To the power of X

At Google I/O 2017, Sundar Pichai announced Google's fundamental shift from a mobile-first to an AI-first strategy. Look at how much has happened in the twelve months since.

250 million new internet users, taking us to 4 billion, so for the first time more than half of the world's 7.6 billion people have access, and 93 percent of them reach it through a mobile phone. 360 million new social media users, taking us to 3.2 billion, with more than half of the new users coming from four countries: China, India, Indonesia and Iran. Four out of five internet users are active on social media. 175 billion apps downloaded worldwide, sixty percent more than the year before, and 86 billion dollars spent on them, more than double the year before.

The deep tech news is a plethora, and this is only a sample. IBM has started to provide quantum computing from the cloud. At CES in January, Intel unveiled a 49-qubit quantum chip, 49 being, according to Google, the threshold of quantum supremacy, where quantum devices solve problems classical computers practically cannot. In March, Google presented Bristlecone, a 72-qubit chip, so the race is on. And I cannot go without mentioning crypto: initial coin offerings have delivered three and a half times more capital to blockchain startups than venture capital has since 2017, almost 5 billion dollars.

Sundar's AI-first direction shows across the whole industry, not only at Google. We are witnessing a dramatic shift away from classical deep learning, which needs loads of data, smart data scientists and GPU-level compute, toward self-optimising and self-learning models running at the edge rather than out of the cloud. Cloud AutoML automates the design of machine learning models: in lay terms it creates and trains its own child, and applied to vision it outperformed every other computer vision system, so even a lay user can obtain an optimised custom model ready to predict. DeepMind's AlphaGo Zero became its own teacher, no longer constrained by the limits of human knowledge or available training data, and worked from first principles. Over forty days, playing 29 million games against itself, it beat every previous version of AlphaGo and became superhuman, at an Elo rating of 5185. A new breed of company, SparkBeyond among them, automates and scales the creative part of data science, and some say it disrupts the data scientist himself. And at Microsoft's Build conference this week, they demonstrated readily trained AI models deployed at the very edge of the network onto the growing army of smart devices, cameras and drones. Google calls its version AIY, artificial intelligence yourself, putting a state-of-the-art TensorFlow model for real-time image recognition into the hands of children.

We have also seen the limits of AI, and that we are playing with fire. Sadly, in March an autonomous Uber car killed a woman, the first reported fatal crash involving a self-driving vehicle and a pedestrian in the US, and it has since emerged that the car recognised an object and did not brake, because it had been tuned in favour of driving comfort. Also in March, Science published the first analysis of all verified true and false news stories distributed on Twitter between 2006 and 2017. No real surprise: false news reached far more people than the truth and spread much faster, with the top one percent of false stories reaching between one thousand and one hundred thousand people while true ones rarely passed a thousand. We have all watched Facebook over recent months, where 87 million users had their information shared improperly with Cambridge Analytica, whose chief executive had publicly demonstrated using psychographic profiling to support the Cruz campaign, and which filed for insolvency last week. And machine bias, where algorithms exhibit the bias of their creators or of their input data. The first is the familiar bias-variance tradeoff. The second has dramatic implications: in US criminal justice, a machine learning algorithm predicted high risk of reoffending incorrectly for black defendants at twice the rate it did for white ones, and another study found men six times more likely than women to see Google adverts for high-paying jobs.

Even this small set of examples from twelve months shows what a fast-changing world we live in, and I argue we live in an exponential world. It is fuelled at its roots by what Gordon Moore observed in 1965, that the number of transistors in an integrated circuit doubles approximately every two years. It is further accelerated by distinct fundamental quantum leaps in technology: self-learning AI, gene editing, blockchain, 5G and edge computing, and quantum computing.

What does exponential really mean? Go thirty linear steps of one metre each and you have moved thirty metres. Go thirty exponential steps, doubling the step size every time, and you will have circled the Earth twenty-six times. Our linear brain has been evolutionarily successful at carrying us from today to tomorrow, and it fails drastically at longer-term prediction in a world moving this fast.

