Machines will soon be able to learn without being programmed
Teaching machines to parse through large volumes of data to learn new concepts and rules is a critical area of development in artificial intelligence, experts told CNBC.
That concept is called machine learning, and it’s been a longtime goal for the AI discipline: The term was coined in 1959 by AI pioneer Arthur Samuel who defined it as a computer’s ability to learn without being explicitly programmed.
To do that, mathematical models are built and then fed with huge volumes of data, experts said. The algorithms learn to identify patterns and assumptions from those data sets that are then applied to process new information.
“We want to be able to use the machine’s own capability to learn from complex data,” Eric Chang, senior director at Microsoft Research Asia, told CNBC.
One area of machine learning that is being looked at is image recognition. Traditionally, a program would need to be specifically told to look at a facial feature — like a nose — for each photo that comes its way. With machine learning, the program learns from millions of examples what the broad category of a feature — “noses” — looks like, so it can identify new ones in future photos.
Replicate that process for hundreds of features, and you get a powerful tool. Businesses will soon be able to put it to use, experts said.
In an airport lounge, for example, machine learning technology can be used to recognize the faces of every passenger that walks in, according to Ian Massingham, global head of technical and developer evangelism at Amazon Web Services. He explained that it would allow the service staff to pull existing information on each passenger and know their preferences in advance.
“These kinds of services end up playing a role in the decision support or in service support,” Massingham told CNBC. He added that such a facility allows the service staff to concentrate on what they do best — personalized interaction.
“A machine can’t go up to you and warmly say hello, nice to meet you and smile. So, human beings will have that stuff, but they’ll be better informed because they will be using AI and machine learning as part of the customer services workflow.”
Speech recognition and natural language processing are also key areas of machine learning that are being researched today. A common example of how speech recognition works is when smartphone users talk to virtual assistants like Siri or Google Assistant on their devices. The phones pick up and process the audio into text using a series of complex algorithms, then they apply natural language processing to understand what the user meant.
Google researchers, for example, have found a way to develop an algorithm that can pick out a single speaker’s voice in a crowded space, according to a report. Microsoft researchers, on the other hand, are studying multilingual speech to train virtual assistants to process interactions where a user switches back-and-forth between two languages.
A ‘phenomenal pace’
Machine learning is possible today due to a few reasons, according to Microsoft’s Chang. First, because of the sheer volume of data that is being generated everyday; secondly more computation power is available due to cloud computing. Finally, he said, newer, more complex algorithms are being created.
Developments in machine learning will progress at a “phenomenal pace this year,” according to professional services company Deloitte’s Technology, Media and Telecommunications Predictions 2018 report.But, the firm noted, the pace of advancement in machine learning will be so rapid that in 50 years, today’s developments would be considered “baby steps.”
Large-and-medium sized businesses are going to use more machine learning technologies this year, Deloitte predicted. It added that the number of projects using the technology will double compared with 2017. At the same time, high-end smartphones will have machine learning chips installed in them but “those chips will not yet be fully utilized,” the company noted. It added that new chips could also “dramatically increase the use of” machine learning by making applications use less power and still be as responsive and flexible.
More broadly, AI will remain a key spending area for companies in the near future, according to the International Data Corporation. In March, IDC predicted that worldwide spending on cognitive and artificial intelligence systems will see a 54.2 percent on-year jump in 2018 to $19.1 billion. That number could grow to $52.2 billion in 2021, IDC said.
Areas that would see the implementation of AI include health care, financial services, transportation and manufacturing, according to Chang.
There’s still some skepticism surrounding the rapid pace of developments in AI.
First, the longstanding concern that machines would ultimately replace humans in the workforce remains. Last year, a report from McKinsey & Company predicted that as many as 800 million workers globally could be replaced at work by robots by 2030. But, the counter-argument many in the industry put forward is that AI will create newer types of jobs.
More than 80 percent of the jobs in 2030 have yet to be invented, according to Amit Midha, president of Dell EMC’s commercial business in Asia Pacific and Japan. “And that is exciting as well as challenging,” he told CNBC. “Exciting in the sense that these jobs are likely to be higher quality jobs where drudgery don’t exist … at the same time, we have to train people for those types of jobs.”
On the other hand, rapid development in AI could lead some investors to expect quick results in areas that require a lot of long-term investments, according to Microsoft’s Chang. “I always tell people that this is a marathon, not a sprint,” he said, adding that another challenge for AI researchers is making the complexities of machine learning more transparent to a wider group of people.
Despite the progress made thus far, some experts think that AI still has a long way to go before achieving what is known as artificial general intelligence, That’s when machines are able to think and act in such a way that they can be mistaken for humans.
Humans can do a variety of things unprompted, according to Tomaso Poggio, director at the Center for Brains, Minds and Machines at MIT. That includes talking about science, technology, the weather, sports, describing objects, sharing about memories, life experiences, motivations and more, he told CNBC in an interview earlier this year.
“We have no idea how to make a machine that can do that. I think we’ll get there, but I don’t think, despite the recent progress, I don’t think that we know or we have any hint of how to do this right now. We’ll get there, but it’s not around the corner yet,” he said.
Dell EMC’s Midha said progress will be made gradually and the focus should remain on using AI to solve societal problems.
“We create (the) future,” he said. “We have to lead towards that. If we make more data available, if we have more compute resources available, and if we keep the focus on solving the problems, which are difficult to solve, we will absolutely make progress.”