Friday, March 13, 2015

I, ROBOT - part 1


The brain is the citadel of the senses: this guides the principle of thought. - Pliny the Elder

Beginning in January, we began exploring recent developments in a long-dreamed-of field of tomorrow: Robotics. The direction of investigation was a natural extension of the birth of the IoT, where the growing ubiquity of wireless tranceivers, microcontrollers and other elements of microelectronics promise to change everything we see, do, own and touch. 

The idea of embedding human-like functionality into instruments, tools and machinery has been the inspiration for a huge body of science fiction literature and film. In fact, the notion of mechanical devices with semi-sentient qualities serving the needs of Gods and Men has been used a storytelling device for millenia, going back at least as far as the ancient Greeks. Yet for even as recently as the 20th century, the conception of such automatons has been somewhat limited, as the following clip illustrates:

Though "Forbidden Planet" was hailed in the late 50's as a highly adventurous and imaginative vision of the future, Robbie the Robot comes across as little more than a quirky and distinctly awkward contraption. Whatever 'intelligence' that one might attribute to the electro-mechanical apparatus is circumscribed by a limited and machine-like repertoire of functions and capabilities. Robbie seems, in essence, much more puppet than human-equivalent android.

For 'tis the mind that makes the body rich. - Shakespeare, "The Taming of the Shrew"

Contrast the above with the following, first broadcast a mere 12 years later in 1968:

The M5, though still sounding very stilted and mechanical in speech while performing its system control tasks with such alacrity that it makes its human masters fear their own impending obsolescence, expresses itself with actions that have an emotional root. Its behavior is flawed and destructive, yet the computer is able to use reason and objectivity to confront its sins, overcome its fear & paranoia and change its behavior. Stated differently: it acts very human.

The contrast between the two videos is definitive and defines a distinct boundary between concepts of robotic design. They illustrate the clash between design methodologies -  one of purely mechanical emulation/imitation and the other a holistic approach for robotic functionality. We can see this very clearly in the development of machine vision and voice recognition described in editorials on this blog since the beginning of the year.

In the first, robots are able to identify items from pattern matching to a library of models and recognize voice commands from a limited word list. This is, at the very most, equivalent to perhaps an insect-level intelligence.

In order for either a biological or synthetic 'system' to reach beyond such limitations, there is a hierarchy of data processing that needs to be satisfied. The interactions of these layers in the hierarchy (thru means that are still mostly unknown) is what seems to flip the switch in the mind to turn on the Light of Reason. What can we generally say about the layers of such a hierarchy?


Minds are like parachutes: they only function when open. - Thomas Dewar


This is the foundational layer of the mind, based mostly in the human brain stem and cerebellum. It is from this part of the brain that we interact with the rest of our body and thru which we receive sensations such as temperature and pain. This 'lower brain' area is also involved in various autonomic functions such as motor control, respiration and heartbeat, as well as sensor activity for vision, speech and hearing. Apparently we inherited this part of our brain from the dinosaurs.

Thru the 'lower brain', we are able to experience and absorb outside inputs and data, though not necessarily understand them. It also plays a major role in our most basic instincts - hunger, fear of predation, the mating drive, fight versus flight and so forth. Exactly how this basic set of behavioral characteristics is stored, how they manifest themselves and the complete extent to which the 'lower brain' governs them is still poorly understood.

To put it more simply: awareness is the most basic part of the mind's hierarchy, the foundational 'feature set', if you will, upon which all of a sentient creature's thoughts and behaviors are based. It helps to think of this underlying feature set of autonomic functions and basic instincts as 'firmware.' Though the ability to understand all outside inputs & data as well as influencing the preset weights and thresholds of this feature set is extremely restricted, this ground level part of the hierarchy is necessary in order to support a more complete framework for learning.


At this level, the human brain compiles inputs from the lower brain and separate sensory input and, along with being aware of this data, assigns a sense of relative importance to it. Our brain is now processing data with the intent of affecting the base weights and thresholds dynamically, as well as preparing decisions on freshly processed output. There is as yet only a very limited understanding of the mechanisms for performing this, however.


