FRAMEWORK SERIES: AI – DATA ANNOTATION

We take a break this week from discussing central banks and want to concentrate on one of our framework ideas. First to define our framework we have looked at what we feel are thematic trends with strong tailwinds over the next decade. Themes like defence, healthcare, protein, materials, energy, infrastructure, cyber security, electronic payments, and AI are expected to grow substantially. Sitting in 1985, personal computers occupied 8% of US households, by 2000, due to exponential growth in PC sales, 51% owned a computer at home. Billions in sales of PCs padded coffers of leading manufacturers and investors. The trend in PC sales is well known, and investments in top 1985 manufacturers IBM, Tandy, Commodore and Apple, would have yielded mixed results over the coming 15yrs, but the dynamics of industry changed substantially. The PC thematic was great, but the right companies with the right management and products had to be picked. Our framework for the next 10yrs highlights some big themes, selecting winners and not being too early will yield results.

As we approach 2030, Artificial Intelligence will become a bigger and bigger part of our daily lives. We are not speaking about the Skynet version from the 1984 and 1991 classics Terminator 1 & 2, but we are speaking about Cortana, Siri and Alexa. Right now, these machines feel like a novelty, but they and appliances like them will become as necessary as computers. This is not a piece on how to gain access to the hardware, by buying Microsoft, Apple and Amazon, that trend is well established. This is more about the investable ideas today that make the AI future possible.

How do you train someone to recognize an image of a cat? Well you simply tell them it is a cat when they ask you what the fury animal that runs your house is. Now how do you teach a computer to recognize the same loveable dictator? That is a much harder problem to solve, as your computer does not recognize language like a human, nor can it make the association. Artificial intelligence overcomes this problem, by writing an algorithm, for example comparing image A of a cat stored in memory to image B of something else and checking if they are equal. The machine then "learns" by doing it millions and millions of times with images of cats, tigers, shower puffs, oranges, etc, and storing the results for future reference.

With the birth of commercial AI products, such as autonomous vehicles, digital assistants, and business intelligence, data volumes are growing 25% a year globally. All that data will be collected, sorted, and labelled to make the machines of tomorrow a reality. What is being done in images, is the same for voice and text. To teach machines how to think their way to a solution a real time "data rush" is under way.

Famous American Author Mark Twain did time in Nevada and California during the 1840's gold rush, and he quipped, "during the gold rush it's a good time to be in the pick and shovel business". Data annotation is the pick and shovels of AI. Why? The human brain is the most efficient machine in our known universe at processing images into responses. Training a computer to do the same thing, takes enormous amounts of data, and this data must be properly organised and labelled. This is where data annotation becomes important. Using the efficiency of the human brain, companies employ thousands of workers to look through images and label them appropriately, so that a clean data set can be presented to the AI algorithms to train them to recognise cats from dogs, and cats from people or fire hydrants due to labelling of the images.

Several companies globally exist in the data annotation space. Here in Australia the best known is Appen (APX:ASX) that counts Microsoft, Amazon, Google, etc as customers and controls a crowd workforce of over 1,000,000 globally. A true global leader in the space, they label language, text and images in native tongues with context so that AI can take flight this decade.