Many people consider artificial intelligence to be a threat. They have been using technology every day for a long time. For some people, it’s even an opportunity to experience the world anew. In a conversation with John Giannandrea, Apple’s AI chief, stated about the opportunities offered by AI – and where the company needs to improve.
Perhaps one of the biggest misconceptions about technology is based on an apology. “I’m sorry, Dave, but I can’t do that,” said the HAL 9000 computer in Stanley Kubrick’s 1968 film “2001: A Space Odyssey” when he defied the commands of the people he was supposed to serve. The machine becomes a killer in space. A creepy idea that continues to shape the image of Artificial Intelligence five decades later.
Hardly any other technology triggers such ambivalence. It makes things possible that a few years ago, one did not dare to dream of. They let the car drive independently through the streets and can breathe new life into 100-year-old photos. However, some people think artificial intelligence is going to be the biggest threat to the future.
It has long since penetrated almost all areas of our everyday life. If we take a photo with a smartphone at night, an AI composes a beautiful photo from individual images in a fraction of a second. If we call something to Siri or Alexa, our sentences are broken up into sound fragments in real-time, analyzed, processed – and, with a little luck, understood.
John Giannandrea: “I don’t like the term AI”
Apple’s AI chief John Giannandrea significantly drove this development. The British computer scientist was the technical director (CTO) of the voice recognition company Tellme Networks. He co-founded the web browser division at Netscape and a senior engineer at General Magic. In 2005 he founded the company Metaweb. And Google bought the company in 2010. There Giannandrea headed the departments for search and artificial intelligence, which is the heart of the section. After almost eight years at Google, he moved to Apple in early 2018. Shortly afterward, he was promoted to Senior Vice President for Machine Learning and Strategy for Artificial Intelligence. There he developed a uniform strategy for artificial intelligence and machine learning across all Apple products and services.
When asked about the prejudices against artificial intelligence, the apple’s AI chief replied in an exclusive conversation. “To be honest, I don’t like the term artificial intelligence Super intelligence replaces, including very mundane things that we have known for decades. “Instead, he prefers the term “machine learning.” “Because that’s exactly what it’s about. Teaching computers to do a particular thing, such as recognizing handwriting.”
What Apple does differently than Google
John Giannandrea, who is called JG by everyone in Cupertino, does not choose the example by chance. Apple introduced handwriting recognition to the iPad. It enables the tablet to independently convert handwritten notes made with the Apple Pencil into digital text.
The machine had to learn to recognize the handwriting strokes while someone is writing. To teach that, the company had people all over the world write things – sometimes quickly, sometimes slowly, sometimes weird. A more complex but more promising approach than simply scanning existing manuscripts.
But it is only in combination with the syntax of a language that the iPad can predict which line or which character is likely to follow next. A huge amount of statistical calculations required for this occurs on the tablet and not in a data center. All data remain on the device – that is also a separate approach.
“We are increasingly building functions into Apple products that require machine learning. And I think we do it differently than our competitors,”said apple’s AI chief. “This is partly due to our entangled integration of hardware, software, and services. But also our focus on data protection. I think that’s the one thing that we do fundamentally differently.”
The power of data
The Apple manager does not say it explicitly, but it is clear he wants to differentiate himself from his former employer Google, among others. “Artificial intelligence is one of the most important things, humanity works on. More profound than electricity or fire,” said Sundar Pichai, head of Google parent Alphabet, in early 2018. His group is driving development on many fronts with the Waymo company. He wants to make autonomous driving socially acceptable. And the Deep Mind subsidiary is developing complex language models, among other things.
Google is the leader in many of these areas and criticized because it systematically collects and evaluates user data like no other company. “With the many individuals, closely interlinked services, Google has a huge surveillance machine,” denounced Johannes Caspar, Hamburg’s data protection officer, in a conversation a few years ago. The collection of personal information is also a central component of Facebook and Amazon. Trade data for convenience – that is the unspoken deal on the net.
“This is technically wrong.”
Apple distances itself from this principle whenever possible. Company boss Tim Cook even describes privacy as a “basic human right.” However, an honorable approach ensures that Apple’s services are not as advanced as those of the competition, claim critics. One example of this is Apple’s voice assistant Siri, who ain’t useful as its competitor Amazon Alexa and Google Assistant.
Giannandrea contradicts this emphatically: “There is an assumption out there that machine learning requires a lot of data and that this contradicts Apple’s data protection standards. That is fundamentally wrong. I think we have proven that can built valuable functions.” For example, face unlocking has been developed in modern iPhones without tapping into private user data.
There are many ways to collect data and at the same time protect privacy, continues Giannandrea. “There are advanced techniques like differential privacy and federated learning, where you can learn from the user’s behavior with their permission, without revealing any of their data. Therefore, I believe that the assumption that AI puts data collection in one central place in a data center ahead is technically wrong.”
“We still have a lot to do”
But anyone who uses Apple’s devices and services for longer knows that they don’t always work as expected. A construction site is – at least in this country – the word suggestions. “If you enter text on the iPhone using the keyboard, Siri anticipates the word you will most likely type next,” advertises Apple on its homepage. This sometimes leads to curious results in everyday life. If the author of these lines types “Sincerely, Chri” into his phone, Siri should know after ten years that “Christopher” would be the right suggestion. But what is offered instead? “Christian.”
When asked about this, Giannandrea admitted frankly: “We still have a lot to do.” But he paints a silver lining: “If we have this conversation in a few years, you will be impressed with the advances in keyboard prediction. I hope we will continue to improve machine learning significantly in the coming releases.” In the future, not only the suggestion, hopefully correct name appear there, also restaurants. For example, which you have previously discussed with your partner. “All of that information is already on your iPhone today. We need to make better use of it. Craig’s teams and my teams are working hard to make the experience better.”
