Silicon Valley Insights– Artificial Intelligence [AI]
Case Study: Amazon Alexa’s AI Platform
Hackerbay delivers an exclusive SVI on why developments in NLU and ASR innovations are creating huge market opportunities for enterprises and developers alike…
In our tech-enabled world, buzzwords such as ‘artificial intelligence’, ‘deep learning’ and ‘machine learning’ are thrown around not only within tech circles, but also in general cultural discussion. Artificial intelligence, or AI, is a hot conversation topic as innovation is rapidly developing in both b2b settings and the general consumer marketplace. This SVI report goes deeper to reveal exactly where this buzz is coming from, and more specifically, where the market opportunity of AI technology lies not only for Silicon Valley companies, but for enterprises of all sizes.
SVI Snapshot
1: ASR & NLU
Recently, two main advancements under the AI umbrella have been achieved; in automatic speech recognition (ASR) and natural language understanding (NLU).
What do these breakthroughs look like?
Regarding ASR, a KPI is accuracy. A certain amount of audio input is translated into written text (which is then run through NLU), and the accuracy with which the computer identifies the language of the audio is monitored. ASR accuracy has jumped from approximately 30% to 90%*+. This is due to AI and deep learning algorithm development.
Regarding NLU, an established KPI is word-error rate (definition: out of a certain amount of words that are written, the word-error rate is the amount of words that a computer misunderstands). Word-error rate went from 18% average to 6% average in the last 36 months due to a deep learning algorithm methodology.
How are these breakthroughs occurring?
There are hundreds of PHDs from esteemed institutions such as Stanford, MIT etc, who are working in the research departments of the largest companies in the tech industry. These experts are the best in their field, and through focusing all of their efforts and research on this innovation space, they become the primary driving force behind AI transformation in consumer-end products and user experience. Their progress is propelling technological innovation forward incrementally via the leveraging of the 3 Hacker Laws.
2: Amazon’s AI revolution:
What did Amazon do? They took the research insights on ASR and NLU, and identified the market opportunity. They built a product based on the innovation to build a consumer product (leveraging Moore’s Law) to bring this innovation to a mass market of 11 million users via consumer hardware. They did not stop there, however. Due to Metcalfe’s Law (network effects), they commoditized and democratised the deep learning break-throughs that Alexa pioneered in order to build a platform for 3rd-party developers to operate on. This platform is known as ‘Lex’.
3: How Hackerbay fits in:
As a result of Amazon’s 3rd-party Lex platform innovation, Hackerbay has the opportunity to build unique Alexa ‘skills’ on an on-demand basis. HB applies Lex innovation to leverage the market opportunity in a new and innovative way. We are among the first in the industry to identify the niche demand for these skills.
Why is this useful?
It makes no sense for corporations to attempt hiring experts in Alexa because the technological landscape is rapidly evolving. It makes morse send to collaborate with free-floating experts on an on-demand basis who can adapt to specific projects and tasks.
Hackerbay delivers AI technology to corporations on an on-demand, per-project basis via 3rd-party developer innovation platforms such as Amazon Lex.
SVI: AI Full Report
Over the past years, there have been monumental leaps in performance and ability of computing power. The advent and progression of AI technologies specifically have excited research and computer science experts, but when consumer tech giant Amazon started to develop Alexa, a fully integrated AI system focused on great customer experience, the buzz increased on a mass-market scale.
According to statistics generated by investment banking analysts at Morgan Stanley, Amazon has now sold more than 11 million Alexa powered Echo Devices.
Because of an increase in customer numbers, Amazon’s Alexa seized the opportunity window to become the “iPhone” of voice and artificial intelligence innovation. Amazon Alexa is not only a hardware product, but a platform with 7000+ ‘skills’.
April 2017: Amazon launches an AI developer platform to the general public
As a Silicon Valley giant which successfully understands and leverages the 3 laws of Hacker Culture (that of Moore, Metcalfe and the Power Law), Amazon did not stop at creating an innovative AI product (a direct beneficiary of Moore’s Law). They saw the exponential growth potential of integrating the product into a platform in a manner not dissimilar to Steve Jobs, whose decision to turn the Apple iPhone into a powerful platform from a hardware product revolutionized the landscape of consumer technology.
Amazon has heavily invested research, resources and funding into a developer platform they have called “Lex”. At their developer conference (Re:Invent) in December 2016, they debuted the private version. Take a look at the presentation here:
After a rapid four-month turnaround, Amazon Web Services made Lex available to the general public on the 19th April 2017
Lex has several benefits for third-party developers to kick-off a two-pronged platform network effect.
Firstly, it’s easy to use. Secondly, due to the technological infrastructure of AWS (Amazon Web Services), all the software written on Lex is seamlessly deployed and scaled on AWS servers. As a consequence, developers can quickly scale their Alexa apps to millions of users without additional server hosting problems or scale issues, which was one of the major problems with Apple’s iPhone platform in 2007. The key innovation in Amazon Lex is the high-quality integration of text, as well as speech understanding capabilities. For third-party developers, this provides on-demand access to the same technology available that powers Amazon Alexa itself.
