Part one of a multi-part series commenting on next steps in the evolution of human-machine interfaces and the rise of “apparent” machine intelligence.
We all remember The Terminator and Skynet – the evil machine intelligence that attempted to retroactively (via time travel) kill its nemesis, John Conner. http://en.wikipedia.org/wiki/The_Terminator
What we might not think about is that many of the concepts in the movies are based upon science fact, not science fiction. And, no, I’m not talking about time travel – as attractive as the concept might be.
I’m talking about the human-machine interfaces and autonomous robotics concepts that were demonstrated both directly and indirectly in the series.
In the fictional not-too-distant future, Skynet began “life” as a military expert system, designed to control robot military systems. Many of these systems were simply unmanned autonomous or remotely operated robotic killing and spying machines – not unlike what the military is developing and deploying today. The most interesting, though, was the Terminator. The Terminator was an “infiltration unit”, designed to interact with humans. Its mission was to spy on the remaining population, infiltrate their positions, and then bring in killing robots to eradicate them.
The Terminators were designed to interact with humans without being detectable as agents of Skynet. As such, they looked human, acted human, and even smelled human – “bad breath and all”. In the story, they were capable of autonomous operation within defined mission parameters. They were capable of voice-based communication, and they were capable of gesture-based communication. They used what might be called expert system technologies to interpret what was going on around them so as to fulfill their assigned individual missions.
Much of what enabled these infiltration units to operate is being developed today – albeit in an uncoordinated manner.
EXPERT SYSTEMS
IBM is perhaps preeminent in the development of expert systems that are publically visible – that is, unclassified. Their Deep Blue chess systems play at the Grand Master level. Their Jeopardy system, Watson, plays at an expert level – in a field that requires not only massive information levels, but also the ability to make sense of complex and occasionally obscure relationships.
One can be sure that IBM and other companies have developed and continue to develop expert systems for commercial and military applications. One example - airplanes that can land themselves.
A particularly interesting kind of expert system, one that most of us already use on a regular basis, is the navigation system commonly called a “GPS”. This system is tightly coupled to its operational domain. It has knowledge of location, direction, speed, and a database that describes the surrounding environment, including roads and commercial facilities. It also has access to real-time updates concerning road conditions and traffic conditions. Within its limited domain of operation, these units can make complex decisions to choose optimal navigation paths.
Perhaps more interesting is that these systems, like the other expert systems described above, are relatively self contained. They can be effectively embedded into other higher-level expert systems.
The concept of “nested” or embedded expert systems (with higher level expert systems invoking lower level “components” to solve problems in specific domains) gives us what might be the beginning of machine intelligence – at least the raw material of expertise in multiple domains and the concept of expert systems that can identify when a sub-domain is being referenced (like in Jeopardy).
MACHINE-HUMAN INTERFACES
In the Terminator movies, the ability to interact with humans was critical to the success of the evil intelligence. The infiltrators had to be able to pass something like an extended Turing Test to be successful. http://en.wikipedia.org/wiki/Turing_test
Intelligence is extremely difficult to concisely describe – like Turing, we often simplify by using a negated test that shifts the burden of definition to someone else – if we cannot tell it is not human, it therefore must be human. If we cannot tell that it is not intelligent, it therefore must be intelligent.
One key to this concept is the human interface.
In order to pass this extended Turing test concept, a machine intelligence, like the Terminator, must be able to pass for human. A successful human interface for a machine intelligence, like the Terminator, must address all aspects of human-to-human communication. It must be able to recognize and convey voice, facial, gesture, and “body language” information simultaneously. It may even convey information by sweating and smelling bad.
Of course, we are a long way from producing a machine infiltrator with this complete capability set. We are, however starting to explore many aspects of this complete human-to-human communication.
Products such as Ford SYNC provide limited voice interaction within the strictly defined domain of in-vehicle communications and vehicle-based entertainment system control. Advanced avionics systems track what a pilot is looking at and interpret commands using this information as context. University and commercial research is being done on reading and conveying facial expressions.
Predictions from NostraDavis
The seeds are being planted, with much more to come. In subsequent articles, we’ll look at what has been done and what is on the drawing board, and how it might all come together. Some predictions:
a) Expert systems for specific domains will continue to be developed and ultimately commoditized. Just as we have navigation systems in a box and chess playing programs on a chip today, we can expect advanced medical expert systems, law systems, vehicle operation, weapons systems, sensing and surveillance systems, and other specialized knowledge domains to be addressed and ultimately ‘packaged’.
b) Human interfaces are moving from theory to practice. Though simplistic, Ford SYNC is an example of an agent with control over multiple domains – in car entertainment, voice communication, navigation to name the basics. Like Watson, it has the ability to recognize the change in context from one domain to another, and to invoke the expert in that domain.
c) Products that track eye movements and facial expression are just down the road.
b) Human interfaces are moving from theory to practice. Though simplistic, Ford SYNC is an example of an agent with control over multiple domains – in car entertainment, voice communication, navigation to name the basics. Like Watson, it has the ability to recognize the change in context from one domain to another, and to invoke the expert in that domain.
c) Products that track eye movements and facial expression are just down the road.
