Depending on who you talk to, artificial intelligence (AI) will either be the saviour or nemesis of mankind.

Elon Musk, of PayPal, Tesla Cars and SpaceX (sometimes described as a real life Iron Man) has likened AI to “summoning a demon” – i.e. once it’s truly here it can’t be controlled. Meanwhile, Demis Hassabis (himself referred to as a superhero of AI), founder of DeepMind Technologies – a British AI start-up acquired by Google in 2014 for £400 million – is on a quest for what he calls “the meta-solution to any problem” through AI.

Debate will rightly rage about the future of AI, but what about the now? Research and development (R&D) into artificial intelligence has been going on for decades. And the result is a steady stream of AI that has been slipping into our lives – sometimes under the radar, sometimes in the full glare of TV cameras.

We park the notion of whether AI will turn out to be a god or Demon, and instead take a look at some of the application of artificial intelligence that is with us in 2016, and the R&D behind it.

Hey Siri! What’s SwiftKey?

Smartphones have been a driver of, and showcase for, so much innovation in the last ten years, and artificial intelligence is no exception. The most prominent example is Siri – Apple’s voice activated virtual assistant who will set your alarm, tell you the weather forecast, add a diary appointment and even deliver a sassy one-liner! In many ways this is the face of AI to the world. A neat, mildly useful, undoubtedly impressive gimmick, albeit with limitations.

So how does it work, where did it come from? You may be surprised to learn that Siri has military origins. It was a project funded by the US DARPA (Defense Advanced Research Projects Agency) and run by a company called SRI before Apple bought them. As an aside, DARPA are the good folks who brought you the Internet and Sat Nav among many other things.

Behind the AI in Siri are some staples of 21st century R&D: voice recognition, cloud computing, application programming interfaces for web integration. Starting with voice recognition, this is a really interesting area of research and development with many problems still to overcome. Accuracy is key and factors like accents and multiple ways of asking the same question are two of the trickiest obstacles. Of course pre-loading set phrases into Siri is an obvious starting point, but to deal with the potential level of variation, Siri also focuses on key words and their context. This allows it to guess questions if it does not understand them. It does not work all the time, as users will testify, but it’s not at all bad.

As hinted at, another aspect of Siri’s AI lies in its cloud computing capability. Siri makes a lightning quick decision as to whether to deal with your prompt in the cloud or locally on the handset. For instance, if you are asking it to simply play a song in your own music collection that will be done on the handset. If the handset can’t cope, then it’s up to the cloud. Up here, Siri will develop its answer partly based upon all the data it has collected from people asking similar questions. That’s machine learning – a key aspect of AI.

Virtual keyboards are another interesting area of artificial intelligence in smartphones. They may not have the personality of Siri, but are used far more widely, so in that sense have a bigger impact on our lives. And in case you need any further persuasion that they are big news, at the start of 2016, Microsoft paid £173 million for British virtual keyboard maker SwiftKey.

So let’s take a look at them. SwiftKey works by learning the personal typing style of the user. Picking up unique habits like nicknames or a favourite sports team for instance, and recognising common typos. Therefore, it gets better over time – a hallmark of artificial intelligence. The result? Swiftkey have saved nearly two trillion keystrokes equating to more than 23,000 years in combined typing time. There is hot competition in this sector though. Fleksy is another virtual keyboard trying to defeat the typo and save time. Their AI doesn’t only look at which buttons you press, but also the pattern in which you press them to understand what you are trying to type – it’s one step on from prediction. They claim to have the fastest keyboard in the world.

Trivial pursuits?

Computer games have long been a bastion of AI. Computer characters that respond to user behaviour to provide more of a challenge or greater realism. It is sometimes argued that this is a completely different field to industrial artificial intelligence. One expert in the field states computer game AI is better described as artificial behaviour. Characters that will respond to what the user does, but within tight parameters. Too much learning would lead to unpredictability in gameplay, potentially disrupting a game’s narrative, and could lead to impossibly difficult ‘baddies’. And where would be the fun in that?

Computer games are not the only pastimes in which AI has been making a splash. In 2011 IBM captured America’s imagination with their latest AI creation, on popular TV gameshow Jeopardy. IBM’s AI machine, Watson, was pitted against two of the most successful contestants in the show’s history – between them they had won $5 million of prize money. With viewing figures up 30% for the encounter, Watson went on to win hands down demonstrating the awesome potential of AI.

And before Watson, let’s not forget Deep Blue, another famous IBM AI machine. Deep Blue became the first computer to win a game and subsequently a match against a chess champion – Garry Kasparov. That was back in the mid-90s and was seeped in controversy at the time. For a more modern AI gaming achievement we turn our attention back to Demis Hassabis’ DeepMind.

