Algorithms
Transcription
In 2002, Google’s revenues were less than half a billion USD. Ten years later, its
revenues had jumped 125 fold, and the company was generating half a billion
USD every three days. At the heart of this staggering growth was the PageRank
algorithm, a novel and original way to rank the popularity and relevance of web
pages. Google has never gauged which page is better from a human perspective;
its algorithms simply respond to what Google has deemed to be most important
in determining what it shows first.
In mathematics and computer science, an algorithm is a self-contained step-by-
step set of operations to be performed. Algorithms perform calculation, data
processing, and/or automated reasoning tasks. It’s basically a recipe with step
by step instructions to get to the desired outcome. For an Exponential
Organization, algorithms are a way to make sense of the abundance of data,
gathered from various internal and external sources. It transforms this data into
business value. This can then be used to improve internal strategy and
operations or to deliver additional value to its customers.
You might not be aware of it, but algorithms have been around us for some time
now.As you turn on Zoom, Netflix or any other video/audio provider, ever
wonder how quickly the audio and video is transmitted? Probably not. But if you
were to go back a few years, you’d notice the abysmal difference in quality.
These providers use audio and video compression algorithms to ensure that,
whatever your connection speed is, you get the best experience possible.
Every time you type in an address on Waze or Google Maps and get directions,
they use a route-finding algorithm. When you start typing on Google search and
autofill kicks in – it’s an algorithm at work, just as when you write a message and
it gets autocorrected. When you go on Amazon and get recommendations – it is
thanks to the Amazon recommendation engine, powered by algorithms. All
these algorithms boil down to the same thing – a set of automated steps to
accomplish a task. Across all sciences, efficient algorithms are needed to
analyze enormous data sets or to select intelligently from a vast number of
possible outcomes.
There are two principles to a successful algorithm. Firstly, that it solves a
problem and secondly that it does so efficiently. And how do we know they are
efficient? Computer scientists use a technique called asymptotic analysis, which
basically compares algorithms, independent of language programmed in, or use
case, confirming if some algorithms are more efficient than others.
Let’s now look more closely on a particular set of algorithms which are at the
frontier of this new world - Machine Learning. Machine learning is a type of
artificial intelligence that uses layers of algorithms to "learn" from the data it is
processing. While artificial intelligence is more of a catch-all term for
simulating human intelligence, machine learning deals with how a
machine acquires information, understands the rules for how it should be used,
and continues improving itself over time. Successful machine learning
applications are applied in areas that involve processing large amounts of data
which today can be said about a growing number of fields. Music streaming
services use machine learning to decide which new songs or artists to
recommend and use algorithms to associate the listener’s preferences with
others that have similar musical taste.
It’s a subjective analysis that delivers the “best” results based on the quality of
available data. We say that a system is capable of “machine learning” when it
performs a function with the data given to it and gets progressively better over
time. If we take the example of customer service, we can see machine learning
being used to support customer service agents.
It shouldn’t be much of a surprise when you factor in that 81% of customers
would rather help themselves than speak with an agent. leading to a significant
preference towards self-service. Thus, we see the use of machine learning in
chatbots, virtual assistants, predictive analysis and even in content creation.
There are different kinds of machine learning such as supervised learning,
unsupervised learning and reinforcement learning. Both supervised learning
and unsupervised learning relies on existing sets of data from which the
‘machine can learn’ and create the correct algorithm to process and extract
value out of this data. The difference between the 2 is that for supervised
learning, we tag the initial set of data and closely monitor whether the
algorithm is picking up the correct traits - those that we want it to focus on.
With unsupervised learning, both the creation of the algorithm as well as the
finding of the patterns themselves is left up to the AI system.
Reinforcement learning is closest to how we as humans learn. Think of a small
baby, learning to walk through trial and error. This type of learning is suitable
for situations where we want AI to learn certain skills that are difficult, or even
impossible for us to accurately describe and understand. Key to reinforcement
learning is setting up the broad objective and then using a set of rewards to
‘motivate’ the AI towards learning the right skills and behaviours.
A great example of machine learning is also Google’s AlphaGo, a computer
programme that Google created using its own neural network that learned to
play the abstract and highly complex board game Go. AlphaGo uses a custom
developed type of machine learning called deep learning. Through this
process, it learned how to play at a level never seen before, by combining large
data sets from past Go games with its own ‘intuition’-based algorithm. This
allowed it to complement the analytics approach of other machine learning
systems with novel creativity and originality as seen in one of the games
played with Lee SeeDol in a world famous duel.
But even more impressive is AlphaGo’s successor called AlphaGo Zero, which
was given no data whatsoever and instead learned how to play Go by playing
against itself millions of times across a space of roughly 1 month. AlphaGo
Zero has then been able to beat AlphaGo by a result of 100 to 0. Following
that, the scope was expanded to include Chess, Shogi and even tackle protein
folding predictions.
To implement algorithms, exponential organizations need to follow the
following steps:
1. List all key sources of digital information available
2. Analyse in which business areas you can gain most value by automating
processes or by extracting value out of data
3. Choose the right kind of algorithm for each priority area
4. Build and test your algorithms
5. Launch and improve over time
Algorithms are usually closely linked with the Interfaces attribute as most well
designed interfaces have a set of sophisticated algorithms running behind the
scenes.
The impending explosion of data, resulting from the billions and trillions of
sensors that will soon be deployed, makes algorithms a critical future
component of every business. Given that they are faster, effective and more
scalable than human beings, algorithms are not only critical for organizations
committed to driving exponential growth, but also represent unprecedented
business opportunities going forward.