The apple’s AI boss rules out that this behavior is a purely phenomenon: “No, no, we are making great efforts to ensure that we use all languages equally. We are investing a lot of money. But we are not yet so far, I totally agree. However, you will see that it will get dramatically better in the next few years.”
Apple’s unusual reluctance
And there is something else that distinguishes Apple from its neighbors in Silicon Valley when it comes to artificial intelligence. While Google explains in detail on the stages of the developer conferences how the technology enables this and that feature. Otherwise not at a loss for superlatives, is comparatively cautious.
Although Apple made voice assistants popular with the masses with Siri in the iPhone 4S in 2011. The group did not enjoy the best reputation in the field of artificial intelligence for a long time. This was mainly due to its own secrecy: For many years, you practically only found out from patent applications what Apple’s AI department was working on. But scientists want to publish and discuss their results. It wasn’t until 2016 that Apple gave in and allowed its employees in this area to publish their results. It was a necessary step to keep up in the battle for the brightest minds in the industry.
But there are also very pragmatic reasons why Tim Cook doesn’t strut across the stage like other CEOs and throw AI buzzwords around, explains Giannandrea. “The end user of an Apple product doesn’t care how a function is implemented, but how well it works. The best technology is the one that disappears that no one needs to understand. Focusing on technology for technology’s sake is nothing that we are interested.”
Learning to experience the world anew
One of the areas that will benefit hugely from machine learning is accessibility features. “Probably the most obvious and descriptive feature is voice control. Which allows you to control the entire iOS and macOS user experience with your voice,” explains apple’s AI chief.
Since last autumn, people with impaired hearing have benefited from noise recognition integrated into iPhones. Artificial intelligence pays attention to predetermined noises. Such as the crying of a baby, the sound of a car horn or a door knock, sends a notification to the phone if the worst comes to the worst. “This is a real game changer for the hearing impaired,” enthuses Apple’s AI boss. “When we think about how machine learning enables computers to understand the world better. It’s not surprising that such features on a smartphone help make the world more understandable for people with disabilities.”
You can also read: These Best Sustainable Kitchen Gadgets Save Time
It can also help in the treatment
Machine learning can also help in the treatment of everyday illnesses. As the digital therapy app Kaia, which has been tested in several clinical studies. She was treated to treat chronic conditions, including back pain, osteoarthritis, and COPD. The app contains instructions for relaxation units and physiotherapy exercises. An AI-assisted assistant uses the camera to analyze the movements and provide feedback – just like a physiotherapist would do.
“We use machine learning to give the user real-time feedback on their exercise form via the smartphone camera,” says Max Strobel, Vice President of Engineering at Kaia. “We started working on it four years ago. But back then, it was just not possible to run such complex models on mobile devices.” Apple now offers a unique interface (Core ML) with which external developers can access the neural processing unit. For Strobel, this is the beginning of a far-reaching change. “It’s pretty exciting what kind of computing power is now available on the latest iPhone and iPad models. The yearly increasing rate is impressive. This allows us to run much more sophisticated models.
Should the technical development continue as in previous years, Strobel expects two trends: “Artificial intelligence and machine learning have the potential to increase the interaction between users and apps. Because the integration of AI in apps essentially increases the ability to make decisions closer to users. And enables applications to react more directly to user input. I also see great potential in the fact that AI can democratize access and reduce costs. The fundamental economic principle that changes machine learning is that suddenly predictions and decision making for specific, well-defined tasks have almost zero costs and are scalable. Now everyone has access to technologies that “
Don’t be afraid of fakes.
The use of artificial intelligence is not only an opportunity. It also offers potential for abuse. Last year, the software company Adobe introduced a new function in its professional software Photoshop. Which you can swap the blue sky in a photo for a romantic sunset with a tap of your finger. This is technically spectacular, but it makes manipulation easier than ever before.
“The concern revolves around what is real and create through machine learning. I still remember four or five years ago when the use of computational photography in the smartphone camera was controversial. Photographers who use medium format or SLR cameras worked, criticizing portraits and white balance corrections using machine learning. People very quickly got used to the fact that their smartphones can take great photos.”
“There is an interesting ethical question about what is too much manipulation. But we have had this problem in our society for many decades. There have already been photoshopped pictures in the press,” continues the apple’s AI chief. “We may need some kind of confirmation that something is real and has not been tampered with.” In his opinion, those robust systems that are able to create forgeries are also the solution to the problem. “There are a lot of artifacts and small details that can be used to tell when images have been machine-generated. If you look at some example videos of politicians that have been artificially generated today, they are pretty obviously bogus.”
“Europe is at the forefront.”
Of course, there are controversial, possibly dangerous developments in the field of artificial intelligence. Companies like Clearview or Palantir demonstrate that the systematic, computer-aided analysis of all available data undermines our privacy. In China, hundreds of millions of cameras are linked together to form a kind of all-seeing eye. Autonomous combat drones could usher in a phase of fully automated warfare.
But apple’s AI chief sees the world as a whole on the right track, especially in Europe. “We are in dialogue with the European Commission and European governments. Europe is at the forefront of defining the rules for the future of machine learning and AI. We are very optimistic that these features will be great for users. But we also think they don’t come without risk. So we’re delighted that policymakers are making sure that we don’t just stumble into the future without thinking about the impact this technology will have on users.
“The apology from HAL 9000 mentioned at the beginning has long been part of pop culture. The phrase is emblazon on T-shirts, mugs, posters. Even Siri knows the saying that was said 43 years before her, well, birth. If you ask Apple’s voice assistant where HAL is, you get the following answer: “Unfortunately, HAL made some bad decisions. But at least he could sing.”
You might also be interested in: