In its brief history as a technology, artificial intelligence has been described as everything from the harmless “computer player” in video games to the technology that will destroy the human race once it achieves sentience – just ask Ray Kurzweil or Elon Musk. For Chris Benson, chief scientist of artificial intelligence and machine learning at Honeywell, and a renowned evangelist for AI, the reality sits somewhere in the middle of those extremes.
“I don’t tend to subscribe to either ‘AI is harmless in all facets,’ and I don’t subscribe to ‘The world is coming to an end’ camp either,” Benson said. “I’m in the middle, saying it’s a tool, and we should be thoughtful in how we’re going to use this tool.”
Benson will provide many of his thoughts on the state of current deep learning technologies and the future, during a fireside chat session at the Robotics & AI Summit on June 18 in Boston. The summit, produced by Robotics Business Review, includes sessions on robotics, industrial automation, and AI applications.
In addition to his work at Honeywell, Benson organizes the Atlanta Deep Learning Meetup and is the host of the new “Practical AI” podcast at Changelog.
A key issue is that many people use AI as an umbrella term, interchangeable with the terms machine learning and deep learning, which causes confusion for many within and outside the industry, Benson said.
“Here in 2018, deep learning is what really people mean when they say AI, but it was a vastly different thing in 1980,” he said. “There’s a very good chance that by 2030, it will be something yet again that’s different. So AI evolves as an umbrella term, whereas deep learning is a precise sub-discipline of machine learning, which is a sub-discipline of data science.”
Simultaneous exponential trends drive AI growth
Benson said three trends are driving the current growth of AI and deep learning: Cloud computing, the creation of modern algorithms, and the availability of big data that is relevant to problems that companies are being solved. In addition, the world is seeing exponential growth in computing capability, thanks to advances in GPUs.
“We’ve never before had a technology emerge that was driven by two simultaneous exponentials that were reinforcing one another,” Benson said. “If anything, we’re speeding up as we explore this.”
As a leading expert on the uses of AI in the world, Benson has seen companies succeed and fail as they dive into the new technology to try and solve business problems. A major key to success is whether companies create a data strategy before they move into a deep learning effort.
“It’s a lot sexier to leap into AI than it is to leap into data strategy by itself, so people do this ‘Cart before the horse’ thing often,” Benson said. “The data strategy that many companies have is insufficient for moving into this new set of tools we’re calling deep learning.”
For example, a lot of companies will take data that’s being generated by Internet of Things devices, or server logs from a data center, and just expect that an algorithm will solve their problem. Benson said companies need to prepare the data, account for input parameters, include dependent variables such as the predictions the neural network will make, and then have a training set ready to go.
“What most companies do is they do either none of that or some of that homework, but they don’t do the full assignment,” Benson said. “Then they get frustrated, and the hype issue becomes a real concern within the business when they have leapt too far, too fast, and they’re not able to get the success they were expecting. In my opinion, that is the single biggest problem in the deep learning field.”
However, Benson said he believes that most companies will learn from their mistakes as they begin to operationalize deep learning and apply those learnings to achieve success.
One application that impressed Benson recently was how medical imaging was using deep learning algorithms to provide a high-resolution look at an unborn baby in the mother’s womb.
Recently featured during the NVIDIA GTC 2018 conference, the technology can be used by doctors to detect health problems earlier, but also to give expectant parents a better view of their child compared with older ultrasound or sonogram images.
“From an imaging standpoint, I already knew they were doing this stuff,” Benson said. “But as I sat there and looked at their demo, it made me think, ‘This affects every human being on the planet who is having a child and getting health care, even if they have no interest in technology. They are being introduced to that unborn child in an incredibly photorealistic way. That has a giant impact on the average person in any country in the world. When you multiply that by millions of use cases, that becomes a really powerful driver of these technologies.”
On a broader scale, the potential for deep learning and AI overall is on the verge of changing the world in almost every industry, Benson said.
“This is going to sound like a complete cop-out, but I see it everywhere,” he said. “I have the opportunity to have many conversations with people from many different industries – I find it hard to imagine an industry that is not looking [at AI]. Every single industry has its toe dipped into the AI world.”
Whether humans are ready for this is another issue, Benson said.
“We’re going into an era that’s going to have more change on an annual basis than had happened across millennia previous to this,” he said. “We’ve had 50,000 years where the human brain has stood out among all the species on the planet since the cognitive revolution happened. If you were to compare the next 1,000 years versus the previous 50,000, you’ll see many orders of magnitude more change.”
The “intelligence” part of the AI term means this is a technology that will continue to involve, which brings additional challenges for humans, Benson said. This is not a technology that’s entirely manual or mundane – it is able to build upon itself.
“This is, by definition, a technology that is getting more and more intelligent over time as we further develop it,” he noted. “There’s a point, potentially, where it gains attributes that we normally associate with human thinking and human feeling.”
If and when that happens, Benson said, humans need to enter into a relationship with AI not unlike relationships with humans – in that AI will be able to influence its own development. “We need to figure out what makes sense – ethically, morally, and be thoughtful as this develops, [because] we may not always see what it develops into until it’s there,” he said.
“Chances are we’ll do some amazing things, and chances are that we’ll make some terrible blunders along the way,” Benson added. “As a society we need to recover from that, and be thoughtful in how we approach [AI development] going forward.”
Chris Benson will speak at 11:30 a.m., on Monday, June 18, at the “Fireside Chat: The State of AI in the Enterprise: How Machine Learning Will Change Everything!” session. Register here to attend the Robotics & AI Summit.