How do we measure intelligence? Is it the capacity to memorise the knowledge of the world, and reproduce it at will? Is it the degree to which solutions can be provided, the speed with which we tackle complex problems?

You see, complexity is ingrained into our understanding of intelligence. The more intricate the grand picture, the more intelligent we are when we unravel the components that made it. However, this is ultimately a simplistic understanding of what intelligence is, because it is anything but a static thing.

Intelligence stands distinct from complexity dictated by complex structures. However, these structures, if static, can be read and deciphered pretty quickly. True intelligence lies in adaptability, flexibility, the capacity to comprehend changes in structures.

The ability to read flux, in patterns and phases, in creating new patterns and then applying them to existing ones – that, requires a very different kind of intelligent learning.

Objectivity is far too overrated when it comes to machines, and technology is slowly moving beyond that realm. Many machines can accurately tell you ‘what is’, giving you a neat, unbiased picture of the state of affairs.

The thermometer tells you the core temperature. The calculator crunches mathematical language and gives you pre-destined responses to questions that cannot swing the other way. The pacemaker regulates the pulses of
the heart, in accordance with certain mechanisms. The GPS shows you the way it thinks is correct and accurate after taking into account certain variables programmed in it.

Imperfection in a machine is when it does something it is not supposed to. Earlier, the machine did not (and many still do not) ‘learn’ their functions – they simply possessed them innately. A pre-programmed existence. The same calculator can throw an error when two-plus-two are divided by zero, but it cannot tell you the why of it.

If you have ever been fascinated by technology, truly fascinated that is, then you must learn to examine machine learning on both, a technical and philosophical level. The crux of this matter is then located in what we call ‘artificial intelligence’, or AI.

Nothing really new?

The AI is not a new concept. Science fiction, cyberpunk and so many tech-based genres have talked about it for decades. Alan Turing, founder of the modern computer, desired a ma chine ‘which thinks’. Stanley Kubrick’s HAL, William Gibson’s ‘Wintermute’, and Eagle Eye’s ‘Aria’ are all fearsome examples of artificial intelligence in fiction.

Yet other examples exist: the Replicant in Blade Runner and modern day television series like Black Mirror all show us the potential dangers and advantages of artificial intelligence blooming in machines.

While the next robot invasion may be long in coming, we witness remarkable attempts at creating AI today. AI here may not be as sentient or as overwhelmingly powerful as we’d have it – but Google’s subsidiary, DeepMind is making progress when it comes to applying AI to normative, everyday problems plaguing several industries.

Here, we explore DeepMind’s AI ‘AlphaGo’, the systems it operates on and the several applications of AI on the whole.

Silicon Valley is awash with AI projects

AI and optimising businesses

A lot of technology has already wormed its way into everyday business operations. From accounting, to data entries, to collation and data analysis – all of these can be handled by the machine.

We talk to more automated voices every day to get our tasks handled. Most smartphones integrate a mixture of interactive and powerful software for schedule management, entertainment and efficiency oriented tasks.

Business Intelligence has been a favourite application of technology, with company-make-or-break decisions being taken by data analysts who can accurately read what the B.I. software tells them.

The current sharing economy wave is also accompanied by the furious integration of tech where manpower is unavailable. Many start-ups operate on the power of mobile platforms and automate most of their functions – from salary payments to feedback and criticism.

Key Performance Indicators are taken very seriously when we discuss solutions for business optimisation. However, we are yet to integrate true AI into these mechanisms. No machine can yet make accurate ‘predictions’ or provide logic behind its own decision-making process.

DeepMind was founded in London, 2010 by a group of enterprising visionaries: Demis Hassabis, Shane Legg, and Mustafa Suleyman. Integrating the organic systems of neuroscience into the synthetic shell of a computer, DeepMind specialises in creating solutions that involve machine learning.

Many computers use algorithms for functionality – however, DeepMind takes it one step further. Our brains function with deep learning mechanisms – repeated, deliberate methods integrate concepts into our understanding which we can then apply.

A very large amount of information can be crystallised into one method by the human brain, resulting in automatic learning – rather than pre-programmed methods. I might be simplifying immensely, but imagine this: how do we teach children otherwise?

By mimesis, distinguishing mechanisms and then application. A dog is not a cat – and vice-versa. This is reinforced into the child’s head by observation, over and over – until they see a thousand dogs and cats. Now, our brains can do this quite easily, but it can be hard to construct algorithms that allow a machine to learn similarly.

DeepMind is attempting to solve problems using a similar, higher grade of mechanism. Here, the machine will be able to read immense amounts of data which will then be reinforced into an algorithm that it can utilise.

This can be done to improve situations, read them accurately and solve problems with the same – for almost any industry.

This is integrated with a ‘neural network’, where each unit, called a ‘neuron’ (like our brains) can weigh variables depending on the programming. Several of these make a network, with the data input on one end and the ‘solution’ on the other, depending on the desired parameters.

Neurons and deep neural networks use adaptive, complex algorithmic learning to arrive at solutions that are far more effective. From taking management decisions in real time to optimising what kind of meal would be healthy for you today – all potential solutions can be arrived at with these methods.

AlphaGo and applications

Google bought DeepMind for $650 million in 2014, indicating a significant step for the company. Since then, the AI race is on – Facebook too, is working on an AI project with Zuckerberg in full swing.

DeepMind has the edge with AlphaGo, an AI that beat the European Go champion Fan Hui. Go is a game that is more complex than chess in its scope, with ample strategy involved. With chess involving 10 to the power 60 variations, computers could beat humans by programming all of the variations beforehand.

Go is a ferociously complicated game, but DeepMind found it pretty straightforward

AlphaGo has been taught this ancient game through DeepMind’s methods – and since the game involves ‘maintaining’ control over half of the board, it requires a lot more thinking to do. Adaptive thinking, to be exact. AlphaGo managed to beat the champion 4 times out of 5, a remarkable achievement by any standard.

The team at DeepMind has utilised this repetitive form of learning for other games too – Space Invader and other such – where the ‘agent’, or the AI is given the game with all its tools without knowing what the game is, or its rules. They simulate the program repeatedly, where the AI learnt the game in over 300-500 tries and perfected it – all due to raw data input and subsequent test runs.

Now imagine this being applied to any real-world scenario. Image recognition and selection, voice sample selection, subway mapping – all of these require huge amounts of data being crunched to learn patterns and preferences. While the initial output may be nonsensical, the AI adapts and evolves to perfection.

Suleyman also mentioned that these deep learning systems are so flexible that they have now been applied for fraud detection, spam detection, hand writing recognition, image search, speech recognition, Street View detection, translation – all with spectacular improvements.

Investors like Elon Musk and venture capital firms like Horizons ventures have already recognised the worth of this tech, investing significantly in it.

The practicalities

Let us skim over two of the most practical applications that DeepMind has been able to demonstrate with AI.

DeepMind announced earlier this year, that the AI was able to increase energy efficiency for all its data centres by using a smattering of data to work out cooling systems and subsequent power consumption – giving an overall 15% reduction.

Another application is DeepMind’s tie-up with the National Health Service in the UK, where it attempts to resolve healthcare issues with smartphone apps. An app, ‘Streams’ is already being tested, where it can detect acute kidney injuries and remotely connect to healthcare systems to improve them overall.

Be it the economy, or the environment, the AI is here to stay. A lot of ethical and technological concerns will obviously be at hand before full-fledged implementation can take place. However, turning a blind eye to technology of this scope may just prove detrimental on the whole for businesses and other sectors alike.

The next step is opening up possible channels of integration for man and machine to combine forces, and generate a better standard of living in a holistic manner.