M+E Daily

Static image recognition is expected to generate the most artificial intelligence (AI) revenue among all use cases on a cumulative basis from 2016-2025, according to projections in a new white paper by market research company Tractica.

Static image recognition, classification and tagging clearly represents the “largest opportunity” for AI during that period, Aditya Kaul, Tractica’s research director, said during an Oct. 6 webinar highlighting key projections and other details from the report. The webinar was sponsored by the AI World conference, which also co-published the white paper.

Image recognition, classification and tagging is already one of the most popular and widely-applied use cases for AI today, Tractica said in the white paper, projecting it will account for $8.1 billion in cumulative spending from 2016-2025.

Companies including Apple, Facebook, Google, Microsoft and Yahoo are “actively using” image recognition and classification algorithms on consumers’ photos, which are either uploaded or stored on devices, the research company said. It pointed out that the main purpose of consumer-oriented applications for image recognition and classification is to help people automatically segment, tag and store images for better data mining and retrieval, which can be done via search or similar photo recommendations. Photo upload sites including Apple Photos, Flickr and Google Photos all use AI image recognition and tagging techniques to automate photo classification and tagging, Tractica said.

The top three AI use cases are expected to each generate more than $7 billion in cumulative revenue during that period, Kaul said on the webinar. The use of AI in algorithmic trading and strategy performance is expected to be No. 2, followed by patient data processing.

Hedge funds are the “main drivers” for the use of AI in the investment sector today, Tractica said, pointing out that hedge funds have a larger “appetite for risk” than investment banks. There are more than $2.3 trillion of hedge fund assets under management today, so “there is an enormous opportunity that lies in identifying the right strategies and algorithms to grow that pool of investment,” it said, adding that companies including Aidyia, Sentient, and Walnut Algorithms are using AI tools to build bots that analyze market indices, market volumes, macroeconomic data, geopolitical events, and other inputs that could convince markets to develop a trading strategy.

Patient data processing is “not very big now,” Kaul said on the webinar, but he added that it “has a lot of potential in the future.” He pointed to IBM’s recent $2.6 billion purchase of Truven Health Analytics as just one example of initiatives in that sector.

The top 15 AI use cases are expected to represent cumulative spending of $58.7 billion from 2016-2025, with the advertising industry expected to see the highest spending at $8 billion, Tractica projected. But there is expected to be a large drop-off between the revenue generated by the top three use cases and the rest of the top 15.

The firm expects the AI market to “make relatively small but significant gains in the next five years, until 2020, after which the market really kicks off,” Kaul said on the webinar. “We see an inflection point happening” in about 2020-2021, he said.

The global size of the AI market in terms of software revenue is “growing at a very, very fast pace,” and is expected to grow from $644 million in 2016 to $37 billion by 2025 with compound annual growth (CAGR) at about 57% in that 9-year period, he said.

AI use cases were categorized into three areas in the white paper: big data, vision (images or video-based applications to recognize images, faces and objects) and language. There’s an “almost even split” between use cases having big data and vision components, with language a “much smaller” use scenario, Kaul said.

The vision component of AI is expected to “drive a lot of hardware requirements with much bigger spend needed around vision algorithms because they require much bigger scale to drive bigger performance,” he said. But AI is shifting away from big data use cases that typically involve only routine processing of data to efforts involving the replication of human abilities, he said.

Despite the huge growth potential of AI, he warned that companies should not rely on it as a “panacea for solving all problems.” AI mostly works in a structured data setting, not for unstructured data, he said. It’s “important to understand the limitations of AI rather than sort of blindly adopting it,” he said, adding: “There is a massive gap between what consumers expect and what is currently achievable, so I think there’s a massive risk of overselling AI.” It’s better to downplay AI’s capabilities and explain its shortcomings rather than oversell it, he said, pointing to text-based chatbots as being oversold and failing to meet the Turing test in which they would be indistinguishable from humans.