Creating some artificial intelligence programs is easy, but turning them into successful businesses is a challenge, experts said this week in Singapore at the annual Innovfest Unbound tech conference.
Improvements in computing power and better availability of data have sped up developments in AI, with large corporations, start-ups, universities and governments all getting involved. Earlier this year, the International Data Corporation predicted that worldwide spending on cognitive and AI systems will grow to $52.2 billion in 2021, up from about $19.1 billion this year.
Since a lot of the basic computer codes used to create AI are widely available, it's easy for developers to come up with programs that can perform certain tasks: For example, a chat bot can take over the basic tasks done by a customer service representative or a program can be developed to scan through hundreds of medical images for irregularities that can save doctors a lot of time.
But that also means many companies are using the technology in areas that do not really require an artificially intelligent program.
"I would argue that AI is overkill for the majority of use cases," Drew Perez, chief executive officer at Adatos, said during a panel discussion and added that the benefits of AI are not instantaneous in many cases. His start-up uses AI to study satellite and drone images of agricultural lands to assess things such as tree counts, soil conditions and plant health. "At the end of the day, if you think about it, it has to have a return on investment."
Perez told an audience that, most AI programs today are in the "lab phase or innovation phase." He explained that even if there are the right conditions — including having the right amount of computing power, sufficient data, the right mix of talent and a culture that readily embraces AI — profitability is not guaranteed.
"You might go for a whole year and find out at the end that, for a million dollars, I'm going to make a hundred thousand [dollars]," he said, explaining that short-term plays in the stock market, for example, could likely give a better return on investment. Unless a company is able to figure out an application for their AI program that can bring in "hundreds of millions, it's just a lab experiment," he said.
Monetizing an AI program is only a problem if it fails to address an important problem, according to Steve Leonard, founding chief executive officer at Singapore-based start-up accelerator SGInnovate.
"Some people say, 'Oh, I'm having a hard time monetizing.' For me, the first question is what led you to work on that problem, so if you're having a hard time monetizing, it must mean that you didn't have a problem that was sufficiently painful for somebody to take an action," Leonard told CNBC.
While building an AI program may not be a significant barrier, companies need a lot of high-quality data to train that algorithm. That, according to Leonard, is both a challenge and an opportunity.
It's an opportunity because large volumes of data can be used to train and improve what those AI programs can do in a more efficient manner. For example, an algorithm could study millions of medical scans of the human brain to learn about irregularities and automatically detect them in future images.
But the challenge is to make sure people feel comfortable with using that data, Leonard said, noting that making the information anonymous is oftentimes beneficial.
He explained that a government's ability to provide high-quality data is defined by its citizen' comfort and appetite to share that information.
Another challenge is in cleaning up data sets they can then be used to train AI programs. Two years ago, IBM estimated that the cost of poor-quality data was about $3.1 trillion a year in the United States alone. Bad data can be expensive because cleaning it up usually takes a lot of time and effort and so that makes it harder for companies to be immediately profitable.
Regulators play an important role in both creating an environment where companies can innovate and also to protect consumers' data rights.
"It's a fine balancing act," said Tan Kiat How, chief executive officer of Singapore's Infocomm Media Development Authority. One of IMDA's many functions is to regulate the use of personal data in Singapore.
Europe rolled out its general data protection regulation laws in late May that saw many tech companies update their terms of services to be more transparent about the information they retain. Many experts have suggested the GDPR laws could provide a guideline for other regions or countries to adopt similar data privacy schemes. That, some said, could limit innovation and growth in many technologies that rely on collecting and using data.
Tan, for his part, said data privacy issues are actively being discussed in Singapore and within the region.
"The important principle around our approach is that we have to be nimble and constantly adjust and ensure that we're ahead," he told CNBC. He explained that having regulatory certainty is very important for companies and consumers. That means the objectives of a regulator like IMDA should be fixed, but the way such goals are attained can be flexible, he said.
Indeed, the IMDA regularly works with the Singapore government, industry, consumers and the academia to shape the discussions around AI and other technologies in the city-state. Recently, Singapore also announced it would set up a council to advise the government on the ethical use of AI and data.
SGInnovate's Leonard said it's important to raise awareness about the benefits of AI. That could make people be more willing to share their personal data in a way that companies can use to train AI programs.
Leonard said it's a matter of helping people understand the benefits versus "AI will eliminate jobs or AI will be sort of a brain in the sky and it will tell us what to do."