Cognitive computing - AI: a once and future saga
Artificial intelligence (AI) is in the midst of a resurgence, after 20-plus years of hanging its head and seeking shelter in the darkest corners of the academy. Famous and thoughtful people are again using the term openly. Popular movies are bringing to life machine-based human companions (or competitors) with starring roles. Farmers are sitting in air-conditioned offices while their tractors drive around the fields attending to the details of precision agriculture. Intelligent flying objects are soon to deliver books and prescription bottles and more to our bedsides. Google cars and Tesla cars—and Detroit cars and German cars and soon maybe even Apple cars—will be self-driving themselves on our highways literally before we know what hit us.
Fortunately, the global media has been staging a very lively public debate on the good and the bad of this new generation of AI. Any of us who care to pay attention can access a stream of ongoing conversation about the many promising innovations emerging in AI and also about the even more numerous pitfalls involved in sharing our accustomed world domination with tomorrow’s random chips of silicon.
Earlier this year, Bill Gates, Elon Musk and Stephen Hawking each made high-profile contributions to a non-profit called the Future of Life Institute, whose programs currently focus “on potential risks from the development of human-level artificial intelligence.” They also joined with Steve Wozniak and more than 1,000 robotics researchers and artificial intelligence experts on an open letter to The International Joint Conference on Artificial Intelligence calling for a ban on offensive autonomous weapons. The nexus of their several concerns is that we will race ahead with technology innovation without pausing to consider how to control—physically or ethically—the outcomes autonomous intelligent machines could produce.
At the same time, Google AI lead Ray Kurzweil is continuing to develop his longstanding view that intelligent machines represent the next phase of human evolution, and that such machines will soon be passing and surpassing the challenges of the Turing Test and beginning to lead the progress of life on the planet. Our strategy should be to learn how to live in peace with these new “organisms” and appreciate the benefits that they will bring—in mental agility, tireless work, innovative insight, etc. While many who are bullish on AI may not go as far down that road as Kurzweil, we need to acknowledge the many positive contributions that are already emerging from even the “low-level” AI technologies we have today.
But what exactly does this AI discussion—high level and high profile though it may be—have to do with cognitive computing, you may be asking? That is an important question to bring into focus, and I will attempt in what follows to share my views on what differentiates the goals and ambitions of cognitive computing from those of AI broadly.
The emerging set of computing strategies or approaches that we refer to as cognitive computing is all about what is practical, doable and effective in practice today. Most “high-level” AI visions are not implementable today. The compute architectures, network sophistication and software intelligence required to create human-acting, “super-intelligent” systems are still very much in the lab at this stage—or perhaps not even discovered yet. Many of the core problems of AI have remained unsolved since Turing and others articulated them in the 1940s and earlier.
Computer science researchers are currently working on new distributed processing architectures and chips that closely mimic the way that the human brain processes the world around it. The net of all this activity is that, in the labs at least, prototype machines can achieve breakthrough speeds while consuming miniscule amounts of power—and perhaps we really are approaching a time when a whole new world of ambitious applications will be able to emerge and bring some of the long-held AI vision of machine intelligence to reality, however unimaginable and unpredictable it may be at the moment.
Cognitive computing is not unimaginable and rarely unpredictable. It takes advantage not only of the tremendous advances in networks and devices in the market today but also of the maturing of the multiple core software infrastructures that have been the focus of ongoing computer science research: natural language understanding, computational linguistics, voice recognition, machine vision, robotics and machine learning in its many—now well-understood if still challenging—approaches. With that broad “cognitive” toolset, grounded on the legacy of enterprise data and knowledge systems, cognitive computing applications are tackling an ever-broader collection of practical problems.
Cognitive computing is not human-level AI. But it does make new kinds of problems computable. It brings a style of interaction with users that is dynamic and adaptable to users’ engagement with the evolving context of the problems they want to solve. I think we can safely understand cognitive computing not as AI, but rather as a series of grounded and practical first steps away from a computing environment that can only take orders or look in the rearview mirror for insights.
The new cognitive computing environment is one that delivers an aggregated, loosely coupled set of technologies to make learning and proactive interaction between people and their many machines a new source of value, not an exercise in frustration. It’s a way to approach data as a valuable shared resource. It’s a way to experience problem solving as an opportunity to accelerate our very human ability to recognize patterns and take actions that have impact.