Artificial Intelligence, Machine Learning, Evolutionary Algorithms, Neural Networks – nope, not taken from 2001 or a William Gibson novel but terminology that we’ve been hearing more and more of these days as technology moves from science fiction into science fact.
We’ve had more and more entrepreneurs come in saying that machine learning will play a part in their product roadmap these days so we recently went to the Machine Intelligence conference organised by Playfair Capital to learn more about the current status, trends and applications.
So what exactly IS machine learning and AI and what could this technology mean to you?
I think, therefore I am…
The pursuit of artificial intelligence could be said to date back to the invention of the computer although the philosophical questions of self-awareness (and what it means to be human) clearly pre-dates this.
In the 50’s, Alan Turing wrote about Machine Intelligence in his work “Computing Machinery and Intelligence”, opening with the claim, “I propose to consider the question ‘Can machines think?'” and stated that by the year 2000, machines would be capable of fooling 30% of human judges after five minutes of questioning (he wasn’t that far off as, in June 2014, it was suggested that Eugene Goodman (an AI bot) had passed the Turing Test, the test put forward by Turing that if a computer can trick a human into thinking he is communicating with another human it would represent viable proof of artificial intelligence (note – you can “talk” to Eugene here)).
Fast forward to the present day and, thanks to advances in computer processing power and lowering costs (à la Moore’s Law), machine learning and AI are going mainstream with the startup world excited by what benefits and new businesses the technology will provide.
What is Machine Learning?
Machine learning is a subset of the wider subject of Artificial Intelligence and can be said to be built upon the fields of computational statistics and pattern recognition. Essentially, it’s the science of getting computers to act without being explicitly programmed and explores the construction of algorithms that can learn from and make predictions on data¹.
Machine learning can be sub-divided into three broad categories²:
- Supervised Learning – The program is “trained” via past experiences based on a pre-defined set of “training examples”, which then facilitate its ability to reach an accurate conclusion when given new data. This can be used to map inputs to outputs to create predictive models (e.g. given the data we’ve collected on people who buy our product, what’s the likelihood of a certain person buying our product?);
- Unsupervised Learning (“Deep Learning” would fall into this category) – The program is given a bunch of data and endeavours to find patterns and relationships itself. This can be used to uncover hidden patterns in the data (e.g. given this bunch of data, what are the characteristics of the person buying our product?).
- Reinforcement Learning – a computer program interacts with a dynamic environment in which it must perform a certain goal, without a teacher explicitly telling it whether it has come close to its goal or not. Another example is learning to play a game by playing against an opponent.
The goal, ultimately, is for computers to think and solve problems for themselves. It’s interesting to see how we’ve borrowed (reverse-engineered?) from the natural world (see evolutionary algorithms and neural networks). The video below shows an example explaining how neural networks and genetic algorithms were used to make the program (MarI/O) learn how play a computer game via unsupervised learning in 24 hours!
Go ahead and watch it now – I’ll wait right here.
Machine Learning in the Real World
Well, “numerous” and “paradigm shifting” are words that spring to mind. Given that one of the foundational building blocks of machine learning is pattern recognition, we already use the technology widely in voice recognition (e.g. Apple’s Siri or Microsoft’s Cortana), spam filtering and optical character recognition. However, machine learning is now being applied to predictive analytics, automation, robotics and a whole host of other areas:
Predictive Analytics – The financial markets already leverage machine learning for algorithmic/high-frequency trading and startups like Kensho are bringing machine learning to bear on better predictive analytics for investment analysis.
Robotics – Google has been experimenting with driverless cars for some time. Baidu, the Chinese equivalent, is jumping on to the bandwagon and BMW, Nissan and a bunch of other car manufacturers are either developing their own technologies or looking to leverage off Google’s growing expertise.
Automation – automation here is not about robotics but more the possibility of replacing traditional human roles with AI i.e. programs that are able to replicate what their biological counterparts can do but faster, more efficiently, and with less error. You would think that the types of roles that could be rendered human obsolete would be manual (and is true to an extent with manufacturing automation) but what about more “professional” roles? Can an accountant be replaced by an “evolved” AI that is able to do the work at a fraction of the time and cost? Not at this time but there are aspects of roles that are being automated, for example Palantir already operates a fraud detection service.
