The progress curve of technological advancements in the world is about to hit a new tangent. The availability, accessibility and affordance of smart devices, autonomous systems and cyber-assistants has led us to reconsider the penetration of technology we want in our surroundings. These technologies represent forces very similar to those during the Industrial Revolution; only that now they will be much more impactful, fast and widespread [∂]. It has and will further, not only shake the socio-economic framework of our society, but also challenge us on a moral and ethical level. Why? Because for so long machines were complimenting the areas of human strength and physical labour. Now, they attempt to complement our notions of human intelligence.
The birth of the idea of machine intelligence, which is so visible now, dates back to the first usage of the prefix 'auto-'. Having its origins in the Greek word autos which means self or by oneself, the word now serves as an adjective to describe anything that is self 'driven' [π]. It has been a part of our vocabulary: automatic, autocorrect, automobile, autonomy, autopilot, etc. are some words which we encounter on a daily basis. Used to define a system that thinks and works on its own, its extent of control were up till now quite decipherable by the human mind. However, in the last two decades, the world has seen an unprecedented change in the way we use and understand technology. These changes prompt us to question technology’s impact in our lives, devise a better definition of the self, what it constitutes. Also freewill.
THE SYNTAX OF TECHNOLOGY
The TASCAM DR-100MKII’s interface is non-intuitive compared to other recording devices available to us like our smartphones; even though it offers more customised controls. Hence, it has a higher learning curve.
If you think carefully, there is nothing 'natural' about the way we communicate. By natural I mean language is not something we are born with (although there are researchers that would argue that it shapes our physiology and hence might get genetically propagated). At least not with the sophistication we live by[1]. Learning and understanding a new language and attaining dexterity in it takes certain amount of time and practice, before we become fluent in it. This is the case with any other technology like learning to type or drive. And while the syntax of spoken language contains utterances called phonemes, which serves as its building blocks, the syntax of driving a car contains fewer analog interaction blocks like steering the wheel, shifting the gear, pressing brakes, etc. (excluding the ‘noise’ in these systems)[2]. Because of the simplicity of this 'language’ and fewer number of permutations, the interaction (or the semantics) between the car and its driver is not as rich or nuanced as say our interaction with a computer.
I am using the word ‘language’ as an example for technology here as we all are familiar with it and use it almost all the time. And it’s the best example of technology that is available to mankind. Why? Firstly, because it has been used through ages. Secondly, it is mercurial in its usage yet stable in its form, it is highly adaptive, available and almost invisible to all its user. The technology of language is so prevalent in our society that we often fail to recognise its presence. And like all other technology, its carry certain objective (that of encoding thoughts) and is plagued with serious limitations (language anomalies). It is also capable of creating or evoking deep emotions in us. Many scientists like Lera Boroditsky also believe that language can change the way we think. In short, it is, where other technologies aim to be. Computers are also programmed using ‘languages’.
So let's talk about the limitations of language and interpolate with it the problems that will arise in any technological solution that can be proposed in the future. In doing so, we might see the problems that these 'newer' technologies like AI have/will introduce and deal with them strategically.
THE 'NEW' TECHNOLOGY
With the birth of computers and then the Internet, the scope and functionality of traditional machinery has increased exponentially. There is a new 'species' of microchips and algorithms that are not just 'doing' things for us, but have started 'thinking' for us as well [3]. That is what computer engineers and scientists around the world are calling AI or 'Artificial Intelligence' [4]. This technology, this 'secondary brain', increases the functional domain of even the most inanimate things around us. It is capable of detecting, decoding and deducing complex patterns from its host’s usage data and respond accordingly. Patterns that are sometimes not noticeable by human beings.
A typical neural network in action. Variables ‘x’ are weighted based on modelling function f(x) and yields output ‘y’.
Facebook, Netflix, Google, etc. invest highly in using these computational models to target personalised ads and curate content on your web feed. Fintech industry uses AI to predict market fluctuations, mitigate risks and calculate individuals’ credit scores.
