During an AI podcast a couple months ago, someone asked me if AI would be the next big tech bubble to burst. This question led me to think about what AI is today and where it’s headed. What is AI, really? It’s a next-generation network and database tool that’s short for “artificial intelligence.” AI just sounds sexier. In fact, AI today is not really what people think it is. AI theories and algorithms have been around for decades. A very simplified description of this next generation network and database tool is it extracts features, converts them to vectors (numbers) in layers, and stores them for easy recall. To complicate things further, we call these “neural networks,” subliminally implying a brain and some biology – which is misleading.
So, what’s changed over the past decades of its development? What’s all the hype about today? For starters, today’s computer power has increased astronomically with GPUs (graphics processing unit, as opposed to CPU, central processing unit), and companies have created open source frameworks and made them public to developers over the past few years. These improvements have fueled much of the hype we’re seeing today.
AI hype resembles the dotcom bubble
Remember the 1990s when you sent your first email and started surfing the web? Altavista and Yahoo! Were just starting up. People were starting to bet on anything that ended with a dotcom in its name. Companies that didn’t know what they were doing were raising hundreds of millions of dollars while companies like Google had difficulty convincing investors. Technical people were just figuring out how to use TCP/IP and HTML.
Was the internet going to change our lives? Yes, eventually but not as quickly and in the way people thought it would. The dotcom crash was about to happen. Similarly, is AI going to change our lives? Yes. Is it going to happen tomorrow? No. How long will it take? We’re probably five to seven years out before it starts impacting people in a more significant way, and probably ten to fifteen years out before there is a major shift. Now, that’s still not a long time, but it’s a long time for investors, and that appears to be where we are with AI today.
What the industry needs to succeed
AI companies need to deliver working and complete products to the market. These products need to solve serious pain points. It’s not enough to have something that’s cool. They need to be quickly deployable and robust. Unfortunately, vertical components and infrastructure for AI products are just not there for all new companies to succeed. It’s like trying to do a mobile app before 3G or trying to do VOIP 20 years ago (slightly different issues, but similar in terms of constraints).
So why is that? Coding languages have been quite similar to each other as they evolved over the past half a century. There are the instructions – the code. When you run the code, the system reads the database(s) and then writes on a database(s). This is pretty much the same for all languages since the 60s. That’s classical coding very simply put. In deep learning and machine learning, developers need to learn how to utilize new frameworks and it has less to do with classical programming and more to do with getting the different types of regressions and layers correct, and pasting components and tools together.
New generations of developers are now getting accustomed to these new platforms and tools. That means you have a serious bottleneck in talent. Even if you get these AI frameworks to work, they’re not deployable products in themselves. Developers still need to have classical coding skills to develop products from these frameworks, which means you need to have teams of AI people and classical coders working in synchrony.
Training is also a common area for error. Now that you have a framework, you need to train it to perform tasks and make predictions. The problem is, there are numerous constraints surrounding training, from the number of classes that you can train, to data collection, to synchronization of data, and getting the neural nets to work as intended. Then you have issues around predictions and latency. How do you know what the thresholds for predicting should be? How does the model work in context and out of context? What’s the accuracy? And what are the consequences of false positives? The AI doesn’t know any of this. You need to configure it constantly; this takes a lot of time and costs a lot of money. Can you build an iOS app without the proper tools? No. Similarly, in AI, the tools don’t really exist yet. AI developers are still improvising.
Overcoming common misconceptions
When you look at AI companies today, you automatically assume they should be founded by Ph.Ds. or at least have a lot of Ph.Ds. on the team. This is a knee-jerk reaction because you assume there is a lot of improvisation, customization, and discovery going on. But is this sustainable? Are today’s AI companies really R&D companies disguised as commercial companies? Maybe, but where does that leave us with real execution and product-market-fit? We’re still at the very early stages.
Another common misconception is we expect top tech companies to totally dominate the AI industry moving forward. It’s like saying the internet is going to be dominated by mobile phone operators or the large Silicon Valley companies of the 90s. This could not be further from the truth. We’ll see a slew of new companies emerging in different fields of AI over the next several years. Yes, the bigger ones will acquire some of them, but I don’t expect the big names to totally dominate this field.
So will artificial intelligence be the next tech bubble to burst? My answer is no. I think it’s here to stay, but we’ll most definitely see corrections and surprises moving forward.
Emrah Gultekin is the CEO of Chooch Intelligence Technologies, a company that produces AI that codes AI.