“Alexa, order me a pizza.”

If you’re anything like me, you’ve already tapped into the power of AI for simple tasks like ordering food or playing your favorite jam.  The truth is artificial intelligence (AI) is behind countless aspects of modern life, both in business and at home. It’s the reason we have personalized social feeds or automated recommendations advertised to us based on previous purchases and pageviews.

Artificial intelligence is here to stay, and it can be immensely helpful across a range of industries. It also has a number of potential pitfalls. But before we delve into those, let’s define the different types of AI.

Defining AI Technology

There are two types of AI technology: narrow (weak) AI, and general (strong) AI. Narrow AI is designed to perform a specific task, like playing poker or facial recognition. Most of the AI technology we use, as of 2018, is considered narrow AI. It’s often swapped with terms like machine learning or automation. Because of its granular applications, narrow AI can be incredibly effective at saving time and money on bland or repetitive tasks.

In contrast, general AI has a more comprehensive knowledge base, similar to that of a human. However, it’s capable of processing information faster and more accurately. General AI is the type often portrayed in sci-fi movies — think Data from Star Trek, or Schwarzenegger’s android character, the “terminator.” Google’s search engine, though immensely complex, is a primitive example of general AI. These days, you can search for items using slang and natural speech, and Google can typically derive your intent from your query. Just a few years ago, Google AI was incapable of understanding nuances in search queries; it could only match keywords with other similar keywords.

The Pitfalls of Artificial Intelligence

People have been talking about the big ethical implications of general AI for years. Obviously, if a self-driving car or piece of medical equipment became self-aware and made an autonomous decision, people could get hurt.

Narrow AI, by contrast, is designed to specialize — meaning we won’t have to worry about it developing consciousness and taking over the world anytime soon. However, there are still pitfalls to narrow AI as we know it, especially when it’s being used to improve workplace efficiency and grow your business. Here are the pitfalls of narrow AI that may not seem as obvious.

Pitfall #1: It takes time to set up.

“Automation” is a promising word. And the businesses who have successfully automated some of their menial tasks are, undoubtedly, saving time and money that they otherwise would have spent on salaries and training.

But implementing new apps, systems, or processes require time to research, set up, and test properly. After that, the AI system carries a learning curve for each human who will be interacting with it — meaning both your employees and your customers will have to learn how to get the most out of it.

This isn’t to say that artificial intelligence is waste of time. However, if you plan on automating some of your manual business tasks, you can count on a mile of preparation and maintenance upfront in exchange for an inch of reward.

Pitfall #2: There are so many possibilities.

The vastness of narrow AI may sound like a positive feature, but when you’re new to AI, it can be hard to sort effective uses from not-so-effective ones. Do you really need to send an automatic email every time a customer takes a specific action? Can (and should) a computer be syndicating social media, reaching out to new leads, nurturing old ones? Once all of the possibilities start to arise, they can feel overwhelming.

It’s also difficult to estimate ROI before you’re actually implemented a new AI system and had some time to gauge whether its speed and accuracy can really replace a human doing the same work.

Pitfall #3: AI may not (yet) be ideal for customer service.

Chatbots have been hyped up over the last couple of years, and for good reason: they’re potentially very profitable for companies who manage a large customer service staff or handle a lot of the same customer inquiries. And chatbots that use deep machine learning, like Cortana or Siri, can be extremely helpful.

However, custom chatbots that are programmed to answer a limited range of questions (specifically, those delivered through messaging apps) can be unhelpful to customers and downright creepy, at worst. Customers seem to be the most frustrated when they’re led to believe a chatbot is a human.

According to a DigitasLBI report, 73 percent of Americans said they wouldn’t use a chatbot again after a bad experience — and further, 61 percent would actually become more frustrated with a chatbot than a human if it couldn’t answer their questions. It seems, then, that when dealing with processes that involve nuances of language, emotional intelligence, or very complicated products, humans are still much better suited for the job.

Like with any cutting-edge technology, the only way for a business owner to reap the rewards of AI is to research, plan, and test. Enlist an automation expert if you don’t have the time to do it yourself, and keep experimenting and tracking ROI to make sure artificial intelligence is serving you rather than costing you.