From Unmanned Systems: Jeopardy champ Watson branches out into business
Many people think of artificial intelligence as something all-consuming, a superior computer thinking ability that could make humans obsolete.
Decades of books, movies and TV shows have warned seems like an inevitable computer or robotic uprising, including the 1920 stage play, “Rossum’s Robots,” which gave us the term robot in the first place.
And tech industry leaders such as Mark Zuckerberg on Musk have been arguing lately about AI and whether it’s a force for good, as Zuckerberg posits, or an existential threat to humanity, as Musk says.
Perhaps the most famous AI in existence right now Watson, IBM’s deep-learning device that gained worldwide renown in 2011 when it beat two human champions on the quiz show Jeopardy!, using its internal memory and without being connected to the internet.
Since then, IBM has been using Watson in a number of surprising ways, such as by helping doctors keep abreast of a tidal wave of information in peer-reviewed reports. It has also been pitching Watson as a way for businesses to incorporate AI into their operations.
That’s the job of Marc Teerlink, the chief business strategist for IBM Watson Group, which urges businesses to “do your best work with Watson.”
That’s possible, Teerlink says, because AI is much simpler than most people think. It’s not a giant brain waiting to take over, it’s more like a screwdriver or a hammer: A specific tool for a specific job. And it’s a tool that everybody can use, he says.
“Five years from now, any important decision is going to be supported by artificial intelligence,” he said earlier this year at a NASA-sponsored talk at Awesome Con, Washington, D.C.’s answer to California’s Comic Con.
North Star blueScope Steel is applying IBM Watson internet of things technology and wearable devices to pioneer novel approaches to help protect workers in extreme environments. Photo: North Star BlueScope Steel
In a later interview with Unmanned Systems, he notes that AI, in its current form, is best used to take over the “heavy lifting” that people and companies can find hard to do. For instance, Watson can sort through hundreds of factors to help match cancer patients with clinical trials that could help them. “All I am saying, is let AI do the heavy lifting, let the AI do the routine processing and have the humans do the thinking,” he says.
Today, about 20 percent of cancer patients in the United States are eligible for clinical trials, yet only about 5 percent of them end up in one, he says, due to the complexity of calculating a variety of factors such as genetic match or simpler factors such as where they live and whether they have someone to drive them. Watson can take on this task, which could take a human 160 to 180 hours, and calculate it much more rapidly.
So in that instance, “we have taken one step out of the process and we have augmented the doctor,” Teerlink says.
That’s the sort of situation where AI can shine. People don’t expect other people to be expert in everything — you don’t talk to your doctor about your stock options — “but we expect AI to be so well trained that it can handle everything in every situation,” he says. Yet mostly AI today is “Artificial Narrow Intelligence,” or AI that specializes in one area. Don’t underestimate this, as it takes away a lot of heavy lifting from you and me in our daily life. And a lot of narrow AI has already surpassed human beings in proficiency in several areas.
Teerlink compares it to a taxi app on your phone that orders the cab for you, and tells you how much longer it will be until a cab arrives, so you don’t have to wait out in the rain. “That’s all it takes to do something special,” he says. “Take a process that you know well, and replace one step with AI.”
Teerlink notes that even the version of Watson that dominated Jeopardy only did so because it had been trained to play Jeopardy. “If you had taken that specific instance of Watson and used it for medical treatment, it would have quoted treatments from Shakespeare and the Bible,” he says, which might not lead to optimal treatment.
Watson case studies
IBM’s Watson is a cloud-based platform for intelligent assistance, consisting of many AIs, each trained for their narrow, specific task. Yet if orchestrated, you could solve problems that are not yet solvable.
With AI, “the basis is that we would be so overwhelmed with information that it would be impossible for us to internalize it, and use it to what its full value could be,” he says.
IBM has several examples of how orchestrating these narrow AI’s together can work, becoming deep learning, or a more generic AI that can tackle a problem, as Watson has been used for everything from analyzing elevator data to keep riders safe, to helping draft professional basketball players, to boosting the “graduation” rates of helper dogs for the blind.
Watson is even a musician of sorts. The AI platform was used by Grammy Award-winning music producer Alex Da Kid to evaluate millions of songs to help write a better one. Watson analyzed years’ worth of text from the New York Times, the Getty Museum, even Supreme Court rulings, and then read the lyrics
of more than 26,000 songs, “finding the most pervasive themes and uncovering the way people talked about them,” according to IBM’s Jeremy Hodge, speaking in a company video.
