Let’s rewind back to around 10 years ago; why did we replace all our feature phones with smartphones?
Because they look cool? Or because it allowed new interactions? I believe the main reason is the practical values of the apps. Most of the users are almost forced to follow while people around them have already changed to smartphones. So the greatest contribution of Steve Jobs is not only the breakthrough in phone design, but to open the door to the ecology of future phones. Even today the energy and imagination of such ecology has remained the same.
The same logic applies to mobile AI, especially when AI algorithms are always diversified and even, weird.
Some people use AI in the medical field, some utilize it in customer service, and some may even use it to predict when your life is coming to an end. The main business in this era is to find out how we can adopt all these crazy ideas to our mobile phones.
But it is not easy, either technically or commercially, to integrate the developers and the ecology of development into one’s platform, and to create industrial barriers for mobile AI in this emerging AI world. The development war has already begun amongst technology giants.
We Need Loads of Talents to Develop AI
Looking at all the so-called “AI phones” launched this year, we can see some interesting phenomenon. Taking HUAWEI Mate 10 as an example, after their introduction of photography features such as scenario identification and data tagging last year, all the flagship models from the major brands have followed suit.
No doubt this is a fantastic idea, and indeed integrating scenario identification into photography can solve quite a lot of problems and offer better experience. But why are all the features so similar? While there are so many exciting AI ideas, merely copying from the competitors instead of focusing on innovation is definitely not helpful to the development of AI technology in smartphones.
In fact, there are a few dozens of successful AI solutions on mobile phones, including a few major types of usage such as image recognition, environment interpretation, image enhancement, NLP and voice processing. These solutions help to improve the performance and experiences in mainstream applications such as live broadcast and short videos, photography, social media, shopping, AR and translation. The possibilities in creating new applications are boundaryless and even more tempting. However, the AI applications that each mobile manufacturer can create is very limited; to make mobile AI truly popular and accepted by ordinary people, we will need loads of talents, or even a new business model based on the AI solutions.
Indeed the reality is cruel when one tries to put a technology dream into practice, or in the applications.
For instance, without the support of specialized AI processing power of the phone, many AI tasks would not be able to perform well, or even run for that matter. And without appropriate platforms and open API, developers would not be able to integrate the AI model into mobile phones. If you look at the whole picture, developers would not risk taking part in mobile AI development without clear future benefits and commercial values.
So this has become the major dilemma in mobile AI: while mobile AI could have diversified applications, developers hesitate to invest in mobile AI development considering the lack of technical support and promise in commercial benefits. If this issue remains unsolved, we will probably end up with the awkward situation of having only two to three new AI applications each year.
Nevertheless, dilemma also means opportunities. This is especially true for those who have already mastered the technological advantage and control over the ecology, as this may be the best prospect for them if they can complete something others in the industry are still struggling to do. A few technological giants are already fighting for their land in this war of mobile AI applications development.
HiAI and TFlite: The War between the giants has begun
Currently, for the developers and mobile fanatics, the most well-known mobile AI development platform would certainly be the HiAI mobile AI framework launched by Huawei.
After Huawei and Huawei Honor have launched three flagship products with AI capabilities, the HiAI framework built on the terminal AI processing power is to explore future development capabilities through a platform for developers.
Till now, the HiAI framework has been upgraded to version 2.0, and is compatible with all the mainstream deep learning development framework. The framework also provides development training and two versions of development board.
Through opening up the capabilities of AI performance, algorithm and applications, the entire HiAI framework can offer 5 relatively complete engines and interface. This allows developers to have a platform targeting the realization of AI capabilities, avoiding the technical difficulties in collecting data and learning from scratch, thus saving a significant amount of time and money.
Currently many popular Chinese applications, such as Kuaishou, Douyin, Meitu as well as other shopping and social media apps have already released their plans in collaborating with Huawei and the HiAI framework. For example Kuaishou is going to develop new live broadcasting effect, gesture and body movement recognition, and scenario identification applications, as well as developing a new compressing model that allows AI effects under poor network connection.
