AI Can’t Detect Sarcasm: Why Content Moderation Needs a Human Touch

The way bros be yapping is low-key changing and affecting the aura – no cap. I mean…

The way people communicate is complicated and constantly changing. Just think about slang. What the “kids” say can be in one day and out the next. AI can’t be one of the cool kids. Whether it’s the lingo of the times or a tone of sarcasm, the technology can’t typically pick up on the nuances.

That’s why customer interactions and content moderation need both the human touch and intelligent technology. Businesses should not just hand over such important processes without keeping humans in the loop.

Building more trust in AI

Certainly, AI is a critical part of the process because it can analyze huge data sets fast, 24 hours a day. But, on top of a lack of emotional intelligence, AI also has a trust issue—counter to what Trust and Safety work seeks to accomplish.

According to Salesforce data, 80% of consumers say it’s important for humans to validate AI outputs, and 89% call for greater transparency. They expect to know when they’re communicating with a bot. 

Business leaders, on the other hand, are confident in what AI can do for their bottom line, as 60% believe AI will improve customer relationships.

Bridging the gap requires building more trust. People driving AI (and not the other way around) is the way to accomplish this.

Freeing people for the complex work

When it comes to content moderation, or the ability to analyze and respond to online customer interactions, some of the more common AI communication issues can also seep their way into an automated content moderation tool. The effectiveness of AI tools is only as good as their Large Language Models (LLMs) or the models that can generate actions based on the data upon which they are trained.

For example, a company may have trained the model on data that doesn’t account for different cultures or colloquialisms. This could create scenarios where an AI content moderation tool incorrectly flags a comment or review.

Let’s take the word “sick” for instance. In certain spaces, the word “sick” is an adjective for something good, not a reference to feeling ill. Someone may use the term to describe a product they were satisfied with or to praise a sales rep for their great customer service. In this scenario, the right response depends on context. If the enterprise hasn’t trained the LLM on this slang term, the content moderation solution may flag a positive comment as either offensive or negative. This could create a scenario in which the customer no longer trusts your online platform enough to leave positive reviews. 

The consequences of customers not believing in the prioritization of their trust or safety in online spaces include:

• Lost trust and confidence in the brand
• Exposure to legal issues, fines, and poor customer satisfaction reviews
• Lost advertising dollars

The risks inherent to poor content moderation are too significant to solely trust AI with the job.

See also: Self-Deprecating Robots are Better Conversationalists

Enhancing content moderation with AI – and the right human touch

With the right combination of AI-driven content moderation and human intervention, enterprises can experience positive business results. The right content moderation approach will be beneficial for both your staff and your customers.

First, the right content moderation platform will allow you to train your LLMs to differentiate where and when the human moderator should step in.

For example, AI should always handle sorting customer requests and responses into different categories so your content moderation staff isn’t stuck reading thousands of customer messages. Also, you should train your LLMs to flag and remove general toxic or inappropriate content that is profane and racially insensitive.

However, if the inappropriateness in a response escalates to a certain level or is unclear to the AI model, then the human is there to step in and remediate the situation. The human moderator should be an expert on nuanced context specific to your business. Armed with the proper business context, they can approach content moderation in a way that puts your brand reputation first.

As your human moderators interact with customer responses, their psychological safety must be of the utmost priority. Although the AI tools will handle much of the incoming content, your team could experience exposure to an inordinate amount of offensive language. This can create damaging mental stress. It’s a good idea to work with your HR and/or people team to enact mental health training and resources that prioritize the mental safety of each content moderator.

Prioritizing customer trust and safety

Research shows consumers place great value on the customer experience. The ability to share feedback on a business’ online platform is an important part of that experience. This is why content moderation should include the right balance of AI and human involvement.

When customers are aware a human moderator is working alongside an AI-powered platform, it can help dissuade AI skepticism while improving each customer interaction. Ultimately, it will display a commitment from you to the customer that their trust and safety is your top priority.