Friday, March 27, 2015

I, ROBOT - part 3


Be stirring as the time; be fire with fire;
Threaten the threat'ner and outface the brow
Of bragging horror: so shall inferior eyes,
That borrow their behaviours from the great,
Grow great by your example and put on
The dauntless spirit of resolution. - Shakespeare, "King John"

Amongst the late period Norse sagas is the tale of Sigrid, often referred to as Sigrid the Haughty or the Proud. The name is a very old one, with meanings attributed to it that include "victory", "wisdom" and "beauty." All such meanings are woefully insufficient, though, as a description of this formidable woman of Norse heroic legend.

The tale concerns a member of Swedish nobility in the 10th century renowned for her great beauty who was also utterly ruthless, vengeful and ambitious. Many were her suitors from Norway, Denmark and Sweden, yet thru cold calculation she played them off against each other to maximum advantage, even burning alive two of her lesser suitors to send a message to the rest that only the greatest were worthy of vying for her hand.

Sigrid made it clear to kings, warriors and chieftains across Scandinavia that she would brook no aspersions to her dignity or honor. She was adept at forming alliances of convenience against her enemies, bringing foes to their death and ruin. Sigrid was also resolutely Nordic, rejecting the wave of Christianity encroaching upon Scandinavia in that era in favor of the traditions of her ancestors so as to honor their heroism and achievements.

Whether Sigrid was a real person or an amalgamation of scandinavian aristocratic women of the era is unknown. She may, in fact, have been a Polish princess who married into Swedish nobility. The historical accuracy of her saga, though, is secondary to the lessons that can be drawn from her chronicle - lessons on the acquisition and use of power. Sigrid was a woman who:

1. Faced the facts of life without fear;
2. Made hard choices firmly and with total commitment;
3. Recruited others to exercise their strength on her behalf;
4. Resolutely adhered to the methods and principles which made her strong.

What is particularly interesting and enlightening about Sigrid and her saga is that a version of her name is in widespread use today. That version is one normally used only by an affectionate parent with their daughter: Siri.


The smartphone digital assistant Siri, originally developed as an offshoot from a DARPA AI project (surprise, surprise!), came to Apple thru an acquisition in 2010 and has been implemented in the iPhone line since 2011, beginning with the iPhone 4S. It is a voice-activated service which is designed to learn the preferences of its owner thru repeated use. 

In its functional menu, Siri greatly resembles its competitors Google Now and Microsoft Cortana. It supports a wide range of languages and is found in all the latest Apple mobile computing devices. Siri is also being targeted at automotive applications.

What is interesting about Siri is the differences in concept and implementation vis-a-vis its rivals. The software design is completely oriented around learning to tailor itself to its owner. The level of emphasis Apple places on this is not evident in competitive offerings. 

Furthermore, Siri operation seems to be much more self-contained. In other words, unlike Google and Microsoft, Apple appears to want each instantiation of Siri to develop as an autonomous unit. Rather than being tied to a central server farm where the primary AI is resident and under further development - as is the case with Google Brain and Microsoft Adam - Siri treats its extended functionality (making restaurant reservations, mapping destinations and so forth) as the delegation of services to 'dumb' utilities and apps. The 'intelligence', such as it is at this point, is intended to reside solely on the user's mobile device. 

The approach is unique and intriguing. It has obvious advantages. Consider: if machines ever do develop independent, self-aware consciousnesses, everyone will feel much more comfortable having a mobile personal processor as their own unique digital 'buddy' instead of being tethered to the equivalent of Skynet.


This does pose a potential competitive weakness for Apple, though, in the form of a growing technology deficiency. After all, Microsoft and Google are using their digital assistants to drive internal AI research thru the challenges presented by an enormous volume of real world inputs and test cases. At the moment, Apple is rather blind to the potential advantages and implications of such interactions to their own nascent AI work. 

Yet one must remember that Apple is never a leader in new technology. Their strength derives from pulling together already extant capabilities in a novel and highly appealing form. 3rd parties do the heavy lifting of early development and Apple integrates it all, then skillfully adds an attractive consumer interface. This is likely to be their plan with AI as well. 

In all respects, Apple's Siri adheres to the example set by its namesake, the legendary Nordic heroine described above. Exactly how Apple's AI saga concludes is a part of the story not yet written; we will thus have to patiently wait and watch for further developments.


The fact that IBM - the creator of modern computing and, arguably, the primary driving force behind the second industrial revolution - would be interested in finding out if it can make its machines self-aware should come as a surprise to nobody. Big Blue's first public demonstration of that interest came to light in 1997, when reigning world champion Garry Kasparov lost (barely) a six game chess match against the Deep Blue massively parallel supercomputer.

However, we can safely conclude, based on deductions we made in previous editorials on the topic, that this example cannot be construed as the manifestation of an intelligent, self-aware, thinking machine. Chess is a very mathematical game with limited and bounded probabilities regarding patterns of moves and counter-moves over time and is thus an activity highly suited to rules-based mathematical analysis by an elaborate Difference Engine. 

The truth is that Kasparov's skill was not even fully tested in the contest. A human opponent would first have to overcome their trepidation in playing such a celebrated grand master of the game. Furthermore, during the course of the contest, there would have been many opportunities for Kasparov to befuddle his opponent with feints & false moves, gestures, postures and the timing of moves in order to rattle the cognitive and interpretive faculties of his opponent. Deep Blue would have been immune to all such stratagems, reducing the game to merely a computational exercise, devoid of the spirit of true contest between living beings.