Furthermore, the complete infrastructure of Amazon Lex (which enables developers to use Alexa and AI technology) is cost-efficient. 1000 requests come out at $0.0004, so 1 million active users costs approximately $2k for the server hosting and maintenance fee. This is a major breakthrough in cost-efficiency of AI technology, especially compared with five years ago when developers had to build their own data-centers, cover maintenance and scale opportunities and ultimately, build their own hardware.
So why have AI developer platforms like Amazon Lex begun trending in Silicon Valley?
According to Amazon’s research paper, Lex is on a mission to democratize deep learning technologies, making them easily available to independent developers.
Lex focuses on two deep learning technologies: Natural Language Understanding (NLU) and Automatic Speech Recognition (ASR).
Amazon is also working on a Text to Speech (TTS) software named Polly, which is currently proprietary with the view to integrate with Lex over the coming months. With regards to performance improvements in NLU, it is possible for computers to analyze text to extract meta-data from content such as concepts, entities, keywords, categories, sentiment, emotion, relations and semantic roles. Major tech companies such as Microsoft, Google, Amazon, IBM and elite universities such as Stanford, MIT and Harvard are investing heavily into technological improvements and research in this field.
Clearly, deep learning has revolutionized the world of artificial intelligence and continues to do so. But how much does it improve performance? Specifically how has computing performance improved at different tasks since the advent and development of deep learning? It seems that deep learning doesn’t seem to excel in the same way as NLU, where trends are positive, tangible and exponentially improving. The Association for Computational Linguistics has provided performance statistics for various NLU tasks:
In January 2017, renowned data scientists from Snip, Caroline Wisniewski, Clément Delpuech, David Leroy, François Pivan and Joseph Dureau, published research insights comparing the NLU capabilities of Amazon Alexa with Snips, API, Siri and Luis. These insights demonstrate Alexa’s NLU accuracy is constantly improving.
So what about performance improvements in Automatic Speech Recognition (ASR)? According to the research by Sarah Constantin of Google, it’s hard to compile definitive results because large corporations who implement the technology do most of their work on vast, proprietary data-sets confidentially. The shared data benchmarks are tiny by comparison and don’t provide detailed recognition for real tasks, however her blog from January 2017 does demonstrate where the identifiable developments in ASR are occurring:
Dictation software company, Nuance, shows a steadily improving rate of performance on word recognition up to the present day, with a plausibly exponential trend.
These are major breakthroughs in ASR, and although Amazon Alexa has not shared performance data, it is of course leveraging general industry trends. What makes Amazon Alexa’s platform so appealing is the neatly-packaged and convenient full-stack combination of a beautiful end consumer device, which is used by millions of users, and a fully-integrated solution with a developer platform, which takes care of all the heavy-lifting.
Leveraging Amazon’s AI platform in the Automotive Industry.
Due to market giants such as Tesla provoking fierce competition through leaps in innovation, the automotive industry as a whole was very quick to include practical-use cases of this newly evolving AI platform, employing developers to run tests on the products available. Innovation managers in the industry see the opportunity to leverage the general trend in AI by incorporating technology like Amazon Alexa, with the view to build great user experiences for end-consumers.
For example, in January 2017, Ford and Amazon revealed their collaboration, where consumers can use Alexa’s cloud-based voice technology to access their car from home, amongst other features. This is a first for the automotive industry. The in-vehicle function via SYNC® 3 AppLink™ enables users to listen to audiobooks, search and transfer destinations to navigation, request news, music, add items to Amazon shopping lists and more. In the car, Alexa allows control of home lighting, security systems, garage doors and more:
In April 2017, Mercedes Benz followed by revealing their fully-integrated in-vehicle Alexa solution:
Leveraging platforms has provoked a decrease in cost for AI integration and intelligence in the automotive industry, opening a plethora of opportunity windows amongst more traditional companies who are looking to branch out into digital innovation spaces.
Hackerbay is moving further into the automotive industry as these technologies improve and proliferate over the course of 2017 and beyond.
So how about competition? Who is set to challenge Amazon in delivering AI technologies to a mass consumer market? Microsoft Cortana and Google are on the come up…
The technological breakthroughs in NLU and ASR have caught the attention of other tech giants like Microsoft and Google, perhaps a little later than Amazon. Google already has a direct competitor with Google Home. However, Hackerbay is keeping close tabs on Microsoft Cortana, their research in AI and their proprietary technology platform, Luis.ai.
Currently, Amazon retains a strong market position with their fully-integrated solution, but Cortana is catching up. Cortana did not release the cooperation with Harman/Kardon, but according to coverage in The Verge from December 2016, Microsoft is expecting a release of an Amazon Alexa competitor, which comes as no surprise as developments in AI and subsidiary features continue to snowball.
In conclusion, we have established that Artificial Intelligence is a trend to continue watching over the course of 2017 and most likely, beyond. As shown in this exclusive Hackerbay SVI Report, AI has opened up a channel of great market opportunity for third-party developers with the opportunity to leverage AI platforms– a ‘luxury’ that was previously unfeasible.
Developers from all industries can design and realize end-consumer experiences at a low price, built on top of a technological infrastructure that employs the newest scientific research in the field of AI and deep learning.
When applying these new advantages across a number of industries, developers have an abundance of new opportunities via which innovative and ground-breaking technologies will undoubtedly be discovered. Be certain to watch this space.
