It’s well known that Ali Baba spent most of his time while on business travels collecting treasures. What is less well known is that he also gathered a large library while he was away, eventually building one of the greatest libraries in the world. Similarly, Jack Ma, the founder of Alibaba, has invested heavily in AI research and development. His recent comments at the SXSW Festival in Austin, Texas are a good indicator of his continued interest in the subject:
“AI is going to change our world, and we need to be thinking about how we can use it for good. There is a lot of AI being developed for national security and law enforcement use, but there also needs to be a lot of AI being developed for good. Social good. We should not fear AI. We should try and use it wisely, because, as human beings, we have a lot to learn from it.”
In keeping with this theme, let’s take a quick look at what Alibaba is up to in the realm of AI.
Why AI, And Not VR Or ELSA)?
VR and ELSA (enterprise-focused AI) each have their place in the AI picture, but Alibaba is clearly laying the groundwork for something more comprehensive. What is it that they see that we don’t?
First, and most obviously, they have honed their AI strategy to deliver fantastic results with minimum cost and maximum speed. In his keynote speech, Ma pointed out that the Chinese search engine Baidu spends 400 man-years per week just to keep up with the country’s high consumer demand for search engines. This is in comparison to the 20 man-years that would be required to develop a new search engine from scratch.
One would assume that as a company primarily focused on commerce, Alibaba would want to own a part of the process from start to finish. The reality, though, is that they largely outsource their AI work to contract research vendors and even then, only a fraction of the overall development work is done in-house. This strategy not only keeps costs down and allows for maximum flexibility, but it also means that they can be much more agile. When you outsource the core technology, you are free to iterate and refine the approach as you see fit.
Alibaba’s AI Playlist
Alibaba’s AI strategy is all about creating compelling and valuable products that people want to buy from their website. To that end, they maintain a playlist on their YouTube channel that highlights various projects and programs that they are working on at any given time in the AI space. As with any playlist on YouTube, it’s a mix of old and new videos, with the newest videos at the top. Currently, the playlist includes a variety of projects that span the whole AI spectrum, from natural language processing (NLP) and speech recognition, to computer vision (CV) and machine learning (ML).
For instance, you can look at the playlist’s description for a breakdown of the contents:
“This playlist covers current trends and technologies in AI. From deep learning to NLP, sequence prediction to search, we dive into the field’s major advancements, from theoretical improvements to practical applications. Finally, we’ll shed light on the societal implications of AI and how engineers can have a positive impact with their work.”
The playlist also provides viewers with an overview of the state of the AI field and the various challenges faced by researchers, inventors, and manufacturers alike. Below, we will run down the latest projects and developments that have made it onto the playlist.
The first video on the playlist is “Deep Learning Through Applied AI,” an overview of a research project that combines deep learning with AI techniques from information retrieval to automated classification to create a comprehensive suite of AI tools for business (https://alibaba.github.io/DL-Ai-Business/).
In this project, the researchers trained a machine learning algorithm – a deep neural network – to perform content selection and classification on large volumes of unstructured data. Content selection entails identifying the most important and relevant content in a given corpus of text, while content classification entails assigning a category (or class) to a piece of content, based on its context in a given document.
The researchers then applied these AI tools to improve the performance of their websites, increase conversion rates, and drive down costs. To give you some idea of how this might work in practice, here’s an example of a tool created for this project, which searches for the terms ‘airline’ and ‘travel’ in the UNWTO Global Outlook Travel & Tourism Report 2017 and categorizes the search results according to the ‘airline market share’ and ‘destination market share’ fields, contained within the report.
While not all AI tools will be applicable to every business scenario, the potential to apply these types of solutions to any situation is substantial.
Speech And Language Processing
The second video on the playlist is “Speech And Language Processing With Deep Learning And Microsoft Azure,” an overview of a research project that teaches a computer system to recognize spoken words and phrases (https://alibaba.github.io/speech-nlp-azure/).
This project combined deep learning and Microsoft Azure, a cloud computing platform, to achieve real-time word recognition, speech understanding, and text-to-speech synthesis in Chinese.
In the video, the researchers introduce the various modules that make up this AI solution, which consists of a natural language processor (NLP), an HMM-based speech recognition system, and a text-to-speech (TTS) synthesizer. They also describe how these modules work together to provide robust speech recognition and understandment capabilities.
NLP is a broad field that covers the various methods that computers use to identify and decode human language. In short, NLP provides the tools for computers to understand the content of human language and speak back to us in our own tongue. As with any other form of AI, the applications for NLP are endless, from simple to complex, automated text messaging and email communication, to more elaborate tasks such as dictation and transcription, content analysis, and semantic search. In addition to the applications mentioned above, NLP has also been used to create intelligent personal assistants like Microsoft’s Cortana and Baidu’s Xiaoice.
The third video on the playlist is “Automated Classification Using TensorFlow,” an overview of a research project that teaches computers to classify images and videos into one of several predefined categories (https://alibaba.github.io/automat-classify/).
In this project, the researchers used TensorFlow, an open-source software library for machine learning developed by Google, to build an automated image classification solution that is capable of identifying objects (e.g., vehicles, animals, people, places) and scenes (e.g., landscapes, cities, mountains) in an image or video.
The team then tested this AI tool on several publicly available datasets, with impressive results. They reported that the image classification solution had an accuracy rate of more than 96% on average and was able to detect objects and scenes with an accuracy rate of more than 90%.
Given the ever-increasing amounts of data that businesses collect and store, it’s essential that these collections are searchable. Automated classification offers a quick, easy, and effective means of achieving this goal, with the added bonus of being able to scale your solution as your company grows.
The fourth video on the playlist is “Effective Search for eCommerce Purposes Using TensorFlow and Custom Schemas,” an overview of a research project that teaches computers to perform keyword searches on large amounts of unstructured data stored in repositories like Google Cloud (https://alibaba.github.io/search-nlp-gcloud-tf/).
In this project, the researchers used TensorFlow and custom schemas to build an effective full-text search engine that can index and search through large quantities of unstructured data.
To follow the convention of the previous three videos, let’s take a look at the contents of this video: