Think Twice About a Company When You Keep Hearing These 4 AI Buzzwords
For every business that's using AI to do something groundbreaking, there are more that, well, aren't. Adobe’s 2018 "Digital Trends" report found that while only 15 percent of companies are currently using AI, 31 percent have it on the agenda for the next year -- but that rising demand doesn’t necessarily correlate with a rise in high-quality AI products. Much of it is fluff.
I learned this the hard way in 2016. We were looking to license a model for a very specific task, and it led to horror. We're an AI company ourselves, but we focus exclusively on language understanding and NLP. One of our clients was looking to add image recognition to one of our big AI models, so we started looking for partners that were good at image recognition. And it was really hard. We ended up settling on a vendor that our client recommended.On paper, the company seemed good. The sales team showed us case studies, and I'd heard the names of the data scientists mentioned in Slack channels about machine learning. But when we actually plugged into the AI, the results were wholly unsatisfying. We should have spent more time doing due diligence and speaking with the machine learning team (or at least an informed executive). When we tried to address the problem with the sales team it became clear they had no idea what the tech did. This separation between the sales team and the engineering team is a huge problem with AI -- the people selling the solutions have to understand how the AI actually works.
Because AI technology is getting so much hype, many companies are pivoting to AI without any experience in the field. And when you combine the complexity of the technology with the number of people getting in on the game, it's a recipe for disaster. It’s hard to validate what good performance looks like; there aren't that many widely established benchmarks for AI confidence or accuracy across different verticals, and some companies are taking advantage of the confusion. And AI is complicated enough without having to sort out bandwagon hustlers.
AI's turbulent hype cycle
AI has huge potential -- its market value is expected to reach more than $190 billion by 2025, according to Research and Markets -- which has given rise to a frothy hype cycle the likes of which the tech world hasn't seen. Some newbies to the the industry would have you believe they can solve all your problems with an off-the-shelf recurring neural network when, in reality, AI works best when it's hyperspecialized.
Discerning between real and fake means digging deeper for a closer look at the implementation and engineering teams. There are differences between data scientists, QA testers, linguists, data engineers and machine learning engineers. Figuring out whether these people truly know AI or are just trying to cash in is imperative -- it will save you heartache down the road. Being on the lookout for these buzzwords should help you distinguish between the hype and the real.
Anyone who says he or she created some AI-powered sentient thing that will magically solve every problem in the world is lying. AI and machine learning models work well when they're focused on specific tasks. We're not in the "Age of Ultron." Like the "geniuses" at Apple’s Genius Bar, AI can only do what it’s trained to do.
The fantasy of a "genius" AI is a marketing tactic that actually obscures how valuable AI can really be. When a model is supposed to do it all and actually falls flat, people trust AI less. Even tech giants with well-paid AI talent have trouble when they go too far too fast. In 2016, for example, a self-driving Uber powered by image recognition AI ran red lights in San Francisco.
Deep learning is powerful, but many companies throw the term around without warrant. There's an industry joke about this: "Deep learning is like sex in high school. Everyone says they’re doing it, but only a few actually are, and they're probably doing it poorly."
Fake AI companies love putting “deep learning” on their homepage because it makes it seem like their AI is "super deep, man." Deep learning can, in fact, be a powerful tool, but sometimes it just doesn't make sense for the problem at hand. Taking cooking, for example. When a scientist tried teaching a deep learning model to put together recipes, the results were hilariously bad.
If a company says it's using deep learning on problems where there aren't extremely large data sets, it's probably faking it. Deep learning can change your life, but it can also waste your time if you use a deep neural network just for the hype factor. There are companies using deep learning to write speeches even though it doesn’t work.
Natural language understanding
NLU is, simply put, AI's ability to process and comprehend language and carry out a command based on that language. NLU can be a necessary part of business, and the technology has come a long way in just a few years. Consider the difference between Siri stumbling out of the gate before straightening things out. The capacity for AI to process an idea relayed from a person and then attach a response to it that solves the person's problem is mind-boggling. But you still have to beware of the charlatans.
A lot of consultants seem to think they've magically encapsulated AI expertise, and they throw around NLU to describe their work, but they’re not doing NLU unless they’re actually training the model, not just adding a service layer on top. This can be helpful in some instances, but the term itself doesn't always mean what it seems. To determine whether a vendor is legit, ask what approach it used to tune its training data.
Non-tech people try to latch onto the AI trend by advertising "augmented intelligence," as in "we will augment your intelligence with AI" -- but AI already does that out of the box. That's the idea. Adding a middleman that doesn't actively touch the algorithms can add a lot of trash in an industry that's supposed to eliminate waste.
Steer away from companies that sold digital stickers 24 months ago but call themselves AI experts today. Many of these companies are hiring "machine learning specialists," a job title that didn’t exist even six months ago. Machine learning engineering job listings have grown nearly tenfold since 2012, according to LinkedIn’s 2017 "U.S. Emerging Jobs Report." It’s great to see growth in the industry, but can all those engineers write algorithms and train models? If not, be cautious.
There are plenty of AI companies doing groundbreaking things that may very well change lives for the better. But every success story spawns a dozen new "know-it-alls." Make sure you ask the hard questions to ensure you’re talking to the right people -- or at least enjoy laughing at the plethora of huckster pitch decks you'll now be able to identify.