Aug 9, 2017

I’m Sick and Tired of Hearing about “Killer Robots”

P.V. Kannan, Co-founder and CEO

If you’ve seen the movie “Snakes on a Plane,” you’ll recall the scene where Samuel L. Jackson uses some colorful language to describe how fed up he is with all the snakes on that plane. While I’m going to use less colorful language here, substitute “killer robots” for “snakes” and “the conversation about AI” for the word “plane,” and you’ll understand how I feel. Even if you haven’t seen it, you get the idea.


In recent weeks, Elon Musk has once again sounded the alarm and drawn others into the discussion about the dangers of AI. While I have great respect for Musk, I feel he is actually doing the industry a disservice. I get why he’s doing it. It’s provocative. It’s the stuff that nightmares are made of – autonomous AI beings that suddenly realize they don’t need humans anymore. This concept has propelled Hollywood fantasies for decades – from the original Westworld movie (1973) to the Terminator series, to… well the new Westworld. While the questions Musk raises are valid, the way he’s doing it leads to confusion.


The problem is that when a high-powered tech exec tells a room full of un-tech-sophisticated elected officials that this is a real possibility, it has the potential to halt years of progress. Musk has called for establishing new government regulations that would force companies to slow down their work on artificial intelligence technologies. “When the regulator is convinced it’s safe to proceed then you can go, but otherwise slow down,” he said. That’s the stupidest thing I’ve ever heard.


I understand that many people have tried to educate Musk on the real vs. imaginary dangers of AI, but it doesn’t seem to be working. As such, it’s critical that the rest of us business and tech leaders do our best to educate others. Thankfully, reporters such as @TomSimonite at WIRED are helping to do this by setting the record straight on stories such as Facebook’s bots going rogue and creating their own language.


Chatbot technology, particularly in the domain of customer service, is where AI can be practically applied today, and artificially intelligent chatbots hold tremendous promise to improve our daily lives. I’m personally working with several major companies on that right now, which I’ll describe in future articles.


Where AI Stands Today


The fact is, we’re a long way from the future that Musk envisions. Artificial Intelligence has been around for a long time, but it’s becoming more popular now because companies are discovering highly useful business applications for the technology. The awareness is fueled by several things:


  • Big Data - Companies can now make sense of a tremendous amount of data (social, mobile, on the machines themselves). The rapid evolution of the data science discipline in the last couple of years has propelled AI to the forefront.


  • Big Marketing - Facebook, Google, IBM, Microsoft and others are making a lot of noise about it. These companies have tremendous marketing power that drives the industry dialogue.


  • Big Blockbusters - Media are excited because consumers (their readers) can relate to it because they’ve been conditioned by Hollywood, and not just in a bad way. Think about J.A.R.V.I.S. from Iron Man, for example. It’s an intriguing concept that AI can do everything for us.


But alas, there’s good and bad about all this interest. The good news is that we’re seeing a massive amount of investment in AI technology. The bad news is there’s a great lack of clarity out there about both the state of AI (e.g., chatbots) and how companies can deploy them. So let’s get into debunking some of the common misconceptions…



Top Myths Debunked


MYTH #1: AI Eliminates the Need for Humans


  • FACT: AI is Critically Dependent on Humans. There’s no such thing as a system that builds on its own without humans. A simple example to illustrate this would be an airline chatbot. The interface needs to be integrated into multiple systems including flight information, current pricing, and the seat reservation system. Each of those pieces requires integration with a different system, and there’s a lot of heavy lifting, not to mention continuous tuning and optimization. No machine can do that now, and that’s just a simple example.


MYTH #2: One Size AI Fits All.


  • FACT: There are different types of AI. The two main types of AI are General Purpose and Domain Specific. In order to apply AI effectively, companies need domain specific expertise. For example, a customer journey in financial services is going to be different than that for hospitality or travel. An AI system needs to understand financial terminology, and the way that humans interact with financial systems. What you’ll see is domain specific AI that works with general purpose AI (such as IBM Watson). In the next few years, you’ll see standardization of APIs so that a variety of microservices become available.


MYTH #3: Any Company can do AI


  • FACT: Most Companies are in the “Crawl” Stage Today. If you think of the three phases of AI as “crawl, walk, run,” the vast majority of companies are in the “crawl” stage. While many companies advertise themselves as AI companies, their story quickly unravels as people ask questions that aren’t in the system. It’s easy for three guys with a computer science degree to come up with a system that does some assisted learning, but that’s a far cry from companies that employ large teams of data scientists working on complex problems who have achieved semi-supervised learning. If you think of automated tagging for example, there needs to be a system in place when confidence in the tagging is low, where humans can go back and discern intents. This creates continuous optimization and tuning. However, that’s a long way from autonomous machine learning.


MYTH #4: Most chatbots today are powered by AI


  • FACT: Not All Chatbots are Created Equal. Is all software artificially intelligent? No. Think of the airline example I mentioned. Let’s say that later in the conversation you ask a chatbot “is dinner served on that trip?” Most chatbots would have no idea what “that trip” is. A conversational chatbot will, but that requires a system that knows who you are, validate that you’re in the system, pulls your reservation, and checks to see if seats are available. That’s much harder than it sounds.


I’ve just provided you with a few examples to demonstrate how hard AI really is today. If you can appreciate how hard it is to make AI work in these examples, you can appreciate how difficult it is to create an environment in which unsupervised machine learning takes place, without human involvement. Once you appreciate how difficult these things are, you’ll realize how far away we are from needing to worry about “killer robots.”


For now, let’s resume the conversation on how AI can make our lives better.