Undeniably, SpaceX is one of the most ambitious projects of recent decades. Improving efficiency of rocket engines, so that there is enough fuel for powered landing, is an objective that has not been accomplished by other giants of the space industry [Fortune, 2016]. Reducing the cost of a certain technology is perhaps nothing innovative, but we are talking about rocketry. Building a financially sound space business sounds exciting when we consider the benefits and opportunities resulting from space travel.
Douglas Lenat http://www.cyc.com/ had a similarly ambitious vision about 30 years ago. He attacked head-on one of the toughest problems in Artificial Intelligence i.e. how to equip computers with common knowledge. Let’s look at a few examples of common knowledge to realize that machine learning (deep learning) techniques are not a silver bullet solution to common sense knowledge acquisition.
“Once people die, they stay dead”
“Can I eat if not awake?”
“Can I remember events that have not happened yet?”
“If you cut a lump of peanut butter in half, each half is also a lump of peanut butter; but if you cut a table in half, neither half is a table.”
[Lenat, 1995], [Panton et al., 2015]
The Cyc project was created to manually enter the enormous amount of common sense fact into a database, utilize predicate logic as a representation and apply reasoning algorithms to produce answers or inferences.
Why the manual approach to knowledge extraction?
Automatic knowledge extraction, as opposed to information retrieval, has always been a very difficult task. From the perspective of 30 years, which in computing equals to centuries, Douglas Lenat (and many others scientist) was right about the scale of the problem. Common sense knowledge organized into useful ontologies is hard to generate automatically for many reasons. We have ample evidence that most spectacular language processing and cognitive computing systems of today often break down when asked simple questions that require child level common knowledge. Often, information retrieval - based techniques without deeper reasoning algorithms work well in simple circumstances (surface level semantics sometimes works) but are easy to defeat with slightly more complicated questions. This is where Cyc or similar systems will have to interject.
Cyc has been largely under the radar for some time. My guess is that it is not because the project was unsuccessful. I think that there are enormously important applications for common sense reasoning modules in natural language processing, robotics or autonomous vehicles. Perhaps new applications will require additional knowledge modules. Cyc designers might not have predicted 10 years ago that self-driving cars will become a reality so soon. Military applications of general purpose knowledge ontologies for autonomous drones are obvious too. Without extensive knowledge, as a foundation for reasoning and planning [LaValle, 2006], lethal autonomous systems will break down easily with grave consequences.
As an AI system designer and programmer, I learned firsthand how difficult (and expensive) it is to build natural language processing systems. Even for very limited domains, it takes a very considerable effort to ensure decent natural language understanding. Question answering systems belong to the most popular applications today for mobile devices. These systems will be improving quickly given the resources that major players can through at the problem.
There is no doubt that in the coming years we will be seeing revolutionary changes in the mode of interaction with our computerized devices. Cycorp might not have landed on Mars yet, but they have the right kind of rocket motors to get there. I’m sure that SpaceX could use some of that technology when they start sending robotic missions to the red planet.
Fortune Magazine, Nov 2016, The Great Rocket Race: http://fortune.com/spacex-ula-lockheed-boeing-rocket-race/
LaValle, S.M., 2006. Planning algorithms. Cambridge university press.
Lenat, D. B. (1995). CYC: A large-scale investment in knowledge infrastructure. Communications of the ACM, 38(11), 33-38.
Panton. K, Matuszek C., Lenat D., Schneider D., Witbrock M., Siegel N., Shepard B., (2015). Common Sense Reasoning – From Cyc to Intelligent Assistant; http://www.cyc.com/wp-content/uploads/2015/06/CycToIntel-1.pdf