AI is not smart enough to solve Meta's content-policing problems, whistleblowers say

• Facebook whistleblower Frances Haugen appeared at an event with ex-Facebook moderator Daniel Motaung.
• Haugen said although Meta talks up AI as a moderation tool, it is too blunt an instrument.

Artificial intelligence is nowhere near good enough to address problems facing content moderation on Facebook, according to whistleblower Frances Haugen.

Haugen appeared at an event in London Tuesday evening with Daniel Motaung, a former Facebook moderator who is suing the company in Kenya accusing it of human trafficking.

Meta has praised the efficacy of its AI systems in the past. CEO Mark Zuckerberg told a Congressional hearing in March 2021 the company relies on AI to weed out over 95% of "hate speech content." In February this year Zuckerberg said the company wants to get its AI to a "human level" of intelligence.

But according to Haugen and Motaung, this is a smokescreen that obscures the work done by thousands of human moderators, some of whom suffer severe mental health issues they say come as a result of their work.

AI is just not as smart as Silicon Valley makes out

Haugen said AI at the moment is nowhere near intelligent enough to interpret the nuances of human speech.

She said using AI for moderation means there's always a tradeoff between having an overbearing algorithm that removes content that doesn't violate content policies, and having an algorithm that misses instances of such content.

"Those AIs can only be so precise," she said. "People who actually work on it call it machine learning, because it is not intelligent."

"It is only learning inferences, statistical inferences from the data that is fed in."

Her comments chime with an ex-Facebook employee who said in an internal farewell note in late 2020: "AI will not save us."

Meta's focus on AI is in part driven by a desire to cut costs, Haugen claimed.

"Facebook is always trying to ... have more of the work, be done by computers, by AI, and less being done by people," Haugen said, saying this is because employing human moderators is more expensive than using automated systems.

When Facebook's AI systems can't make a judgement call on a piece of content, it's up to human moderators to review it.

Motaung, who worked as a contracted Facebook moderator in Nairobi until he was fired in 2019, said he doesn't believe AI systems can do the same job as a human.

"The element of what is moral, what is ethical, comes in — and artificial intelligence cannot do that," Motaung said.

Human moderators pay the price with their mental health

For moderators like Motaung the work can take a severe toll on their mental health.

Motaung said in his lawsuit against Meta that working as a moderator and viewing graphic, disturbing content all day resulted in him getting PTSD.

Motaung said Tuesday the work left him "broken."

During Tuesday's event both Motaung and Haugen said Meta could bring about changes to make work for its content moderators safer.

"Even saying: 'Hey, you are looking at things that traumatize you every day. We're going to pay you for a full week, but you only have to come in every other day ... that is a real intervention," Haugen said.

"That would radically reduce the harm, and they choose not to spend even that small amount of money," she added.

Motaung and Haugen both said Meta outsources its moderation work to contractors in order to limit liability.

Around 15,000 people globally work as moderators for Facebook and Instagram, a 2020 report from the New York Unversity Stern Center for Business and Human Rights said. Most are employed by third-party contractors.

In a message to Mark Zuckerberg, Motaung said he wants to know if Zuckerberg would intervene to help moderators. "Go and take care of it. Do what's right," Motaung said.

Meta did not immediately respond when contacted by Insider for comment.