Nonetheless, IBM's research efforts in AI did not end in 1997. A new system called "Watson" made its debut in 2011 when it participated in the "Jeopardy" television quiz show against the two best players in the TV program's history. Watson proceeded to wipe the floor with its rivals and won the competition handily.

There were several factors which gave Watson the advantage in the contest and led to its overwhelming victory. First of all, Watson's hardware was, as in the case of Deep Blue, a supercomputer. The architecture is based on the POWER7 superscalar multithreading CPU. 32 such cores are employed in a ten rack configuration of the Power 750 server, with 90 servers clustered. Since each core supports 4 threads and runs at 3.5GHz, the parallel computational capability of the server cluster is enormous.

The software running on this supercomputing platform is called DeepQA. It is designed to understand natural language queries to which it responds by sorting thru a vast database of encyclopedias, dictionaries, thesauri, articles and the like. An analysis and scoring system was applied to information gathered from an interpretation of the spoken query to compute a most likely answer. The general software architecture is illustrated below.

The mind is its own place, and in itself
Can make a heaven of hell, a hell of heaven. - Milton, "Paradise Lost"

On the surface, it appears that IBM has achieved much in AI development by setting humans and machines against each other in these intellectual duels. Yet after some quiet contemplation, the defects in these contests become evident. Both the Deep Blue and Watson challenges concern very restricted tasks and constrained situations. In a sense, the competition in both cases is not whether a machine can beat a man at thinking, but whether a man can beat a machine at automatic large volume parallel computational processing. Rather than building a new, independent intelligence, IBM is simply pushing the limits of emulation, where a machine is - under very carefully bounded conditions - only imitating humans. It's Robbie the Robot taken to its ultimate extreme.

Stated differently: the perspective of the researchers and their approach to the problem of creating an artificial intelligence is fundamentally wrong. Man is not simply a super-complex collection of switches, but a living being capable of dealing in the abstract and even able to profoundly alter his conclusions and decisions based on the same stimuli yet occurring under different circumstances - especially in the temporal domain. Furthermore, two people can observe the same phenomena under the same circumstances and arrive at completely different conclusions. We are, after all, not at all the biological equivalent of machines.

In other words, man's behavior is not calculable. The same calculation can yield different results, depending on a great variety of other factors. Furthermore, the outcome of those calculations cannot be mapped to a linear decision tree. We are nonlinear by nature and constitution, even if our thinking tends towards interpreting the world in a linear, cause-and-proportional-effect fashion.

Falling Short

Don't fear failure. Not failure, but low aim, is the crime. In great attempts it is glorious even to fail. - Bruce Lee

Researchers attest that they are trying to create a thinking machine - a sentient electromechanical life form, if you will - and do indeed spend some time learning from medical researchers who are delving into the mysteries of the human brain. But AI scientists are only tackling the easy stuff - the stuff with which they are familiar.

Current efforts focus on creating the electronic equivalent of neurons and synapses, or simply agglomerating computational horsepower and lines of code to emulate the speech output and multimedia processing capabilities of the mind. But they're not tackling the hard questions - ones such as the following:

1. What is the 'base feature set' or basic group of instructions that makes living things struggle to survive? The search for and identification of nutrients, contending with rivals and predators, the imperative to reproduce - none of this is really being examined by AI researchers. It's vital, as they'll have to find machine equivalents.

2. What are the higher level feature sets, and what are the rules for their interaction with the base set? Remember that we studied chaotic/nonlinear systems earlier this year and saw how, from a small set of boundary conditions or rules, extremely intricate patterns can emerge from a single, repeated input. For living things, the inputs are not simple, but rich in meaning; furthermore, there are always multiple kinds of inputs from the world external to the organism.

Let's take the simplest example of an organism: a bacteria swimming in a water droplet. Some would view such a creature as so primitive as to be insensate and hardly more than a mobile version of a grain of sand. This, however, is a very shallow understanding. 

A bacterium has multiple sensory inputs - pressure, temperature, and chemical receptors on its cell wall that are to some extent equivalent to a combined sense of taste and smell. If threatened or menaced by its environment or another creature, the microbe flees; when it detects potential food, it moves towards it; and at a certain point, when it has the proper energy stores, it will either reproduce asexually thru mitosis or, in some cases, seek out another of its kind with which to exchange genetic material. All during these activities, it constantly checks its surroundings with its senses and comes to a decision on future courses of action, dynamically adjusting its behavior to suit external and internal conditions, taking into account its relative priorities.

All of the above is a set of behaviors that are simple compared to those of humans, yet are beyond the capabilities of any AI project. This should be the very first goal of AI - to create a machine which independently behaves like a bacteria, given the necessary external stimuli. Only from such a basic level of living creature can higher order life be realized. Then, one day, instead of developing more and more sophisticated automatons that perform amazing parlor tricks, we may reach a level of synthetic organism which fits the following scenario imagined by one of science fiction's most influential writers:

Someday a human being, named perhaps Fred White, may shoot a robot named Pete Something-or-Other, which has come out of a General Electrics factory, and to his surprise see it weep and bleed. And the dying robot may shoot back and, to its surprise, see a wisp of gray smoke arise from the electric pump that it supposed was Mr. White's beating heart. It would be rather a great moment of truth for both of them. - Phillip K. Dick

Next week we'll look at two more AI developers - this time, at the most important level of functionality: the chip level.

Dear readers,
Don't forget the Amazon and FlipKart banners! Many thanks to those who already have clicked thru them to do their shopping, and I hope more will join in to help keep this blog going. :-)
p.s. If any of you would like those banners to hilite particular products or categories, please let me know.