One implication of exponentially rising computing power is an exponentially falling unit cost. A 3D printer cost 40,000 dollars in 2007 and now costs 100. A drone cost 100,000 dollars in 2008 and now costs a few hundred. A lidar sensor, the key enabler for self-driving cars, cost 200,000 dollars in 2009 and now costs 100. Sequencing your entire DNA cost 10 million dollars in 2007 and now costs 99 with 23andMe. Solar has fallen from 30 dollars per kilowatt hour to below 10 cents.

Companies fail because their managers are humans with a linear brain, and Kodak is the classic example. The dominant leader in photography commanded 90 percent of film sales and 85 percent of camera sales in the US in 1976. In that same year, an engineer called Steven Sasson presented the first self-contained digital camera in Kodak's boardroom. It weighed three and a half kilos and had a resolution of one hundredth of a megapixel. The managers naturally extrapolated linearly and discarded it, because it would never reach the quality of their film, and anyway it would only cannibalise a business built on paper. In 2012, Kodak filed for Chapter 11 bankruptcy protection.

The implications of our exponential world are a tremendous opportunity for all of us, companies and individuals alike, and they come in four parts.

First, abundance removes the barriers to entry, so everyone can do it. Connectivity and speed, with almost ubiquitous high-speed coverage, and spectral efficiency doubling with every mobile generation. Information, where the best of the world's knowledge can be accessed from remote parts of the world, through Khan Academy and its like. Computing power, where you no longer buy hardware but spin up a virtual server, free at entry level. Availability of tools, with open source supporting the entire DevOps value chain. And everything as a service, infrastructure, platform, software and analytics, on demand and paid as you grow.

Second, the only constant is change, and the rate of change is increasing. This is your chance. Only 14 percent of the companies in the 1955 Fortune 500 are still alive today, while new companies rise at an ever-increasing pace. A typical Fortune 500 company took 20 years to reach a billion dollars of market cap. Google, founded in 1998, took eight. Facebook, 2004, took five. Uber and WhatsApp, both 2009, took two. Snapchat, 2011, took one year. So you have to learn constantly: the half-life of business competency has dropped from 30 years in 1984 to five years today, and for technology competency it is surely no more than one.

Third, every existing company can be disrupted and no company is immune. Everyone knows the classical examples: Amazon against the retailer, Uber against the taxi, Airbnb against the hotel. What they have in common is that they established platform businesses, took the wastage and the unnecessary middlemen out of the system, and made the user experience far more pleasant. New platform enablers are emerging, and blockchain technology with Ethereum and distributed applications is on the verge of disrupting industries such as banking and finance. They go beyond the central platform players, and might even disrupt the likes of Facebook in a decentralised manner.

Fourth, the competition is no longer large organisations. It is the explosion of exponentially empowered entrepreneurs. Small teams: when WhatsApp was acquired by Facebook for 19 billion dollars in 2014, it had 35 engineers and more than 450 million users. Startups succeed with a handful of people, and unicorns typically have only a few hundred staff. Larger organisations such as ourselves empower what we call intrapreneurs to set up and run innovative businesses, and at Dialog we are shifting from hierarchical structures to sets of small, agile, cross-functional teams that recruit the best-suited domain experts: product managers, UX and analytics experts, front-end and back-end developers, QA.

It is on us to leverage the power of exponential, through the use of abundance, fast learning, constant reinvention, new technologies and tight networking between people and teams. And it is on us to apply the power of exponential to real-world problems, not only in the lab. At Dialog we have an amazing team of data scientists and data engineers working on intelligent bots for personal customer service, omnichannel customer experience optimisation, real-time recommendations, data-driven decision making, retail outlet planning, network planning and network fault prevention. We are excited to work with you all, through our innovation fund, our startup and developer programme, or directly in our core businesses.

A word on where my own passion for this comes from. In my first physics semesters at the University of Ulm I was fortunate to work as a part-time student for Professor Peter Fromherz, who had just demonstrated the first successful coupling between a neuron and a silicon transistor, and in his team I programmed self-organising maps, commonly known as Kohonen networks. For my diploma thesis I worked at Caltech in the Koch lab of neuromorphic engineering, building brain-like real-time parallel processing structures in CMOS, for motion detection among other things. At the University of Munich I then built what was then the smallest transistor, with a gate length of a few atomic layers, admittedly only working at a few tens of millikelvin, but showing quantum transport and electron spin effects that preceded the qubit. Pick your passion, and define your MTP, your massive transformative purpose.