AI research is also going on in the field of technologies that compliment traditional technologies like driverless cars. The complex activity of manoeuvring the vehicle, navigating through the traffic on road, following the traffic rules, etc. might soon be taken up by a machine. A moving car on a straight empty road is not very different from a aeroplane flying in the sky. And 'auto pilot' is relatively a well-known and trusted technology.
Given a certain values of altitude and speed (and maybe some other values), the program moves the plane from point A to point B in the sky, while the pilot could be gone away to sip a cup of tea. Similar auto feature is present in the air-conditioners which maintain a certain temperature. But what if the aeroplane (or the machine in it) could understand the flight schedule of the destination airport, extrapolate the fuel consumption of the plane, consider the position of other flights and the weather in the near vicinity and strategically speed up/slow down itself for a safer and more comfortable landing! That would be an intelligent plane. Similar logic applies for a driverless car with an added complexity (of taking pedestrians, jay-walkers, etc. into account).
A self-driving car detects objects on its way in real-time, reads and evaluates distances in between and navigates (steers and manages speed) its way through them. Over the course of many journeys the car becomes smarter at responding such obstacles and move from source to destination. Source: Youtube
Like all the problems in the world, the success of these algorithms rests in the definition of these measurable parameters. I already described a few examples above for an 'intelligent' aeroplane. But, our past experience of flying or driving or weather prediction has made us well aware of most of these formal parameters. And hence these problems are immediately solvable via AI. However, there are still various situations where AI does not have all the necessary classifiers to run its predictions and hence produces false or inaccurate results.[5] One can say that machines are yet to develop this sophisticated 'vocabulary' to understand and make sense of the world around them. But AI definitely brings them one step closer to achieving that.
A 'SMARTER' FUTURE?
According to the story of Babel, a united humanity in the generations following the Great Flood, speaking a single language and migrating westward, comes to the land of Shinar (שִׁנְעָר). There they agree to build a city and a tower tall enough to reach heaven. God, observing their city and tower, confounds their speech so that they can no longer understand each other, and scatters them around the world.
If we look at the world of computers through the lens of its language, we will notice that it is a Tower of Babel situation. All machines understand each other perfectly, they are quick in action and their creator (us humans) is overwhelmed by their progress.
The obvious and interesting question that this idea raises is - How will our world look in the future in the presence of these machines? How will AI shape the way humans learn, live, love, work and rule? Will these 'intelligent' machines usher an age of utopia or their limited scope and reliance give rise to a host of new problems? In other words, is it wise to teach the machines this 'new language'? A language capable of giving them the freedom of deducing and representing complex meanings out of a context. The answer is not easy.
Sure, the machines will get smart. But how do we define this smartness. Let's try to derive analogies from a person who has just learnt to read a new and rich language. A language that gives him access to a wider context and meaning than what he is accustomed to. Does learning this language enables free thinking in him? Can learning more about his surroundings pave a way to unlocking his inner consciousness and self-awareness? More fundamentally, does a language has the power to do so?
Twentieth century Austrian philosopher Ludwig Wittgenstein once wrote “The limits of my language means the limits of my world”. In his book, Tractatus Logico Philosophicus, Wittgenstein argued for a representational theory of language. He described this as a ‘picture theory’ of language: reality (‘the world’) is a vast collection of facts that we can picture in language, assuming that our language has an adequate logical form. This theory was known as logical positivism which advocated verificationism, a theory of knowledge which asserted that only statements verifiable through empirical observation are meaningful.
Logical Positivism (1900-40)
B.Russell, Vienna Circle, young Wittgenstein
Computers are trained nowadays to derive meaning out of a scene. This uses technologies like computer vision, image processing and machine learning.
This sounds familiar, right? Very much like the methods we use to train our computers today. But human language and its usage is very different from machine language. Our world and its meanings are not completely based on empirical data and objective truth. Abstract ideas like ‘mind’, ‘love’, ‘pain’, and other qualia based subjective experiences are not verifiable and that is where machines fail to grasp the context of the situation.
