Apple is building its artificial intelligence (AI) expertise brick by brick, with Turi being its latest AI-related acquisition. Over the last few years, Apple has also acquired VocalIQ (voice recognition) and Perceptio (image classification) in an effort to bolster its AI expertise. Turi, formerly known as Dato, is a curious addition for Apple. Turi provides a machine learning (ML) platform and is best known as an academic toolkit for teaching ML.
I have personal experience in using the Turi machine learning platform, specifically GraphLab Create, as a part of a Coursera Machine Learning specialization I completed recently. Turi was co-founded by Carlos Guesterin, an Amazon Professor of Machine Learning at the University of Washington, who is also one of the course instructors. GraphLab Create is a powerful and easy to use tool built around Python, the most widely used coding language for ML. Turi has a number of ML toolkits built in to the platform such as clustering, regression, recommenders, classifiers, and others. These toolkits allow data scientists, developers, and academics to deploy ML-based applications and web services. One of the best features of GraphLab is SFrame, which is an open-source toolkit for managing large-scale tabular data structures, similar to pandas in Python. Turi also has a commercial business focused on ML as a service, licensing its platform to companies trying to build applications and analytics using ML.
Apple has not yet articulated its plans for Turi, however here are a few thoughts on where the new assets might fit in relation to Apple’s AI plans.
- Apple is likely to integrate Turi’s ML algorithms as a part of its developer tools like Swift and XCode, which form the foundation of the macOS, iOS, watchOS, and tvOS platforms. GraphLab will allow its developers to build applications for image classification, speech analysis, text analysis, or other ML techniques. We expect to see GraphLab APIs become integrated into Apple’s developer platforms. With Turi, Apple developers will be able to empower their applications on iOS, macOS, watchOS, and tvOS with AI capabilities.
- Turi could also be used to improve iOS itself by improving energy consumption, managing applications, or improving the user interface (UI). The same could be applied to watchOS, macOS, and tvOS, although optimizing and improving iOS will be the first priority since mobile is Apple’s cash cow.
- Turi has a predictive service capability, which can be accessed through RESTful APIs. These could be used internally by Apple to improve business functions like marketing and sales using customer segmentation, sentiment analysis, or churn reduction tools. These could also be used for predictive modeling of supply chain and logistics, or to help reduce the company’s energy consumption within its facilities, which is already a priority area for Apple. Notably, DeepMind has recently announced the use of deep learning to reduce energy consumption at Google’s data centers.
On the whole, Turi does bring significant ML expertise to Apple, and its most immediate use case is empowering Apple developers with AI tools. However, even with Turi’s ML expertise in house, it’s hard to see how Apple can narrow the gap with the likes of Facebook and Google on AI. Apple’s Turi acquisition suggests that Apple is taking the path of least resistance for enabling AI, and is unwilling to spend big R&D dollars on developing AI algorithms internally. Both Facebook and Google have dedicated R&D centers for AI, which are advancing the state of the art, participating in academic conferences, and making their AI tools open source. For example, Google’s Deep Learning toolkit TensorFlow has become the most popular ML framework on Github in a matter of months.
For Apple developers who want access to the latest AI algorithms, they will be limited to Apple’s closed ecosystem, of which Turi is forming the basic foundation. Also, Turi isn’t particularly strong in deep learning applications such as image classification, speech recognition, or text recognition when compared to TensorFlow. In another example, generative adversarial networks (GANs) are the most talked about AI advancement today, which has already been implemented in TensorFlow but is unlikely to be the case for Turi. The openness of AI is what is driving its rapid evolution today. Facebook’s FAIR, Google’s Brain group and DeepMind, Amazon’s DSSTNE, Elon Musk’s OpenAI, and Microsoft AI Research are all pushing the boundaries of AI and doing it by embracing openness. Apple doesn’t participate in any of the leading AI conferences and prefers instead to perform is AI work in secret. Turi, which was partly open, especially to academics and students, is now going to become a closed framework, and rather than incorporating the best of what AI has to offer, it is likely to be left behind. Unless Apple embraces openness, especially around AI, and pours some of its $200 billion in cash into pushing AI research, I don’t see the company being able to compete in the same arena with the likes of Google, Facebook, Microsoft, Amazon, or Tesla.