On the day that Twitter acquired London machine learning startup Magic Pony Technology, Alex Wood catches up with Suranga Chandratillake, to get to the bottom of this new area of technology, and find out why large American companies are getting so excited by British machine learning startups.
Suranga is a computer scientist and entrepreneur turned investor, and is now general partner of Balderton Capital, who invested in Magic Pony Technology last year.
Alex Wood: What is machine learning?
Suranga Chandratillake: In the past, computer programmers needed to tell computers exactly what to do. The computers would then follow these instructed steps over and over. This limits computers to being dumb followers of instructions.
Machine Learning aims to help computers learn about the world for themselves, by creating a more generic algorithm. These algorithms can be shown a lot of data, and they then figure out how to react to, or process the data for themselves.
AW: How does it work?
SC: There are a number of different approaches to creating these self-learning algorithms.
Among the best known are artificial Neural Networks. These networks are modelled on the way biological neurons work. They take inputs (or signals) at one end, and allow the patterns that are ‘seen’ to affect the outputs to which they are connected.
These outputs are connected to yet more neurons, and therefore become inputs. And this continues, the original signal is propagated through multiple layers of these neurons, creating a network.
The networks get smarter by refining their connections to each other. They do this by being shown lots of pieces of signals (often hundreds of millions), over time.
This process enables them to do surprisingly human things like describe the narrative in an image or play a video game without even being taught its rules.
Are machine learning and Artificial Intelligence the same thing?
Strictly speaking, Machine Learning is a subset of Artificial Intelligence (AI). AI is a broader term that captures all manner of efforts to get computers to display a form of intelligence that is or appears to be human-like.
AI was arguably invented by Alan Turing in a particularly awesome paper written in 1950.
In general, computing owes rather a lot to Turing.
What is Machine Learning good at?
Machine learning is best at fuzzier, uncertain problems.
Traditional software is very good at dealing with highly specific jobs, and cannot deal with uncertainty.
One of the limitations of Machine Learning is that a vast volume of data is required for the algorithms to learn from.
As mentioned above, the networks get smarter by processing lots of different signals. Therefore, Machine Learning works best on problems for which there is a lot of easily accessible data.
Why are huge US technology companies acquiring these companies?
This is partly due to the massive potential for this new method of solving problems.
Microsoft’s Office suite already makes billions of dollars selling software that helps information workers read, understand, and work on fuzzy, unstructured problems. If they build Machine Learning into these tools, the users will become even more efficient.
Similarly, Google’s mission is about understanding and organising the world’s information.
Until recently this was mainly done through the (relatively) simple concept of a keyword search index that looked up where specific searched-for words appeared on the web.
With Machine Learning, Google will be able to extend that sort of technology into much fuzzier forms of understanding, and therefore find increasingly more useful results and matches.
The other reason that the large players are particularly focused on Machine Learning is that their existing scale means they have access to large datasets that very few others do.
Google has the largest index of the web on the planet, Facebook has the largest dataset of humans and their everyday lives and connections. And by recently acquiring LinkedIn, Microsoft now knows more about workers, their jobs, and their employers than anyone else.
That’s a lot of data.
Will machine learning change our lives for the better?
There are many jobs that are fuzzy enough that they still depend on human workers to undertake fairly simplistic tasks.
Good examples include things like driving a car or certain white-collar tasks like basic accounting or administrative work.
The hope of machine learning is that computers will be able to do these tasks for us over time, freeing the people involved to engage in the more interesting, more creative aspects of the job.
Is there a downside? Will the robots take our jobs?
The potential downside of the machine learning revolution is that when applied to industry, the computers may take some of the jobs that are today undertaken by people.
In this way, machine learning looks like it could impact white-collar industries in much the way robots have impacted blue-collar ones.
Proponents of machine learning say that this is a good thing – it will create a more plentiful economy with more money, and more spare time for us all to enjoy.
The naysayers suggest that it will push people out of jobs, and the wealth it creates may concentrate in the hands of few.
As is often the case with technology, I don’t imagine the industry will stop innovating.
This is the right time for society to begin to understand its implications, and to decide how to use ML to benefit as many people as possible.
Are the Robots going to be indistinguishable from humans, like in Terminator or Bladerunner?
Everything we’ve discussed so far is ‘applied’ or ‘narrow’ Machine Learning.
This is where an algorithm and lots of data are used to teach a computer to mimic one, very specific task.
Even if narrow ML gets really good at this one task, it hasn’t been trained to learn others.
Completely human-like AI will only happen when we crack General AI, which still feels quite a long way off.
The post What the heck is… Machine Learning appeared first on The Memo.