AI – An evening of Insights

James set the scene by explaining that Aigen focuses on helping organisations to understand the potential of AI and its application to real world business problems. Essentially, James and his team look beyond the superficial promises of AI to concentrate on solving real, practical problems.

Aigen is ‘AI technology agnostic’ and openly partners with several AI technology providers. Probably the closest of these relationships is with Rainbird – the company’s owner and one of the country’s leading AI platform providers.

James quickly homed in on the fascinating question of: what’s new in AI? He made the point that AI is moving so rapidly that major game-changing developments now happen in a matter of weeks; not months or years. And to prove this point he reeled off some of the changes that have very recently hit the headlines:

• This very week, in Davos, the CEO of IBM announced that Watson is now reaching 1billion people around the world. Clearly, AI has come of age.

• Medical science is also making news with pioneering AI technologies that read blood results and predict death dates for heart disease patients. Indeed, IBM now claims to have 80% coverage of oncology.

• AI is also being used to optimise participation at conferences and workshops. Using an analogue of emotional intelligence, a newly-developed AI monitoring system is radically improving audience concentration and participation levels.

• AI systems are also layering learning on learning. Hassabi’s DeepMind team proved this point when their program AlphaGo recently beat Lee Sedol – the world-ranked GO champion – in a head-to-head tournament.

• A new generation of AI bots are also being developed and the promise of asynchronous communication is being headlined as the way forward.

• And streaming deep learning is now going real time… MIT has just announced that it is “reading the internet” with real time machine learning.

But James saw natural language and theme creation tools as possibly the most exciting areas of future development…

He pointed to an example (which surfaced in this week’s news) of a major social network platform turning itself inside out to become AI driven.

Similarly, in financial services, AI is addressing some of today’s greatest challenges – such as fraud identification, risk assessment and making critical lending decisions. Most importantly of all, these new AI tools can actually explain how they arrive at their conclusions. This ability to rationalise actions is becoming increasingly important both from a regulatory perspective and in terms of customer expectations.

And James’ final point was perhaps his most thought-provoking and contentious comment: “the journey to singularity is not the right journey.” James was emphatic that the focus should really be on creating collections of applied AI apps that work together to tackle complex challenges.

Ben opened with a bold assertion: “Rainbird has created the AI platform that enterprise clients have always wanted – one that delivers consistency of decision making and scalability.”

To illustrate this point, he explained how Rainbird adopts a “human down” rather than “data up” approach. It takes people’s expertise and develops tools that distil and use this knowledge. And because these tools function exactly like humans, people seamlessly and confidently connect with them. However, Ben stressed that the purpose of Rainbird AI is to support and augment the human workforce; not to replace it.

Accountability is core to Rainbird’s functionality.

Every decision this platform makes can be precisely explained – an unprecedented level of process transparency.

Ben then gave two examples of how experts now use the platform to encode their knowledge in a very natural way…

1. A major UK retail bank is now using Rainbird to build a Chatbot for bank staff. Utilising IBM Watson at the front end, it offers access to the very best advice from their most experienced colleagues (without, of course, those colleagues being present).

2. Mastercard also adopted the platform to encode expertise on their internal sales process. This is now used to drive consistency and generate huge efficiencies in the sales process.

By way of introduction, Wolfgang explained that Zuhlke is a broad-based team of engineers focused on helping clients to innovate using agile techniques and especially a subset of AI known as machine learning.

Wolfgang then demolished a myth…

The big buzzword of 2015 and 2016 was Big Data. With an unquestioning belief in the maxim: ‘the bigger the data, the better the intelligence’, companies became obsessed with data collection. But, as their data lakes grew into small oceans, the size and cost of building and processing these infrastructures spiraled out of control.

It was untenable… and totally unnecessary!

Wolfgang maintains that Big Data itself is useless. Data is only valuable when it is transformed into actionable insights and intelligence – and the more data a company has to sift, the more difficult that goal is to achieve.

Traditional approaches to dealing with this problem have proved to be difficult, expensive and very time-consuming – requiring a great deal of engineering expertise and human effort. By contrast, however, machine learning can deliver results with much less effort… and, most importantly, much less data.