It also analyzed those songs to find patterns of keys, chord progressions and other data points. Da Kid also used a new IBM tool, Watson Beat, a cloud-based app that’s still under development.
“Watson Beat knows what sounds good to us. And based on input, it knows, for example that if a user wants a ‘dark’ or ‘gloomy’ song, that it should use a minor key,” says Janani Mukundan, one of its developers, in an IBM blog post.
Users can input the beginnings of a song, and “in as few as eight bars or 15 seconds of input, it can churn out a multi-instrument, minutes-long track that can be adjusted according to the mood and genre any budding Beethoven chooses,” the blog entry says. The users can then change the song to their liking.
Da Kid then used Watson’s insights and technology to write the aptly named song “Not Easy,” and recorded it with Elle King and Wiz Khalifa. The result is available on iTunes.
“It’s almost like having a million [copies] of yourself reading a million different books at once, a million articles, and understanding social media and just conversation in general. I could never do that,” Da Kid says in the video. “… It just took everything to another level.”
Watson has also been working with E & J Gallo Winery to develop an intelligent irrigation system that increases the quality of the winemaker’s grapes. The system mapped a 30-by-30-meter grid and couples that with Watson’s ability to read an enormous amount of satellite weather data.
“Each block of vines in the grid gets its own personal irrigation plan based on weather data and soil moisture levels,” the Watson website explains. “This allows the exact needed amount of water— based on highly targeted irrigation requirements — to be dispensed to each grapevine. As the weather changes, the irrigation methods react to ensure vines only receive water when needed.”
Working with Watson
Before trying to incorporate AI into their workflow, Teerlink says companies need to ask themselves some basic questions: What are the “heavy lift” items that I need help with? If I had that help, what added value would that bring? Do I want to have a conversation with customers, or do I want to compare multiple documents? Do I want to share my data as I “train” my AI system, or do I want to keep it to myself to maintain my competitive edge of “my secret sauce?”
IBM offers a variety of products related to Watson, starting at lower levels of AI, where “each [application programming interface] does one thing and does it really well,” Teerlink says.
Then there are starter kits and accelerator packages, where other companies have input their knowledge to help train the AI to perform specific functions. Beyond that are apps that have even more expert training and which offer more AI functions.
Thousands of companies work with the system, including some household names such as H&R Block, which uses it to help identify credits and deductions; Sesame Workshop, which uses it to develop educational platforms for preschoolers around the world; and The North Face, which developed a dialogue-based chat system to help users shop.
Other users include the Weather Company, which uses AI to improve weather forecasts; North Star BlueScope Steel, which monitors the status of hits workers using sensors that monitor their heart rates, body temperature and other data points to develop personalized safety guidelines; and Macy’s, which allows customers to input questions in natural language about Macy’s store products, so they can receive a customized response and go to the right location.
There is a “healthy mix” of IBM reaching out to potential clients, and companies approaching Watson seeking help, Teerlink says.
Predicting the future of AI today is tricky; compare it to the few that could have predicted all the things made possible by the internet, or by smart phones.
Teerlink notes that IBM is a member of the Partnership on AI to Benefit People and Society, a group formed by industry heavy-hitters to develop and share best practices, boost public understanding of the technology and provide an “open and inclusive” platform for discussing it. Members include Amazon, Apple, Facebook, Google and Microsoft.
“We call them principles for AI, for a cognitive era, to guide what you do,” Teerlink says. “It’s frankly our responsibility as leaders out there that are putting these technologies out to guide them in their entry into the world in a safe way.”
He also says that an ethical framework needs to be in place before there can be widespread adoption of AI systems. Someone needs to help answer the question of how AI systems should behave in certain situations, such as the pending crash of a self-driving car. Should the car hit an old pedestrian, a young pedestrian, or avoid them both and kill its passenger?
“If we want the AI in the car to make a decision, we need some kind of government-blessed framework that helps to make the decision,” he says. “What do you want the technology to do? … How do we deal with that, or how do we deal with the liability?”
That holds true in other areas, such as using AI to suggest medical treatments or give legal advice. “There will be regulations and rules” around responsible AI, Teerlink says. “We’re still at the beginning of that robust dialogue.”
“It’s going to take some time before we have fully automated AI, because of the liability and the [legal] responsibility of the decision making,” he says. “We fix that, governance wise and ethical wise, we go on to the next step. But Before that, I think AI is mostly going to be augmenting our jobs, in a pretty significant way, towards a near future, where the AI does our heavy lifting and offer the humans more time to do the [creative] thinking.”