Stephen Hawking first met Gordon Moore, the cofounder of Intel, at a conference in 1997. Moore noticed that Hawking’s computer, which he used to communicate, had an AMD processor and asked him if he preferred instead a “real computer” with an Intel micro-processor. Intel has been providing Hawking with customized PCs and technical support since then, replacing his computer every two years.
Hawking lost his ability to speak in 1985, when, on a trip to CERN in Geneva, he caught pneumonia. In the hospital, he was put on a ventilator. His condition was critical. The doctors asked Hawking’s then-wife, Jane, whether they should turn off the life support. She vehemently refused. Hawking was flown to Addenbrooke’s Hospital, in Cambridge, where the doctors managed to contain the infection. To help him breathe, they also performed a tracheotomy, which involved cutting a hole in his neck and placing a tube into his windpipe. As a result, Hawking irreversibly lost the ability to speak.
For a while, Hawking communicated using a spelling card, patiently indicating letters and forming words with a lift of his eyebrows. Martin King, a physicist who had been working with Hawking on a new communication system, contacted a California-based company called Words Plus, whose computer program Equalizer allowed the user to select words and commands on a computer using a hand clicker. King spoke to the CEO of Words Plus, Walter Woltosz, and asked if the software could help a physics professor in England with ALS. Woltosz had created an earlier version of Equalizer to help his mother-in-law, who also suffered from ALS and had lost her ability to speak and write. “I asked if it was Stephen Hawking, but he couldn’t give me a name without permission,” says Woltosz. “He called me the next day and confirmed it. I said I would donate whatever was needed.”
Equalizer first ran on an Apple II computer linked to a speech synthesizer made by a company called Speech Plus. This system was then adapted by David Mason, the engineer husband of one of Hawking’s nurses, to a portable system that could be mounted on one of the arms of a wheelchair. With this new system, Hawking was able to communicate at a rate of 15 words per minute.
However, the nerve that allowed him to move his thumbs kept degrading. By 2008, Hawking’s hand was too weak to use the clicker. His graduate assistant at the time then devised a switching device called the “cheek switch.” Attached to his glasses, it could detect, via a low infrared beam, when Hawking tensed his cheek muscle. Since then, Hawking has achieved the feat of writing emails, browsing the internet, writing books and speaking using only one muscle. Nevertheless, his ability to communicate continued to decline. By 2011, he managed only about one or two words per minute, so he sent a letter to Moore, saying: “My speech input is very, very slow these days. Is there any way Intel could help?”
Moore asked Justin Rattner, then Intel’s CTO, to look into the problem. Rattner assembled a team of experts on human-computer interaction from Intel Labs, which he brought over to Cambridge for Hawking’s 70th birthday conference, “The State of the Universe,” on January 8, 2012. “I brought a group of specialists with me from Intel Labs,” Rattner told the audience. “We’re going to be looking carefully at applying some state-of-the-art computing technology to improve Stephen’s communicating speed. We hope that this team has a breakthrough and identifies a technique that allows him to communicate at levels he had a few years ago.”
Hawking had been too ill to attend his own birthday party, so he met the Intel experts some weeks later at his office in the department of applied mathematics and theoretical physics at the University of Cambridge. The team of five included Horst Haussecker, the director of the Experience Technology Lab, Lama Nachman, the director of the Anticipatory Computing Lab and project head, and Pete Denman, an interaction designer. “Stephen has always been inspirational to me,” says Denman, who also uses a wheelchair. “After I broke my neck and became paralyzed, my mother gave me a copy of A Brief History of Time, which had just come out. She told me that people in wheelchairs can still do amazing things. Looking back, I realize how prophetic that was.”
After the Intel team introduced themselves, Haussecker took the lead, explaining why they were there and what their plans were. Haussecker continued speaking for 20 minutes, when, suddenly, Hawking spoke.
“He welcomed us and expressed how happy he was that we were there,” says Denman. “Unbeknown to us, he had been typing all that time. It took him 20 minutes to write a salutation of about 30 words. It stopped us all in our tracks. It was poignant. We now realized that this was going to be a much bigger problem than we thought.”
At the time, Hawking’s computer interface was a program called EZ Keys, an upgrade from the previous softwares and also designed by Words Plus. It provided him with a keyboard on the screen and a basic word-prediction algorithm. A cursor automatically scanned across the keyboard by row or by column and he could select a character by moving his cheek to stop the cursor. EZ Keys also allowed Hawking to control the mouse in Windows and operate other applications in his computer. He surfed the web with Firefox and wrote his lectures using Notepad. He also had a webcam that he used with Skype.
