It might have been a forcing function of the global pandemic, or it could have been accelerated by ongoing advances in compute power. Whatever the cause, artificial intelligence and machine learning have become key information technology assets for many large enterprises.

A McKinsey & Co. survey released in November found that half of respondents had adopted AI in at least one function, and half reported revenue increases as a result. The technology is being widely deployed today to fill prescriptions, predict the weather, recommend movies, grow crops and deliver pizzas. In other words, it has made itself at home in the enterprise, from the mailroom to the boardroom — not just for whizzy new apps but mainstream business applications.

“When I go to business leaders, I ask what their biggest business problems are,” said Andrew Ng, founder and chief executive of Landing AI. “I don’t ask what their biggest AI problems are.”

Ng, one of the pioneers in the AI field having led previous initiatives for Baidu and Google LLC, spoke during the recent virtual EmTech Digital conference, hosted by MIT Technology Review in late March. His remarks echoed not only McKinsey’s findings, but those of a number of entrepreneurs and scientists who are at the forefront of the AI field.

Understanding complex text

General acceptance of AI is powered by the reality that progress is accelerating and use cases are growing as a result.

Researchers at OpenAI are continuing to work on their lofty goal of developing artificial general intelligence, machine technology with the reasoning and learning capability of the human brain. Microsoft Corp. and Elon Musk are major investors and last year the company announced a commercialized release of GPT-3, OpenAI’s language model that employs deep learning to produce humanlike text.

Ilya Sutskever, co-founder and chief scientist at OpenAI, indicated during the EmTech Digital event that an iteration of GPT-3 – DALL-E – is capable of creating accurate images from fairly complex textual commands. An instruction to “create a paring of an owl sitting in a field at sunrise in a pop art style” resulted in a believable rendering, based on the examples displayed by Sutskever during the virtual MIT event.

“This is not the endpoint, it is a first step toward multimodel systems,” Sutskever said. “Deep learning is the one idea you don’t want to bet against.”

There are signs that the AI field is moving beyond the research and development lab and into operational support. Scale AI Inc. was founded in 2016 to address the issue of data labeling, the time-consuming work needed to make correct identifications of still images, voice, video and text for users to build machine learning models.

The company has reached a $3.5 billion valuation and its labelling technology has become especially popular within the autonomous vehicle world, in use by clients such as General Motors, Honda and Lyft. Being able to scale machine learning infrastructure has been a key factor behind Scale AI’s growth.

“We think this paradigm is critical for the mass expansion of AI,” said Alexandr Wang, chief executive and founder of Scale AI. “We’re at this tipping point for the deployment of AI throughout many different industries.”

Chip technology for AI

Meeting the significant computing demands of building AI programs and models may be in for a radical change soon as well. Lightmatter Inc., a startup incubated at MIT, is on track to ship its first light-based AI chip this fall.

Lightmatter has designed its unusual technology around using wavelengths of light for simultaneous computation, a process known as photonic computing. Only a limited amount of energy is required, because light is cooler than electrical power.

The company claims that its chip, called Envise, can run a natural language model five times faster and at one-sixth the power of top-of-the-line AI chips on the market today.

“With light, you can have data represented in different colors,” explained Nicholas Harris, founder and CEO of Lightmatter. “We can actually send multiple datasets through the system at the same time. It’s like virtual cores, you get big energy savings and big throughput.”

Even with this and other potentially groundbreaking advances still on the horizon, adoption of AI and machine learning for an expanding number of use cases has not been held back.

AI is now widely deployed by global credit card companies, such as Mastercard Inc., to spot fraudulent transactions. Mastercard has taken an approach where it uses AI to let as many card purchases through as possible, and must render a decision on each one in 50 milliseconds or .05 seconds, according to Ed McLaughlin, president of operations and technology at Mastercard.

“AI isn’t a strategy,” said McLaughlin. “That’s like saying the strategy of a ship is to float. It’s just part of the environment, tools that we have.”

A plan for machine learning

Amazon.com Inc. made a decision more than 20 years ago to adopt AI and machine learning for nearly every aspect of its business and then turned that into a portfolio of products for its customers. In November of 2018, Amazon Web Services Inc. introduced 13 new machine learning products in one event alone.

Amazon’s AI and machine learning customers range from Domino’s Pizza which uses the technology to forecast demand and have orders out the door within minutes, to support for prescription processing within Britain’s National Health Service, to helping farmers make smarter and more precise decisions about their crops through a Bayer Digital initiative.

“Ten years ago, every business unit at Amazon was asked how they planned to use machine learning,” said Michelle Lee, vice president of the Machine Learning Solutions Lab at AWS. “It was not acceptable to say you didn’t have a plan.”

While use of AI and machine learning continues to grow, so have concerns around how the technology will be used. Although an investor himself in a number of AI-related projects, Tesla Inc.’s Musk told a SXSW gathering in 2018 that AI was “far more dangerous than nukes.”

Concerns such has these have prompted 42 nations of the Organization for Economic Cooperation and Development, including the United States, to agree on a set of principles for responsible stewardship of trustworthy AI. Approximately 30 countries have adopted AI policies to govern how the technology will be used for education, economic policy and as an algorithm within government agencies.

Potentially more significant is new regulation by the European Union which is making its way toward potential adoption later this year. Early drafts of the proposal appear to be following a path similar to General Data Privacy Regulation or GDPR which affected all companies doing business in the European Union.

“This is a technology at the very early stage of its implementation,” said Julia Reinhardt, international strategy consultant for tech and policy, Fellow-in-Residence at Mozilla. “It’s already having an impact on people and society. How do we make sure AI works equally for all in a way that preserves our fundamental values and the rule of law?”

What pervades the AI business right now is a sense that growing enterprise adoption will fuel a renaissance in the computing world as more firms find new uses for the technology and an expanding number of eager entrepreneurs build novel ways to deliver on that.

“In the end, AI is math and AI is software,” said Dario Gil, senior vice president and director of IBM Research. “The scientific method should be a tool for business. What we’re about to enter is an era where discovery is at the forefront. We’re living in the most exciting time in computing since the advent of computing.”

Image: Pixabay Commons