Google has launched the TensorFlow Lite late last year. The concept is similar to how Huawei utilizes AI at the terminals, but the actual operation is quite different. While HiAI builds upon Huawei’s AI capabilities and products, TFlite is basically based on the deep learning development framework of TensorFlow, with its aim to help developers to develop and run learning models on their own devices. Thus it is more about algorithm development than application and commercialization.
Some reports state that a number of applications in Pixel 2 were developed based on TFlite, and it is likely that the next Google mobile AI will be integrated with TFlite. At the meantime, many experienced users on TensorFlow have already created lots of exciting application models with TFlite. Although these applications may not be useful for Chinese users, these ideas provide significant reference values for Chinese developers.
This doesn’t mean that Apple does not care about developers; instead, it has released the Core ML, a kind of machine learning ability, on the iOS development platform last June. Currently it provides two APIs, the Vision API and Natural Language Processing API, which allow developers to develop machine vision and natural language processing functionality.
It seems like Apple is leaning towards developing low-level capability in small scale, to allow developers to improve the application experience on iOS, rather than aiming at revolutionary development. From previous experience, Apple always waits until the technology is mature before it releases a groundbreaking development with excellent engineering.
The giants have chosen their directions in this AI development war. As they are fighting for their own ground, they are testing the market with their own technological advantages and strategic demands. While there is no recognized standard in the industry, this also gives developers unlimited opportunities. This war of the future has the target confirmed.
Mobile AI War: What is It All About?
When the demand and understanding of a new technology increase, a few factors determine the occurrence of the ecology of development among different platforms and developers.
First thing first, is to lower the barrier.
This barrier could involve different levels, such as the technological entry barriers, the cost of trialing, learning cost, cost of transfer and compatibility cost. Developers could not afford the time and money to figure out all these uncertainties in the market and would not want to dig a hole for themselves when dealing with platform compatibility and framework transfer. Developers who have little knowledge on algorithm may also want to contribute to this area. The platform should be responsible for providing solutions to all these barriers.
For example, one of the major capabilities of Huawei’s HiAI framework is to allow developers with little algorithm knowledge to develop high-quality AI applications according to their own needs through releasing the application API. This also enables developers to focus on application experience and business development. For applications that are relatively mature, they can improve the AI functionality quickly and avoid high learning cost and long development cycle.
Of course this doesn’t mean that developers do not need to understand algorithm. Keeping in mind that developers may have their own needs and values, they must be provided with options so they can find their own way to flourish.
A functional ecology should consider empowering solutions, such as profit sharing and market guidance. Market values will always be the driving force for all developers, especially those in a highly competitive market such as China. While providing good technical support, the platform should also offer effective guidance on how to gain reasonable traffic and commercial return.
Given the low market recognition, a groundbreaking case may be more persuasive than all the technical parameters and market analysis. Once a distinctively attractive application is born, the commercial values of AI capabilities and AI mobile development framework would emerge to surface, and the appropriate way to enter the mobile AI era would then be confirmed. We all know what would happen in the future following this logic, but in reality we still need a catalyst to lead us to the future.
Mobile AI development is like a castle of the future. We all know that diversity is the key of AI, and we also know that giants with technological advantage have already made their move. The mobile AI development needs just needs a little bit of time.
The HiAI framework seems to be a better option for today’s Chinese developers, while advanced developers with better technical knowledge may be able to borrow from different development solutions and form their own development system.
There is an example which can illustrate the values of AI development ecology for our mobiles: a phone would be useless if it is not compatible with WeChat, regardless of all the claims and brags.
On the same line, the value of AI phone lies in its ability to adopt future AI applications. So the strategic barrier of mobile AI is not to put on the “AI hat” in the marketing campaign, but the technological entry barriers in development and construction of the ecology.
So far this is still a technology game between the giants however, it is possible that everything could change.
