustration of a state-of-the-art "static" learning system. The term "static" is connected to the fact that researchers had to feed information to Watson, teach it how to play the Jeopardy! game and tweak the programming when they spotted flaws in Watson's game play.
In a second phase, which we call "dynamic," the systems will constantly mine information on their own from multiple domains via multiple sources, including text, video and audio. They'll engage in deeper reasoning, taking advantage the ability to performer higher levels of semantic abstraction to better understand how pieces of information relate to one another.
The third phase would involve "autonomous" learning systems. In this phase, the systems would achieve understanding of natural language, image, voice, emotion, and other sensory information; be able to self-formulate hypotheses and generate questions across arbitrary domains; and utilize the selection of multiple algorithms to learn autonomously.
At IBM, we believe that exponential growth in our industry has been achieved by a combination of continual improvement and disruptive innovation. Today, it's time for a huge disruption-learning systems. What are your ideas? What grand challenge should we choose?