Tech companies' legal leaders and policy experts gathered at Santa Clara University School of Law on Feb. 2 for a day of panels on content moderation and removal. The school's High Tech Law Institute brought representatives from Facebook, Google, Reddit and others to discuss many facets of content moderation, including mental health, artificial intelligence and transparency.

Here are some big takeaways from a day of discussion featuring some of Silicon Valley's most famous companies:

1. Artificial Intelligence Can Remove Content Faster, but Not Better

When it comes to content moderation, context is key. But, as Facebook Inc. public policy manager Neil Potts said Friday, ”Automation is not great at context yet.”

At one panel session, “Humans vs. Machines,” Potts and other panelists agreed that while AI can be useful for black-and-white cases such as posts that endanger or exploit children, a human eye is needed to make most removal decisions accurately.

“It's hard to tell if a review is racist, or if the review is describing a company that was racist,” Yelp Inc. deputy GC Aaron Schur said, providing an example. “[That] requires a human eye for judgment.”

Panelists also noted that changing laws—such as those in Europe, and particularly the U.K.—forcing sites to take down harmful content in a short period of time could force platforms to rely on less accurate, nonhuman content moderators.

2. Human Content Moderators Are Only Human

Spending all day, every workday, looking at graphic images and disturbing posts online takes a toll on mental health. Human content moderators can get worn down from constant exposure to the worst parts of the internet, according to panelists at another session, “Employee/Contractor Hiring, Training and Mental Well-Being.” If moderators feel burnt out, their work may suffer.

Panelists said their companies have a variety of ways to help moderators stay healthy, including counseling and massage therapy.

“We have a lot of different wellness-oriented perks because we really believe that human moderators are the key to having high-quality moderation,” said Charlotte Willner, trust and safety manager of Pinterest. She recommended that companies “invest in their skill set, teach them to become familiar with this type of content, [and] invest in their long-term health.”

3. The Community Knows Best (Kind Of)

During another session, on if and when to outsource moderation, panelists from Reddit, Wikimedia and Nextdoor offered similar advice—turn to the community first. They all said community self-regulation is their sites' most common form of content moderation.

“We let communities decide what the rules are for that community, decide what should go in and out there,” Reddit counsel Zac Cox said. “People who join [the] community can follow the rules and help enforce them by flagging content or commenting in a way that reinforces those norms.”

Many platforms, such as Reddit and Yelp, include features that allow users to up-vote a post or mark it as useful. This is another form of self-moderation, Cox said, as it allows popular content to rise to the top and pushes content the community doesn't want to see down.

But most of the companies said they did have clear processes in place for escalating a content management situation that's too large to be handled by the community alone, or isn't being handled properly.

4. Be Transparent, but Don't Help the Bots

Companies have a fine line to walk with content moderation rules. The guidelines should be clear enough that users know the repercussions of abuse on the platform, but not so clear that users, or bots, can game the system, panelists at the event said.

If users and bots can figure out that harmful content won't be removed unless it contains a specific word or slur, according to panelists, they may continue posting disturbing content that doesn't technically violate guidelines.

When a post does get removed, Patreon Inc.'s head of legal Colin Sullivan said it's crucial that moderators take the time to hear out users' appeals and explain what happened.

“Creators think we're making a fair decision about them, so they feel like we're making fair decisions in general,” Sullivan said.

Sullivan and Medium's head of legal Alex Feerst said transparency becomes more difficult when the user seems to be a bot. Alerting a bot that's account has been shut down or has violated guidelines may push the bot to create another account. In this case, they said, it could be better to isolate the account and not inform the bot.

5. Diversity Is Key

Diversity is important in every part of the company, and that holds true for trust and safety teams. Content moderators should be well-trained, but panelists at Friday's event noted that having moderators from different backgrounds can lead to a better conversation about what is or isn't harmful content.

Feerst said Medium has rotations into the moderator role so that people from around the company can spend time doing trust and safety work.

“Trust and safety and content moderation is a field that, if you don't have gender and ethnic and other [types of diversity], you can't do it,” Feerst said. “Because you don't have enough perspective to generate the cultural competencies, and, most importantly, you don't know what you don't know.”