The (Near) Future of AI (and what this means for startups)
Currently a lot of startups are looking to either build businesses around machine learning (Kensho above or The Grid – an automated AI driven website designer that is still in beta at the time of writing) or leverage the technology to boost the performance of their startups.
The question therefore is will there be any advantage or point of differentiation as the technology continues to enter the mainstream? There could well be commoditisation (open source APIs with everyone having access to the black box) and differentiation will come down to commercial factors (such as the business model and brand building) or algorithmic specificity (bespoke algorithms with access to private or the widest datasets to learn from – little surprise that Google, Baidu et al are all investing heavily in this area).
So why is all of this important now? Well, the key factor here is that there won’t be a vertical, segment or business that won’t have some kind of application of machine learning and AI over the next few years. Therefore it makes sense for entrepreneurs to understand what machine learning and AI can do now and how these tools can be used to make their businesses run better, faster, and at a lower cost or face being left behind by competitors that are more progressive.
AI and the Internet of Things
From the perspective of finding startup opportunities, it’s also worth thinking about the applications of machine learning on more subtle levels. At the conference, Dr Rand Hindi (Founder and CEO of Snips) provided some interesting insights about the potential synergy between AI and Ubiquitous Computing (aka the Internet of Things (“IoT”).
As machine learning and AI continues to improve and our world becomes increasingly connected, the possibility exists of AI helping to facilitate our lives by learning our behavioural patterns and creating more digestible information feeds presented to us at the optimal time, automatically dealing with communication archiving etc (think of a personal, personal assistant), thereby reducing the friction and inefficiencies created by our relationships with these devices – especially information overload (how much time do you spend reading notifications, checking email etc?) – resulting in MORE time pursuing the things that matter and LESS time spent staring at a screen. Who wouldn’t pay for a service like that?
The (Far) Future of AI (and what this means for all of us)
So looking even further ahead and on a wider level, what does all this mean for us and what happens if/when? machines surpass humans in general intelligence?
At the conference, there was a lot of discussion around the applications and socioeconomic impact of machine intelligence and an increasing divide between peoples expectations of what our future will look like.
“By the time we get to the 2040s, we’ll be able to multiply human intelligence a billionfold. That will be a profound change that’s singular in nature. Computers are going to keep getting smaller and smaller. Ultimately, they will go inside our bodies and brains and make us healthier, make us smarter.” – Ray Kurzweil
Looking at the outer extremes, on one side is the utopian view (which would include that of transhumanists like Raymond Kurzweil) who believe that AI will revolutionise human civilisation culminating in a technological singularity (due to occur anytime between 2029, 1,000 years from now or never depending on who you ask), a paradigm shift in human civilisation that could be represented by the birth of Artificial General Intelligence (or “Strong AI”), able to do anything a human can and responsible for its own evolution (which could be a thousand years of evolution in a year of “real” time). After this event, immortality (or at least longevity) will be within reach and technological feats we cannot comprehend (given the ‘limitations’ of our feeble biological brains) will be possible.
“With artificial intelligence, we are summoning the demon….In all those stories where there’s the guy with the pentagram and the holy water, it’s like yeah he’s sure he can control the demon….Didn’t work out.” – Elon Musk
On the other side is the dystopian view – with many prominent technological figures including Elon Musk, Bill Gates, Stephen Hawking and Nick Bostrom (author of Superintelligence) all having expressed concerns around the potential negative ramifications of AI, losing control of the machine and the existential risk with Prof Hawking stating that machines with AI could “spell the end of the human race”.
Utopia or Dystopia?
Whether you believe whether we are blessed or damned, the topic definitely raises some interesting questions. For instance, what will be the socio-economic impact of this type of technology? e.g. would driverless cars mean that Uber won’t need its drivers? If traditional roles like accounting can be replaced with machines, what does that mean for the incumbent workforce? Will large numbers of the population suddenly become unemployed / unemployable? What are the moral and ethical questions arising from self-aware AI? Should it/they have rights?
It’s easy to get lost in the possibilities but Mustafa Suleyman (Co-Founder and Chief Product Officer of Google DeepMind) provided everyone at the conference with a well needed reality check by stating that we have much more urgent existential issues that we should be dealing with at this time – for example, climate change or the fact almost a billion people don’t have access to clean water.
Being an optimist, machine learning and AI will clearly provide benefits that we are only just beginning to harness and, whatever happens, it’s too late to put the genie back in the bottle…