Apple Buys Turi: Why It Matters
In early August, Apple announced that they were purchasing a Machine Learning startup named Turi, in a reported $200M deal. This move slipped under the radar of most analysts, but it is important, both for Machine Learning and for Apple. Here’s why this deal matters.
The story begins with Carlos Guesterin, a professor at Carnegie Mellon, who created an Open Source toolkit for Machine Learning named GraphLab, which enables data scientists to easily create and deploy ML-based applications and web services. GraphLab made life easier for data scientists by addressing some of the thorny issues they face:
- The data storage engine and the machine learning libraries are combined in one package. Other solutions, such as Hadoop’s Mahout or Spark’s MLlib view data storage and data analysis as two separate steps to be handled by two separate tools, each with their own complexities and nuances. GraphLab provides a built-in graph database as well as a large-scale tabular database, and combines it with a number of built-in Machine Learning algorithms such as click-through predictors, churn predictors, sentiment analysis, fraud detection systems and recommendation engines, based on state-of-the-art AI technologies such as regression analysis and deep learning. The data storage engine and the algorithms work together seamlessly.
- Each ML algorithm must be tuned by defining values for a number of parameters. GraphLab provides reasonable default values for these parameters, and exposes key parameters so they can be tweaked by the data scientist.
- The toolkit is built around Python, the most widely used coding language for data science.
Guesterin subsequently moved to the University of Washington and turned the GraphLab OpenSource project into a company called Dato, subsequently renamed to Turi. Turi’s business model was based on licensing their toolkit to third party service companies that would use Turi’s toolkits to solve enterprise data science problems. It seems this business model was not panning out for Turi – the licensing fee for the toolkit was far too low to cover the company’s expenses. Hence the decision to sell to Apple.
Turi’s value lies in its integration of various data science databases, algorithms and tools into a single unified product. This is part of a larger trend towards consolidation in the data science market. Existing products are difficult to use, and skilled data scientists who know how to use these tools effectively are as hard to find as hens’ teeth. So, there is an emerging market for “data science in a box” products, which target a specific problem domain (e.g. retail or manufacturing) and combine a set of tools together into a simple-to-use product that lowers the bar for data scientists to get meaningful results. Turi was an excellent basis for building such products. Since we can assume that Apple will remove Turi from general availability to the marketplace, Apple’s purchase of Turi is a loss for data scientists.
What can we learn about Apple’s strategy from their purchase of Turi? Apple has always designed and built outstanding hardware, but never really saw itself as just a consumer electronics company. To differentiate itself, Apple brought the first voice-operated digital assistant to the market with the launch of Siri in 2011. Since then, other digital assistants such as Google Now, Cortana and Alexa have appeared, and Siri now seems to lag behind in terms of understanding and capabilities. Google Now can predict what you want before you even ask for it, Cortana is tightly integrated with PC business applications, and Alexa is an integral part of the smart home that can dim a lamp or turn up the air conditioning. Siri seems to be stuck in 2011, making the same comprehension mistakes and providing the same basic capabilities. That’s the most likely motivation for Apple’s purchase of Turi: to help Siri catch up with her competitors.
What do the competitors have that Apple lacks? Their secret sauce is a strong background in artificial intelligence and machine learning, along with a deep commitment to the OpenSource community. Some examples:
- Google’s AlphaGo, based on deep learning algorithms which they share with the AI community, was the first computer program to ever beat a professional player at the game of Go. Their OpenSource data science toolkit TensorFlow rapidly became the most popular ML framework on Github.
- Facebook AI Research (FAIR) Lab does state-of-the-art research in full cooperation with the academic community. As their director, Yann LeCun, says, ““If your dream is to solve AI, then Facebook— with its incredible infrastructure, rich data and top talent—is simply the most exciting place to be.”
- Amazon created the DSSTNE (pronounced Destiny) library for building deep learning engines. Amazon then released the library to the OpenSource community, so that researchers around the world can collaborate to improve it and innovate by using it in new domains.
Apple’s competitors are all pushing the boundaries of AI by embracing openness. Apple, on the other hand, has a different corporate DNA, based on a closed, walled garden of products that provide a smooth user experience by keeping out any technologies not sanctioned by Apple. Apple’s approach to AI has followed a similar pattern – Apple does not participate in any major AI conference or OpenSource AI projects. Instead, all of Apple’s AI work is done in secret, based on purchased AI companies such as machine learning startup Perceptio, speech AI firm VocalIQ, and now Turi.
It’s hard to argue with Apple’s closed approach in areas like hardware design and user experience – they have propelled Apple to its status as one of the world’s largest companies. But, AI is different, because of its rapid pace of evolution and its dependency on nascent academic research. Without ties to academia and the OpenSource world, Apple’s purchased AI companies will be hopelessly out of date within 6 months. Until Apple changes its AI approach to openness, Siri will continue to fall behind the other digital assistants, despite Apple’s splashy AI purchases.