It is at this topmost level of the hierarchy, where Awareness combines with Cognition/Perception and correlates them with one's own existence and surroundings, that Mind flowers. Here both self-awareness and an understanding of unique identity are generated, as well as a comprehension of temporal situational context. The subjectivity of weights and thresholds on inputs is strongest at this level, including the suppression of automatically triggered behaviors and active control that overrides the lower brain. Since the Mind functions within a temporal context, experience acts as a modifier to weights and thresholds. In other words, the Mind permits extensive learning to modify future behavior.

Mind is the Master power that moulds and makes,
And Man is Mind, and evermore he takes
The tool of Thought, and, shaping what he wills,
Brings forth a thousand joys, a thousand ills: —
He thinks in secret, and it comes to pass:
Environment is but his looking-glass. - James Allen

Here we can see how mathematicians and researchers are drawing inferences between today's conceptions of the formation of the Mind and chaotic/nonlinear mathematical principles. From a reduced set of basic principles (instincts) with boundary conditions for growth over time, a much higher order of behaviors is generated, each being different from any other due to differences in starting conditions. In the end, the whole is obviously much greater than the sum of its parts.

Yet it is, paradoxically, part of the way the human mind functions that has prevented progress in robotics research. One of the 'software subroutines' that help organize our minds and facilitate learning is that our brains are ever sifting thru data and memories to form connections between them. The purpose of this is to discover patterns in the data.

Pattern recognition proved vital to our survival as a species. From the dawn of Homo Sapiens some 250,000 years ago up to the period shortly before the birth of agriculture around 12,000 -13,000 B.C., the earth's climate was unstable, with extreme temperature swings. By carefully observing nature and drawing inferences from the body of accumulated information, early humans were able to ascertain with a fair amount of accuracy whether and when certain plants, bushes and trees would produce edible fruits, vegetables or tubers, what the likely migration routes and congregation zones of herd animals would be, where predators were most likely to hide and so forth.

This critical feature of our brains - a capability which was key to our survival as a species in the pre-historical period - has been the primary roadblock to the advancement of robotics. Despite the nonlinear character of our minds, we have a burning need to impose order, patterns and predictability on the world around us. We desperately want to create a world that behaves in a linear manner - including our machines.

As we've seen over the last 3 months, Machine Vision and Voice Recognition are moving towards nonlinear mathematical foundations in order to capture the entire range of variables and their dynamic interactions, as well as their variations in weights, thresholds and initial conditions. The weights and thresholds are set, reset and continually reevaluated based on machine learning.

What has remained unrecognized until recently is that neither Machine Vision or Voice Recognition will ever reach their full potential if developed in isolation. Both require a more comprehensive/global learning capability - the capacity to weigh, judge, determine value or harm and adjust conceptions accordingly. Judgement thus requires the ability to interpret - an ability that changes with experience, evolving needs, desires and conditions outside of intelligence which impose on the faculty of interpretation certain drives, urges, dangers, opportunities and so forth. Thus, there needs to be an artificial version of that which governs, coordinates and shapes all of it (though certain base weights and thresholds may be so strong that it may be very difficult to override or recalibrate them permanently or even consistently.) 

What we are talking about, of course, is AI. It is this capability which must become a Reality so that we can provide that 'spark of life' which will turn inert machines into thinking beings.

What has been facilitating the revolutionary work of the last half decade in machine vision, voice recognition/activation and AI is the enormous growth in computational capability, memory and storage in individual devices & systems as well as distributed computational resources accessible thru the web. Evidently the computing power to create Asimov's robots and Philip K. Dick's androids is now available. 

There is no shortage of research initiatives dedicated to realizing such a breakthrough. Efforts are underway in the labs of Google, IBM, Qualcomm, NVidia and other private & public organizations. However, dear reader, the exploration of these will have to wait for future posts. ;-)

Dear readers,
As some of you have noticed, there are new banners on this blog for Amazon and FlipKart. If you shop on either of those sites by clicking on those banners, it will help a great deal in defraying some of the costs of maintaining this blog and providing fresh content. Many thanks to those who already have, and I hope more will join in helping to keep this blog going. :-)