Even Wittgenstein soon realised that this picture theory of language was quite wrong. Wittgenstein’s shift in thinking, between the Tractatus and the Investigations, maps the general shift in 20th century philosophy from logical positivism to behaviourism and pragmatism. It is a shift from seeing language as a fixed structure imposed upon the world to seeing it as a fluid structure that is intimately bound up with our everyday practices and forms of life. For later Wittgenstein, creating meaningful statements is not a matter of mapping the logical form of the world. ‘In most cases, the meaning of a word is its use’, Wittgenstein claimed, in perhaps the most famous passage in the Investigations. It ain’t what you say, it’s the way that you say it, and the context in which you say it. Words are how you use them.
The Blind Man and The Lame Man, Aesop’s Tale
TLDR: So there we are. Machines in the world speak a common language. This is understandable to many humans (a.k.a programmers) also. It’s syntactical restriction causes serious limitations in their meaning making capabilities but is a necessary evil for them to function. If computers are to experience the world like the humans, their underlying language must be modeled to handled the fluidity of human languages. Else computers will look at nothing past the ‘face value’ of a situation. So it is safe to say that they will never be more ‘capable of deriving meaning out of context’ than their creator/s. Hence, its response is also a reflection of it’s creator’s thoughts.
This must also clarify the notion that machines can take over humankind is actually a question on the morality of the individual/s that wield its control. The combination of human and machine is somewhat like a fabled team of a blind man and a crippled one. One has a vision but not the means and vice versa. Together both can benefit from each other but alone they are powerless. The clarity with which the blind man moves forward depends entirely upon the ‘descriptions’ of the lame man. The blind man is slowly delegating some of his tasks to the crippled man (machine learning) but cannot make him see the world as he does.
[∂] According to a McKinsey Global Institute report, some 375 million jobs worldwide will vanish by 2030. An Oxford study in 2017 predicted job losses of up to 47% within 50 years. A report by the U.S. Bureau of Labor Statistics says a startling 1.4 million jobs in the U.S. will be gone in just eight years.
[π] A better understanding of these etymological usages can be found here: http://membean.com/wrotds/auto-self
[1] Sure we do exhibit certain movements of our body when triggered with certain emotions. Our lips curve upwards and eyes squint when we are happy, we get wide eyed and red when angry. Others human beings can see this, register it and relate to it; and in the process term it a language (or popularly ‘body language’). But these are not conscious modes of communication. And sign languages and dance forms again have to be learnt.
[2] Shannon, C. E., A Mathematical Theory of Communication Urbana, IL: University of Illinois Press, 1949 (reprinted 1998).
When designing the interaction in a car, the principles of high fidelity, low noise and ease of usage apply, which holds true for any language that works in a high risk, high reward environment. The noise might be caused by the rust in the axle, the loose steering wheel, etc.
[3] A primitive example of this species might be a calculator which solved simple arithmetical problems and nothing more. These problems were, as it turns out, faster to solve in binary representations of 1 & 0. Over the coming years, the language of computer programming and its cognitive abilities got more complex and computers nowadays can solve problems, like check the route to one’s office and play one’s favourite playlist in the car, by taking just one sentence as input. Even smarter ones can detect the contents of your personal photos and tag and sort them based on location, people and even emotions (based on smiles in the photo). We call these systems 'Weak AI' or narrow AI. They are capable of mimicking specific portions of human intelligence and have been around for a long time now. Strong AI, on the other hand, is a system or machine that is able to think the way our brains think (only difference is that they are inorganic in nature). We are yet to design a strong AI system.
[4] Instead of specifying the machine a specific parameter for event A to occur, we train the machine to define its own parameters for the occurrence of event A through a trial-n-error method. We feed the machine billions of training data to induce such intelligence. In time, the machine develops its own model to detect these events and respond accordingly. Read the history of AI here.
[5] Such is the case with stock markets were even veteran capital investors and brokers are unable to predict the next big trend accurately. Overlooking even the most trivial information or leaving out any unknown 'ingredient’ leaves an AI recipe useless. And this is very common in financial sector as there is no formal rule to decide how the market behaves. This can be regarded as the limitation of AI.
[6] One thing that sets apart a machine from a human is that machines are built for a purpose. And it is in the hands of the creator of these machines to set this purpose. A calculator cannot function as a car which cannot function as a washing machine.