What has happened between 20 Aug 2011 and 12 May 2017? And what next?
“Software is Eating the World” — Marc Andreessen, 20 Aug 2011
“Software Is Eating the World but AI Is Going to Eat Software” — Jensen Huang, 12 May 2017
The Marc Andreessen article originally published in the WSJ on 20 Aug 2011 presented views on the impact of a software revolution and how this revolution alters the value chain of industries that primarily exist in the physical world. Summarily, as software companies build out ‘real, high-growth, high-margin, highly defensible businesses’, the article postulates how future growth of world’s economies will change. On 17 May 2017, Nvidia CEO Jensen Huang spoke to MIT Technology about how the machine-learning revolution is just starting. So what has happened between these dates?
One. The makeup of the world’s top 5 companies, by market capitalisation, has changed. As at 2Q2011, the top 5 companies were Exxon Mobil, Apple, Petro China, ICBC and BHP Billiton . Compared to the list as at May 2017, it reads Apple, Alphabet, Microsoft, Amazon and Facebook .
Back in 2011, there was still plenty of debate on the intrinsic valuation of Silicon Valley’s technology companies simply because the perceived intrinsic value of Exxon Mobil, Petro China, ICBC and BHP Billiton is tangible. The value is tangible because their investors are able to form on their intrinsic value based on the future cashflow from the tangible assets on their balanced sheets. Even at a pinch, these tangible assets can be easily valued by the price a third party is willing to pay in an arm’s length transaction. Six years later, the debate is no longer about tangible assets but rather on quantum, sustainability and transparency of cashflow, the cash burn rate and viability of business model. In response, Google reorganized as Alphabet thus separating its core businesses from the moonshot ones and giving more financial transparency to its results .
Two. The manner we consume information has changed. Back in 2011, the desktop was primary gateway to the Internet and mobile devices are still in its infancy. In less than 6 years, mobile devices are the dominant gateway . Wrapped around this statistic is a generation of internet users in Developing Economies like China, India and Indonesia who probably never used a computer in their lives and have gone straight to mobile devices. This ubiquitous access made data abundant.
In a typical minute of 2016, 21m messages were sent on Whatsapp, 700k Facebook logins, 38k posts to Instagram, 70k hours of content were streamed on Netflix and 2.4m search queries were made on Google. Including other services not mentioned, ubiquitous access is generating volumes of structured and unstructured data at increasing velocity and with much variability, variety and veracity .
Three. It is universally accepted that those who control the flow of oil, other key commodities and capital gave them enormous powers. As at 2Q2011, the identities of the world’s top 5 most valuable companies is consistent with such conclusion. With control of the abundant data from ubiquitous access, ‘the seat of power’ shifted. “By collecting more data, a firm has more scope to improve its products, which attracts more users, generating even more data, and so on .” Technology companies benefit from the network effects of its user base and with more data, the network effects are multiplied and the defensibility of their business grows in the recurrent cycle. Extend that intuition and one might comprehend why Facebook paid a king’s ransom (~10% of its market value then) for Whatsapp or Amazon paid a 27% premium for Whole Foods.
Cue Nvidia CEO’s “Software Is Eating the World but AI Is Going to Eat Software” and the tipping point is Google DeepMind Challenge Match which was declared as a significant milestone in artificial intelligence research . The application of AI is increasingly ubiquitous and like how software alters the value chain of industries that primarily exist in the physical world, AI disrupting with a different flavor. Software relies on a rules-based approach to solving problem whilst AI solve the same problem by generalizing much like how a human being learns to complete a task like driving a car. This makes AI a ‘black box’ solution that is potentially so complex that its engineers or programmer may “struggle to isolate the reason for any single action .” An interim relief in the self-driving AI use case is a neural network, developed by Nvidia, to visually highlight what the ‘black box’ is paying attention to whilst in use .
On the other end of the use case spectrum, Google DeepMind’s Deep Q-learning algorithm can be trained to play video games like Atari’s Breakout or Catch with superhuman level results . In the Catch use case, this is achievable with 500 lines of code in Python or less depending on proficiency . And Training a machine to play Catch is conceptually no different from training your dog to play Catch but sans a Chuckit!, a bag of doggie treats and other tangible experiences.
So what next? The potential for AI and the extent it can change the world is unfathomable. In healthcare where data is messy and unstructured, the application of deep learning AI systems to analyse medical imaging can detect cancer faster and with better accuracy than human professionals . Similar systems are getting FDA approval and many more applications are in the pipeline. This does not diminish an oncologist’s work does but it highlights how AI will empower doctors to do better work. Perhaps the clear and present certainty in all this is to rethink the phase “Artificial Intelligence is no match for __________”.
 Example — Keras deep Q-network for Catch, Gulli, Antonio; Pal, Sujit. Deep Learning with Keras (Kindle Location 4032). Packt Publishing. Kindle Edition.