“There’s intelligence in AI systems, if you want to call it that. But it’s not organic intelligence, and it doesn’t play by the same rules humans do.” — James Vincent, The Verge
A funny thing happens every couple of years in the tech sector: A new buzzword emerges, and it starts appearing everywhere. And the more it shows up, the more it gets misused.
We love artificial intelligence. AI is absolutely at the core of Directly’s value proposition.
But, sadly, the term “AI” has become widely misused and misunderstood. Especially the “intelligence” part. There’s a perception that you can just magically turn on AI and machine learning algorithms and — poof! — instant intelligence. And that AI in business will replace the need for human intelligence.
It’s simply not true. Because it’s the human that makes for truly smart AI.
Let me explain.
The 3 key ingredients to truly smart AI in practice
“AI is akin to building a rocket ship. You need a huge engine and a lot of fuel. The rocket engine is the learning algorithms but the fuel is the huge amounts of data we can feed to these algorithms.” — Andrew Ng, Wired
Artificial intelligence starts, simply, as code. Usually in the form of an algorithm, this code has the capacity for learning. But there’s nothing smart about the code on its own -- except the developer that wrote it. The algorithm needs two more pieces before it can become truly smart AI in practice.
It needs a data set. And the bigger the data set, the better. The machine-learning powered engine that drives Netflix’s recommendations is based partly on viewing data from millions of other members. Once you apply an algorithm to a data set — and give it a task (e.g. “Go find a movie you think I’ll like”) — it can start to function. The dataset might be a database of movie titles it can crunch. But even if you have massive amounts of data (which most companies don’t!), the algorithm won’t be smart about that task yet because it doesn’t know anything about you. Netflix, as an example, wouldn’t be able to personalize any recommendations if it didn’t know what you watch.
It needs training. This is where the human comes in. When I tell Pandora that I like a song, it’s AI engine will be smarter about future song choices it makes. And the more TV shows I watch on Netflix, the more data the recommendation engine has to suggest titles it thinks I’ll like. I’m training these AI services as I use them, giving them the intelligence they need to get better at their intended purpose.
In some ways, AI is like a baby. A newborn has a brain on board, but until that baby experiences and learns from life events — including guidance from experts like its parents — the brain won’t be intelligent.
Why (expert) human training matters: AI gone wrong
“It has to be said that chatbots, though they are selling like hot cakes today, are mostly stupid and disappointing.” — Thomas Gouritin, TheNextWeb
There are many, many examples of AI failures -- often problems rooted in a lack of expert human training.
Remember in 2016, when Facebook Messenger opened up its platform for developers to create custom chatbots? It spurred a digital gold rush because thousands of businesses couldn’t resist the chance to reach the 1.3 million active Messenger users. The problem was that excited users expected intelligent interactions. And these bots mostly failed to deliver on those expectations. The bots simply can’t learn or be trained, so we’re left with thousands of simple-minded digital denizens that are, as the quote above says, “mostly stupid and disappointing.”
And then there was Microsoft’s AI-powered Twitter chatbot Tay. Tay was smart in that it had the capacity to be trained. Unfortunately, it got trained by the wrong people. Shortly after its release in 2016, a few Twitter users manipulated the bot into sharing hate speech. As The Verge reported, “when Tay was fed sexist, racist, and other awful lines on Twitter, the bot began to parrot those vile utterances and, later, began to adopt anti-feminist and pro-Nazi stances.”
The takeaway here: Good training -- expert training -- matters.
What our expert-trained AI looks like
“To nail a conversation, a digital assistant needs to be told over and over when it’s failed.” — Matthew Hutson, Bloomberg Businessweek
There are many different approaches to training AI, depending upon the application. Some AI learns from monitoring online human behavior, just as Amazon might suggest purchases based on what you bought in the past. Some companies will train AI using internal team members. Other companies will hire “clickworkers” — people who get paid to train AI on everything from image recognition and text analysis to audio recordings.
Directly takes AI training a step further. Our AI-powered CX automation platform is smart because it deploys expert super-users on the products and services our clients sell. For example, who better to answer support questions about driving for DoorDash than... a driver for DoorDash? That’s what Jon Gallez does, as reported in the San Francisco Chronicle. On our platform, these experts train AI as it responds to different customer support inquiries. Multiple experts validate each answer and suggest edits when an answer is insufficient. The end-result is that each AI response benefits from thousands of hours of combined expert experience. And through this expert training, our AI gets smarter over time.
“Through such collaborative intelligence, humans and AI actively enhance each other’s complementary strengths: the leadership, teamwork, creativity, and social skills of the former, and the speed, scalability, and quantitative capabilities of the latter.” — H. James Wilson & Paul R. Daugherty, Harvard Business Review
No question, we’re entering exciting times when it comes to the application of AI. Intelligent assistants like Amazon’s Alexa and Apple’s Siri are giving us a taste of that promise. And we’re entering a wave of applied AI in business, where eventually virtually every workplace application will have machine learning built in.
But make no mistake: You can’t have truly smart AI without training. Human training. And even better: expert training.
Want to learn more about AI-powered CX automation in action?
We’ll be publishing more blog posts about the intersection of AI and customer experience, so make sure to join our email list and follow us on Twitter, LinkedIn, and Facebook.
And if you’d like to see AI training in action, contact us today to set up a demo of Directly’s CX automation platform.