In AI, only data matters and a few companies are grabbing it all
In ‘The Only Thing That Matters in Machine Learning is Data’, I argued that the most important asset today is data. Companies that collect the most data win. More data means more accurate systems, means better products, means more customers, means more data. This virtuous cycle is why no other company can catch up with Google in search or Facebook in social. This was a nightmare for Bing and MySpace, but not a problem for Toyota or BP. They went about their business as usual.
In the olden days (90s/00s), market boundaries were clear and competitors pretty much came from the same market. What Internet companies were doing on the Internet was a curiosity. But now Google has Project Loon, Nest, a life science division, autonomous vehicles, and scary robots. Facebook Oculus, Instagram, WhatsApp and Messenger, and ambitions in the payments. Microsoft is looking to solve machine translation in Skype. Baidu is aiming to introduce autonomous buses within the next three years. The vision and scale of these companies are like nothing the business world has ever seen.
It is now dawning on most businesses that transformation is likely to come from outside traditional markets. Real transformation in healthcare is not going to come from Siemens or Medtronic. In the transport industry, it is not going to come from Toyota or Volkswagen. Companies with machine learning talent and mountains of data will swan in with products that incumbents can’t match. It is easy to copy new car designs and smaller ultrasound devices. Much less so when it comes to personalised healthcare treatments and autonomous vehicles.
Good luck trying to copy those without data.
Natural language processing software works on data pretty much regardless of the source of said that – smartphone, games console, home appliance. Data is data and not constrained by the industry it comes from. Computer vision will analyse video, it doesn’t matter if it is video from broadcasts, Periscope, or security cameras. Expertise in machine learning and deep learning is applicable across every industry. Google, Facebook, Microsoft, Baidu, and IBM can apply their technologies across all industries. The virtuous cycle of data means it will likely play out as a winner-takes-virtually-all in each market.
OpenAI aims to distribute AI to everybody
So where does OpenAI fit in? OpenAI is a counterweight to the monopolisation of data and power that is likely to play out across industries. The company’s founders are Ilya Sutskever, a deep learning expert, and Greg Brockman, former CTO of Stripe. The company’s backers include Elon Musk, Sam Altman, Peter Thiel, and Reid Hoffman. The company has over $1 billion in commitments and as a not for profit, it is planning on giving away all research and patents. Note — research and patents, not data.
OpenAI has some serious talent and money at its disposal. The vision is to use this talent and money to distribute AI gains rather than letting it accumulate to a few powerful companies. There is no such thing as altruism in business. OpenAI is no exception. It will bring tremendous benefits to Musk, Altman, and other investors. Simply put, the company is a giant repository for data. Tesla and SpaceX can pool their data with Y Combinator companies such as Airbnb, Dropbox, and Stripe. OpenAI can offer huge data sets for engineers, equal to what Google or Facebook can offer, with the added bonus that OpenAI has a noble cause.
Investors understand that they cannot compete with Google or Facebook for data or talent. OpenAI is their play. By giving away patents and research, OpenAI will not be doing anything Facebook and Google are not already doing. Google has given away TensorFlow. Facebook has open-sourced its deep learning software for Torch and Big Sur, it’s deep learning hardware design.
To truly benefit humanity, OpenAI should share its data. Without sharing the data, OpenAI is like a chef preparing a coq au vin but not sharing the recipe. The coq au vin tastes delicious, sure, but without the ingredients, making one at home will result in garlicky overcooked chicken. Nobody wants garlicky overcooked chicken.