The Intel team envisaged an upheaval of Hawking’s archaic system, which would involve introducing new hardware. “Justin was thinking that we could use technology such as facial-gesture recognition, gaze tracking and brain-computer interfaces,” says Nachman. “Initially we fed him a lot of these wild ideas and tried a lot of off-the-shelf technologies.” Those attempts, more often than not, failed. Gaze tracking couldn’t lock on to Hawking’s gaze, because of the drooping of his eyelids. Before the Intel project, Hawking had tested EEG caps that could read his brainwaves and potentially transmit commands to his computer. Somehow, they couldn’t get a strong enough brain signal. “We would flash letters on the screen and it would try to select the right letter just by registering the brain’s response,” says Wood. “It worked fine with me, then Stephen tried it and it didn’t work well. They weren’t able to get a strong enough signal-to-noise.”
“The more we observed him and listened to his concerns, the more it dawned on us that what he was really asking, in addition to improving how fast he could communicate, was for new features that would let him interact better with his computer,” says Nachman. After returning to Intel Labs and after months of research, Denman prepared a 10-minute video to send to Hawking, delineating which new user-interface prototypes they wanted to implement and soliciting his feedback. “We came up with changes we felt would not drastically change how he used his system, but would still have a large impact,” says Denman. The changes included additions such as a “back button,” which Hawking could use not only to delete characters but to navigate a step back in his user interface; a predictive-word algorithm; and next-word navigation, which would let him choose words one after another rather than typing them.
The main change, in Denman’s view, was a prototype that tackled the biggest problem that Hawking had with his user interface: missed key-hits. “Stephen would often hit the wrong key by hitting the letter adjacent to the one he wanted,” says Denman. “He would miss the letter, go back, miss the letter again, go back. It was unbearably slow and he would get frustrated.” That particular problem was compounded by Hawking’s perfectionism. “It’s really important for him to have his thoughts articulated in exactly the right way and for the punctuation to be absolutely right,” says Nachman. “He learned to be patient enough to still be able to be a perfectionist. He’s not somebody who just wants to get the gist of the message across. He’s somebody who really wants it to be perfect.”
To address the missed key-hits, the Intel team added a prototype that would interpret Hawking’s intentions, rather than his actual input, using an algorithm similar to that used in word processing and mobile phones. “This is a tough interaction to put your faith into,” the video explained. “When the iPhone first entered the market, people complained about predictive text but quickly distrust turned to delight. The problem is that it takes a little time to get used to and you have to release control to let the system do the work. The addition of this feature could increase your speed and let you concentrate on content.”
The video concluded: “What’s your level of excitement or apprehension?” In June that year, Hawking visited Intel Labs, where Denman and his team introduced him to the new system, initially called ASTER (for ASsistive Text EditoR). “Your current piece of software is a little dated,” Denman told him. “Well, it’s very dated, but you’re very used to using it, so we’ve changed the method by which your next-word prediction works and it can pretty much pick up the correct word every single time, even if you’re letters away from it.”
“This is a big improvement over the previous version,” Hawking replied. “I really like it.”
They implemented the new user interface on Hawking’s computer. Denman thought they were on the right path. By September, they began to get feedback: Hawking wasn’t adapting to the new system. It was too complicated. Prototypes such as the back button, and the one addressing “missed key-hits,” proved confusing and had to be scrapped. “He’s one of the brightest guys in the world but we can’t forget that he hasn’t been exposed to modern technology,” says Denman. “He never had the opportunity to use an iPhone. We were trying to teach the world’s most famous and smartest 72-year-old grandfather to learn this new way of interacting with technology.”
Denman and the rest of the team realized that they had to start thinking differently about the problem. “We thought we were designing software in the traditional sense, where you throw out a huge net and try to catch as many fish as you can,” says Denman. “We didn’t realize how much the design would hinge on Stephen. We had to point a laser to study one individual.”
At the end of 2012, the Intel team set up a system that recorded how Hawking interacted with his computer. They recorded tens of hours of video that encompassed a range of different situations: Stephen typing, Stephen typing when tired, Stephen using the mouse, Stephen trying to get a window at just the right size. “I watched the footage over and over,” says Denman.
“Sometimes, I would run it at four times the speed and still find something new.”