Go is an ancient Chinese board game. At first glance, you may think it simpler than Chess – it is played with just black and white tokens, unlike the different pieces on a chess board. However, there are far more possible positions in Go. In fact, more positions than there are atoms in the universe. To put that in context in AI terms, if a computer were to try to win a game of Go using the same techniques with which Deep Blue won at chess (search “trees” if you are interested), it would take millions, maybe billions of years.

Yet DeepMind’s algorithm AlphaGo created shockwaves by defeating grandmaster Lee Sedol 4-1 earlier this month. Prior to that it had beaten the European Go champion 5-0 in a secret showdown in October 2015. And before that had a 99.8% success rate against rival computer programmes. It seems safe to say that AlphaGo has cracked Go. Some are describing the feat as being achieved through intuition which until now has been considered an exclusively human trait.

The AI techniques that AlphaGo used are called advanced tree search and deep neural networks. Complicated stuff that we won’t go into detail about here. However, Demis Hassabis expands on the fascinating story in his blog. The plan in the future is to use such software to solve real-world problems.

Down to business

So far we have had a look at AI in the context of consumer facing technology and in games of one form or another. Next up we look at some business applications that are being used right now, and we start with our old friend Watson.

After his exploits on Jeopardy, his human overlords put him to work on solving business problems around big unstructured data. Behind the term ‘AI’, Watson is using natural language processing to comprehend grammar and context. This allows him to analyse unstructured data like journals, social media posts, news articles and systems data – which accounts for 80% of all data today. Armed with this understanding of the data, the next stage is for Watson to interpret complex questions that are asked of him. Finally, he presents answers and solutions based upon the underlying evidence and information he has discovered.

If that all sounds quite abstract, think of it being applied in the fields of medicine, law or cooking. Taking medicine as an example, a consultant oncologist could take a patient’s casefile, feed it into Watson, who will then process it against medical texts, the latest journal articles and any other relevant data to give a suggestion for the best course of treatment. The consultant, would have the final say, and may well have arrived at the same conclusion without Watson, but the AI allows him to keep abreast of a vast range of ‘unstructured data’ to continuously deliver best practice. Watson is the oncologist’s cognitive assistant.

The potential of Watson is described as allowing any professional to be as good as the best professional in their field, to scaling the greatest minds to every mind. Exciting stuff!

Meet Baxter. He is six foot tall, 21 stone, oh, and he’s an industrial robot. But not your traditional kind. Baxter has a revolutionary AI interface that makes him so easy to programme that pretty much anyone can do it. You can simply drag his arms through a procedure, say moving objects from a box to a conveyor belt, and he will remember the movement and carry on doing it. Here he is in action:

You can also see him making a cup of coffee or folding a shirt – both difficult tasks for a robot. AI combined with robotics is a doubly interesting topic because they represent something of a mirror to our physical and mental selves, with all the concerns and opportunities that this entails. Safety was a key R&D consideration when developing Baxter. Previous industrial robots have to be caged from humans to lower the risk of injury, but Baxter can sense impact and force and instantly stop if a human obstructs his path. This makes him easier and cheaper to integrate in a production line. By overlaying artificial intelligence with robotics it gives the opportunity to cease the race to the bottom of the pay scale with cheap labour in manufacturing, and see humans work alongside machines to achieve competitive production lines.

Time for one last look at Big Data, and this time shine a light on British tech ‘unicorn’  Datasift. Their software VEDO takes unstructured social data and converts it into actionable business insight. It applies context to social posts that can then be routed to relevant functions in a business. For instance, using machine learning, rules based models and their classifier library Datasift can interpret a post commenting that a product is out of stock and send it to the sales team of a business… or a tweet containing a complaint and direct it to customer service. In an age when there is an ever growing array of channels through which customers can communicate with a company, Datasift want to help their clients ensure that fewer opportunities are slipping through the net.

How ForrestBrown can help

As we have seen, AI is here, it’s helping us in all manner of ways, and it is only going to become more influential.

Cloud computing technology is a boon in helping power the research, the big tech companies are hooked and are putting their money where their mouth is: spending vast sums on buying AI start-ups and hiring the best ‘human brains’ to develop the artificial ones.

ForrestBrown are already helping companies in this field to grow by identifying operations that qualify for valuable R&D tax credits. This can effectively pay for up to a third of costs incurred looking to resolve range of scientific and technological challenges that AI presents– a significant contribution to the ultimate success of your project. This benefit could then be reinvested the follow year, supporting continued innovation. As we have seen, artificial intelligence is a wide term that covers a huge number of distinct areas of research and development. We have a lot of experience working with companies involved in projects involving big data and unstructured data among other things. There is so much AI innovation going on, so whether it is data, cloud computing, voice recognition or any other field of artificial intelligence that you are involved in, talk to us to find out how we can help you ensure that you receive the hugely valuable R&D tax relief that you’re entitled to.

Related posts