Artificial Intelligence Will Define Google’s Future. For Now, It’s a Management Challenge.

In the most high-profile incident, last month Google parted ways with a prominent AI researcher, Timnit Gebru, after she turned in studies critical of the company’s approach to AI and complained to colleagues about its diversity efforts. Her research findings concluded that Google wasn’t careful enough in deploying such powerful technology and was callous about the environmental impact of building supercomputers.

Mr. Pichai pledged an investigation into the circumstances around her departure and said he would seek to restore trust. Ms. Gebru’s boss,

Jeff Dean,

told employees that he determined her research was insufficiently rigorous.

Nearly 2,700 Google employees have since signed a public letter that says Ms. Gebru’s departure “heralds danger for people working for ethical and just AI—especially Black people and People of Color—across Google.”

Last week, Google said it was investigating the company’s co-head of ethical AI,

Margaret Mitchell,

for allegedly downloading and sharing internal documents with people outside the company. Ms. Mitchell, who has criticized Mr. Pichai on Twitter for his handling of diversity issues, didn’t respond to requests for comment.

Alphabet’s approach to AI is closely watched because the conglomerate is widely seen as the industry leader in sponsoring research—both internal and external—and developing new applications for the technology, ranging from smart speakers to virtual assistants. The nascent field has raised complex questions about the growing influence of computer algorithms in a wide range of public and private life.

Google has sought to position itself as a standard-bearer for ethical AI. “History is full of examples about how technology’s virtues aren’t guaranteed,” Mr. Pichai said in remarks to a Brussels think tank last week. “While AI promises enormous benefits for Europe and the world, there are real concerns about the potential negative consequences.”

Google’s push to advance in AI through acquisition has also added to its management challenges. In a previously unreported move,

Mustafa Suleyman,

co-founder of Google’s London-based artificial-intelligence arm, DeepMind, was stripped in late 2019 of most management responsibilities after complaints that he bullied staff, according to people familiar with the matter.

DeepMind, which Google bought in 2014, hired an outside law firm to conduct an independent probe into the complaints. At the end of 2019, Mr. Suleyman was moved to a different executive role within the AI team at Google.

DeepMind and Google, in a joint statement, confirmed the investigation into Mr. Suleyman’s behavior and declined to say what it found. The statement said that as a result of the probe Mr. Suleyman “undertook professional development training to address areas of concern, which continues, and is not managing large teams.” The companies said that in Mr. Suleyman’s current Google role, as vice president of artificial-intelligence policy, “he makes valued contributions on AI policy and regulation.”

In response to questions from The Wall Street Journal, Mr. Suleyman said that he “accepted feedback that, as a co-founder at DeepMind, I drove people too hard and at times my management style was not constructive.” He added, “I apologize unequivocally to those who were affected.”

Mr. Suleyman’s appointment to a high-profile role at Google bothered some DeepMind staff, as did a tweet from Mr. Dean welcoming him to his new role at Google, according to current and former employees.

“I’m looking forward to working with you more closely,” Mr. Dean wrote in the tweet.

Unlike in Google search, in which artificial intelligence is employed to reorganize and resurface existing public information, DeepMind has focused largely on health issues, such as crunching vast amounts of patient data to figure out new ways to treat disease.

DeepMind has struggled to meet financial expectations. The unit had a loss of $649 million in 2019, according to a U.K. filing last month. Google forgave more than $1 billion of loans to the unit over the same period, the filing indicates.

Google has tried at various times to construct broader oversight of its artificial-intelligence projects, with mixed success.

A DeepMind independent review board, meant to scrutinize the unit and produce a public annual report, disbanded in late 2018 after board members chafed at what they said was incomplete access to DeepMind’s research and strategic plans.

Months later, an external AI ethics council created by Google was disbanded after a week, following an employee petition and other protests about right-leaning board members. A Google spokesman said at the time that the company would “find different ways of getting outside opinions on these topics.”

Google’s artificial-intelligence efforts date back at least a decade, and the area has been a stand-alone division under Mr. Dean since 2017. Mr. Pichai, in announcing the new structure that year, said that AI would be central to the company’s strategy and operations, with advanced computing—also called machine learning—threaded throughout the company.

University of Washington computer-science professor

Pedro Domingos

said Google has been well aware of the many pitfalls associated with AI. He remembers Alphabet board chairman

John Hennessy

in a chat several years ago describing his biggest fears as the search giant intensified its push in the field.

Mr. Domingos says Mr. Hennessy told him that if anything went wrong with AI—even at other companies—Google would be blamed.

“We are one misstep away from everything blowing up in our faces,” Mr. Hennessy said, according to Mr. Domingos. Mr. Hennessy said he doesn’t recall the specifics of the conversation but said Mr. Domingos’s recollection is probably right.

One early misstep came in 2015, when several Black Google users were surprised to see that the artificial-intelligence technology in the company’s free photography software had automatically labeled one of their albums of human photos, “gorilla,” based on skin color.

Google apologized and fixed the problem, and publicly redoubled promises to build internal safeguards to ensure its software was programmed ethically.

The effort involved hiring researchers such as Ms. Gebru, the former co-head of the company’s Ethical Artificial Intelligence team, who has been outspoken on the limitations of facial-recognition software in identifying darker-skinned individuals.

She was one of hundreds of employees who produced research, sometimes with academic institutions, on AI. Unlike Google’s other engineers, the AI group functioned more like an academic department—tasked with debating larger issues rather than troubleshooting products.

Reuters later reported that before Ms. Gebru’s departure, Google had launched a review of sensitive topics and in at least three cases asked staff not to cast its technology in a negative light. Google declined to comment on that.

Ms. Gebru and Mr. Dean didn’t respond to requests for comment for this article.

Mr. Domingos said Google’s artificial-intelligence advancements have helped create faster and more accurate search results—as well as more relevant advertising.

Mr. Domingos said most of Google’s problems related to AI are rooted in the company’s approach to managing staff, adding that science, and not ideology, should guide ethical debates.

“Google is the coddler-in-chief,” he said. “Their employees are so coddled that they feel entitled to make more and more demands” regarding how the company approaches AI and related issues.

Ms. Gebru’s departure highlights the challenge Google leadership faces, as the two sides can’t even agree on whether she resigned, as the company says, or was fired during her vacation, as she has said.

Write to Rob Copeland at rob.copeland@wsj.com and Parmy Olson at parmy.olson@wsj.com