The announcement sounded very promising: With the aid of artificial intelligence, Huawei claimed it had finally written the conclusion to Schubert’s famous Unfinished Symphony. For 197 years, the work remained incomplete, for the composer had not even written three movements before he turned his attention to his ninth symphony. Until his death in 1828, Schubert’s eighth symphony had remained unfinished.
The newly finished work was now awaiting its premiere at London’s venerable Cadogan Hall. Huawei invited Techgarage to visit London in early February to hear the piece performed.
Of course, the suspense was high: would a computer be able to imitate human ingenuity? Would we notice a difference between Schubert’s genius and an AI composer? And would the music of the future be determined by algorithms in lieu of human composers putting pen to paper?
But for a more realistic evaluation, we first need to define some basic parameters.
When is intelligence “artificial”?
If there’s one point the tech community agrees upon, it’s that most devices of the future will have some sort of built-in artificial intelligence, or AI. A device without AI will be as cut off from everything else as today’s mobile devices would be without a cellular or Bluetooth connection.
And if there’s one other point of consensus, it’s that tech companies still have no clue about how to pitch AI to the user, let alone explain the benefits and workings of this highly touted technology to a broad user base.
That lies in the nature of the thing: In this sense, “artificial intelligence” doesn’t even exist yet. All we’ve done so far is to train our computers to do a quick and efficient job of repeating lines of code. Until now, computer chips have only reproduced – or occasionally slightly modified – some aspects of human genius. An impressive proof of this was Google’s DeepMind project, in which the company’s Alpha Go software trained itself in the classic board game Go until it had collected enough data points to hold its own against the best human players – in the process improving its algorithm.
And this is where the boundary starts to blur: When a computer begins to use machine learning to change and improve its algorithms on its own in order to execute a task more efficiently in the future, can we really call that “artificial” intelligence?
Huawei’s vision of artificial intelligence
In recent months, one company that has strongly positioned itself with the concept of AI is Huawei. In line with the slogan “fake it till you make it,” the Chinese electronics manufacturer has not shied away from optimizing its devices and services with a portion of alleged “AI.” Their stated vision is clear: “Huawei believes in the power of AI, bringing technology and human expertise together to push the boundaries of what is humanly possible” – in other words, to redefine the limits of the possible through the fusion of technology and human expertise.
How AI is already improving images
Huawei has started using AI to optimize the pictures users take with their smartphones. That’s why, with the 2017 Mate 10, an NPU (neural processing unit) found its way into a Huawei cell phone for the first time. The chip was supposed to boost the device’s computing power so that it could recognize the objects in an image. The camera settings were then automatically adjusted in whatever way was appropriate: The algorithm would automatically color the sky blue in a landscape shot; it would accentuate the colors in a food photo of sushi – or for portrait photography, it would automatically optimize the image upon detecting a face.
Huawei developed the story further when it converted a car for the purpose of demonstrating it at the 2018 Mobile World Congress. The idea was that it would immediately brake or change course if, through the Mate 10 Pro’s camera, it registered the outlines of a dog, bicycle, or other object on the test track. Following this demonstration, Huawei continued to develop its AI image recognition in subsequent models like the Huawei P20 (Pro) or, more recently, the Huawei Mate 20 Pro, which performed amazingly well in our test.
AI as sign language interpreter?
In the winter of 2018, Huawei’s app StorySign demonstrated yet another use for the company’s alleged AI. With the smartphone’s camera, the app could recognize children’s books’ texts and then reinterpret them into sign language by means of an avatar. This would make it easier for deaf children and their families to learn this essential language. The implication is that Huawei’s proprietary algorithm would take over the bulk of the translation work.
So would future cell phone cameras have the ability to translate any text? One thing is clear: Sign language is a complex system that recognizes major regional differences and adapts to the context of the narrative. That’s why simultaneous translation is not yet possible with the current state of technology. Instead, what Huawei’s software does for now is to recognize a specific block of text in a book and then play back the relevant stored video clip. For the moment, though, every new publication still needs to be interpreted by human translators.
The outcome of this, however, is that a full three months after StorySign’s launch in December 2018, the German catalog still features only one title: Peter Hase: Ein Guckloch-Abenteuer (That’s good old “Peter Rabbit” for english readers). Given that AI purportedly never rests, I personally would have expected a bit more of productivity and preserverance in that field.
Now how about Huawei’s AI composer?
And now, on this foggy February evening in London, a new chapter is about to unveil with the completion of Schubert’s famous Unfinished Symphony. For to AI developers, the gift of improvisation is still the Holy Grail. And this gift is especially needed when composing classical music.
So just how does this cell phone manufacturer plan on bringing this about?
Right from the start of the event in time-honored Cadogan Hall, the expectations were somewhat subdued: The piece we were about to hear was not written 1:1 on a cell phone by a smart chip. Rather, the measures of the first two movements of Schubert’s Symphony No. 8 were played for Huawei’s algorithm until the machine-learning algorithm detected a pattern in terms of timbre, tone frequency, and rhythm, similar to how Huawei trains its Master AI image recognition.
Based on this, the algorithm developed additional related sound combinations and melodies. These were then evaluated by well-known composer and Emmy winner Lucas Cantor, who took the most suitable sound segments and manually put them together to create a composition for performance by a classical orchestra.
Huawei admits that this rendition is neither a “definitive” nor “final” version of the symphony but rather “a unique variation” derived from the interplay of AI technology combined with human expertise.
As a journalist with a tech background, I won’t venture to make a judgment on the quality of the version performed here. However, classical music connoisseurs appeared relatively unimpressed with the result, especially with the fourth movement. Some called the outcome “film music”, describing it as “jarring,” and even an inexperienced listener would be much more likely to recognize the auditory parallels to a James Bond chase scene than a logical extension of the first two movements.
Conclusion: Fake it till you make it
I won’t deny that Huawei is achieving great things with its devices and services. The far eastern company is breathing new life into a saturated market and shaking up the status quo established by previous smartphone manufacturers. That AI is the focal point of their efforts is also absolutely central to their future technical development. For the moment, though, Huawei is still using the concept of “artificial intelligence” in a rather loose sense and with a certain artistic freedom, i.e., whatever machine learning is taking place with the Mate 20 Pro and Co. is not yet happening with any consistency.
There’s no doubt that employing machine learning and artificial intelligence will be an ongoing boon to the user. For the moment, though, Huawei’s algorithm still needs a lot of help from human beings – and at this point in time, this speaks more loudly for the human talent that the Chinese tech company can bring together than for its advances in artificial intelligence.