By September 2013, now with the assistance of Jonathan Wood, Hawking’s graduate assistant, they implemented another iteration of the user interface in Hawking’s computer. “I thought we had it, I thought we were done,” says Denman. However, by the following month, it became clear that, again, Hawking was having trouble adapting. “One of his assistants called it ‘ASTER’ torture,” recalls Denman. “When they said it, Stephen would grin.”
It was many more months before the Intel team came up with a version that pleased Hawking. For instance, Hawking now uses an adaptive word predictor from London startup SwiftKey which allows him to select a word after typing a letter, whereas Hawking’s previous system required him to navigate to the bottom of his user interface and select a word from a list. “His word-prediction system was very old,” says Nachman. “The new system is much faster and efficient, but we had to train Stephen to use it. In the beginning he was complaining about it, and only later I realized why: He already knew which words his previous systems would predict. He was used to predicting his own word predictor.” Intel worked with SwiftKey, incorporating many of Hawking’s documents into the system, so that, in some cases, he no longer needs to type a character before the predictor guesses the word based on context. “The phrase ‘the black hole’ doesn’t require any typing,” says Nachman. “Selecting ‘the’ automatically predicts ‘black’. Selecting ‘black’ automatically predicts ‘hole’.”
The new version of Hawking’s user interface (now called ACAT, after Assistive Contextually Aware Toolkit) includes contextual menus that provide Hawking with various shortcuts to speak, search or email; and a new lecture manager, which gives him control over the timing of his delivery during talks. It also has a mute button, a curious feature that allows Hawking to turn off his speech synthesizer. “Because he operates his switch with his cheek, if he’s eating or traveling, he creates random output,” says Wood. “But there are times when he does like to come up with random speech. He does it all the time and sometimes it’s totally inappropriate. I remember once he randomly typed ‘x x x x’, which, via his speech synthesizer, sounded like ‘sex sex sex sex’.”
Wood’s office is next to Hawking’s. It’s more of a workshop than a study. One wall is heaped with electronic hardware and experimental prototypes. Mounted on the desk is a camera, part of an ongoing project with Intel. “The idea is to have a camera pointed at Stephen’s face to pick up not just his cheek movements but other facial movements,” says Wood. “He could move his jaw sideways, up and down, and drive a mouse and even potentially drive his wheelchair. These are cool ideas but they won’t be coming to completion any time soon.”
Another experimental project, suggested by the manufacturers of Hawking’s wheelchair earlier this year, is a joystick that attaches to Hawking’s chin and allows him to navigate his wheelchair independently. “It’s something that Stephen is very keen on,” says Wood. “The issue was the contact between Stephen’s chin and the joystick. Because he doesn’t have neck movement it is difficult to engage and disengage the joystick.” Wood shows WIRED a video of a recent test trial of this system. In it, you can see Hawking driving his wheelchair across an empty room, in fits and starts. “As you can see, he managed to drive it,” says Wood. “Well, sort of.”
Wood showed WIRED a little grey box, which contained the only copy of Hawking’s speech synthesizer. It’s a CallText 5010, a model given to Hawking in 1988 when he visited the company that manufactured it, Speech Plus. The card inside the synthesizer contains a processor that turns text into speech, a device that was also used for automated telephone answering systems in the 1980s.
“I’m trying to make a software version of Stephen’s voice so that we don’t have to rely on these old hardware cards,” says Wood. To do that, he had to track down the original Speech Plus team. In 1990, Speech Plus was sold to Centigram Communications. Centigram was acquired by Lernout and Hauspie Speech Products, which was acquired by ScanSoft in 2001. ScanSoft was bought by Nuance Communications, a multinational with 35 offices and 1,200 employees. Wood contacted it. “They had software with Stephen’s voice from 1986,” says Wood. “It looks like we may have found it on a backup tape at Nuance.”
Hawking is very attached to his voice: in 1988, when Speech Plus gave him the new synthesizer, the voice was different so he asked them to replace it with the original. His voice had been created in the early ’80s by MIT engineer Dennis Klatt, a pioneer of text-to-speech algorithms. He invented the DECtalk, one of the first devices to translate text into speech. He initially made three voices, from recordings of his wife, daughter and himself. The female’s voice was called “Beautiful Betty”, the child’s “Kit the Kid”, and the male voice, based on his own, “Perfect Paul.” “Perfect Paul” is Hawking’s voice.
This story was first published in WIRED UK issue 01.15
No comments:
Post a Comment