Machine learning is an AI tool that Zuhlke has been utilising for several decades. It has now come of age with a set of tried and thoroughly tested techniques.

Machine learning is particularly good at classification (training the model to recognise and classify things correctly), pattern recognition (training the model to recognise regularity and routine) as well as ranking and filtering data.

Wolfgang illustrated this point with a fascinating case study involving the leasing of railway engines…

In this instance, the locomotive owners were concerned to know whether the companies leasing their costly assets were keeping to the terms of their contract. How was the engine being used…how many miles was it travelling… was it being properly maintained… and were their depreciation assumptions wildly optimistic, overly pessimistic or bang on the money?

Surprisingly, the answers were supplied from the smallest possible data sample…

Every few minutes, a locomotive transmits a fuel gauge reading and these were captured by Zuhlke and subjected to some clever and very complex assessment. Taking this tiny data sample and applying machine learning techniques, the team was able to calculate how hard the train was being worked… how many miles it travelled each day… how frequently it was being serviced… they could even (using online timetable data) deduce the routes the train was travelling. All from a series of small fuel readings.

And, in a final case study, Wolfgang showed how an equally small (and seemingly insignificant) snippet of sample data – the sound made by the opening and closing of a lift door – could predict when an elevator required maintenance. Now that is clever thinking.

Dean opened on the subject of voice technology and why its integration into the total customer and employee experience is becoming so critical…

Automatic Speech Recognition (ASR) has now achieved around 95% accuracy (i.e. it correctly transcribes words 95% of the time). This is projected to reach 99% in the next few years. And this seemingly small 4% uplift will make a greater difference than most people imagine… it will be a game-changer.

But ASR is actually the easy part. By far the hardest challenge is to understand what the words actually mean. This is the area of Natural Language Processing (NLP) that is now focusing most of Amazon’s energies and resources.

By creating more endpoints, more users and the more feedback loops, the NLP system can learn appreciably quicker. And this is where the cloud proves so powerful. It’s the crucial catalyst – the enabler. You simply cannot sell everyone the hardware and software needed to do NLP but you can make it universally accessible via the cloud…

Alexa listens to the user’s speech… converts it into text via ASR… understands the message sentiment with NLP…and then selects the relevant application to generate a response which is finally converted into speech. And it does all of this in under 0.5 of a second (courtesy of AWS).

The real challenge is to make all this as natural as possible – this is the realm of Voice User Experience. And to show just how crucial this is to the process, Dean compared the standard mobile weather app with Alexa. So, imagine you are preparing to fly to Lisbon and need to know whether to pack a raincoat…

With the mobile app, you have to access the phone, open the app, add Lisbon to the list of cities and then tap Lisbon to see the weather forecast. Five years ago – perhaps even one year ago – this would have been acceptable. But no longer – today, the process is truncated to seven words: “Alexa, what’s the weather like in Lisbon?”

Dean then went on to look at the evolution of user experience. He traced developments from the early Graphical User Interfaces (GUI) which were festooned with buttons… through to the mobile app which is touch activated (for example, the Angry Birds’ user experience had a single touch to aim, power and launch a missile. The GUI equivalent would have had a whole series of separate buttons for height, power etc)… and finally Dean brought us right up to date with the simple, staccato Voice User Experience of Alexa.

The goal is to make Alexa ubiquitous. The Alexa Skills Kit empowers people to build new Alexa skills. In turn, this gives Alexa users the chance to access ever more products and services. Examples of companies that have been early adopters of this thinking include National Rail, Just Eat, EDF and Aviva.

Dean also announced that three global car companies – VW, BMW and Ford – have all decided to make Alexa the default ASR system in their future cars.

However, the greatest challenge with Voice is to create the ultimate Voice User Experience. Amazon not only appreciates this truth, it has risen to the challenge. Amazon has launched the Alexa Prize with a $1million bounty. The cash will go to the first person who creates a Bot – using the Alexa Skills Kit – that holds a 20-minute conversation without the user realising they are talking to an AI system. That will be the moment when AI becomes utterly credible and ubiquitous…

And it won’t be long in coming!