Computers that Code Themselves Using AI
What many tech founders who built the incredible digital infrastructure you see today would never openly admit is that they were not born into a world which treated them very well. They didn’t have the best of upbringings because they grew up in an era where an interest in computers was a stigma, and something that could (and often times did) get their asses kicked. All the nerds (many of whom weren’t exactly hitting it out of the park with the ladies) would sit around in little groups of poorly-dressed abject souls, slightly comforted by the fact that they were smarter than their aggressors. HPV wasn’t something they got from one of their girlfriends, but instead something they did with the lads. You see, HPV was part of a much bigger acronym – HPVAC:
Nowadays, most of that stuff will get you thrown in jail, or at the very least on someone’s watch list, but back then this was a sub-culture that acted as a springboard for some of today’s brightest minds in technology. In order to be part of this motley crew, you generally needed to be a social outcast in some way, but that wasn’t the most important qualification. An absolute requirement was that you had to be able to sling some code. You needed to be as comfortable on a command line as you were building a beige box (look it up). Nowadays, there are coding camps where we all hold hands and learn to “become coders”, and everyone is clamoring for a piece of the action making coding one of our 9 jobs of the future. However, we all seem to be ignoring the elephant in the room here. The monster we’ve awakened called artificial intelligence is learning to code.
Before we delve into this notion of computers that can code themselves, we need to consider that traditional coding is becoming less and less useful. Take as an example the latest flavor of Google’s AlphaGo algorithm. It was told what the rules of a game of Go looked like, and then it was told not to lose. The algorithm then created its own method of playing Go as opposed to being “trained” by watching other people play. Think about what an amazing revelation that is and then ask yourself if you think that these algorithms are going to sit down and start writing IF/THEN/ELSE statements like a bunch of primitive chimps. Probably not. The way that your brain works isn’t something that you could create using today’s best coding languages, and neither will AI. With that said, there are some startups out there that are developing technology that can write and rewrite itself. One of these is a startup called Gamalon.
Founded in 2013, Massachusetts startup Gamalon has taken in around $12 million in funding to develop a system that learns “orders of magnitude faster” and with “orders of magnitude less training data” compared to the more traditional machine learning algorithms of today. They’ve taken in the most DARPA funding of any company in the past 3 years to develop their technology which is referred to as Bayesian Program Synthesis (BPS). It’s when machines start to have ideas, which is why they also call their technology “idea learning”.
98% of big data was created in the past two years, which helps to explain why 80% of it is unstructured. In other words, this data doesn’t live in beautifully structured databases like the one seen below:
This is the problem that Gamalon is setting out to solve. If you think about a big collection of academic journals, it’s just blobs of text that need some structure for a computer to make sense of them. Almost every digital system out there needs to consume structured data. This is where Gamalon comes into play, and the following example best explains how powerful their technology is.
One of their customers was spending $4 million a year on a team of 50 people to do “human-in-the-loop” training for AI algorithms to process unstructured data. After 9 months, they achieved an acceptable level of accuracy. Gamalon took that exact same data set and in just a few minutes provided results that were twice as accurate. That’s what we mean when we talk about machines that have ideas.
So how do they do it? Well, you’ll need to have a superior knowledge of math to really understand it but the key takeaway is this. That team of 50 people that their client was using was made up of 20 people from Atlanta and 30 John-in-Mumbai types from Bangalore. If that company had access to Gamalon’s “software that teaches itself” then they would have not needed that team of 50 people which equates to not needing 37.5 man-years of work which is about one employee’s lifetime. That algorithm just replaced one engineer in minutes and that’s just one example. The fact that the computer does something so exponentially well that it renders a team of coders useless is the same net effect as a computer that can code itself. The end result is that we don’t need so many coders.
According to an article by TechCrunch earlier this year, Gamalon already has a few solutions available for customers on Amazon’s cloud and it’s hard to see how they wouldn’t already be doing some incredible things for their customers which translates into cash. The small startup seems to have wrapped up their DARPA contract as of last month, one for which they beat out 78 different competitors to land:
The man behind the company, Ben Vigoda, has a track record of entrepreneurial success and an incredible resume. His last company, Lyric Semiconductor, was the first microprocessor architecture for statistical machine learning, one that was acquired by Analog Devices. If you want to feel like a total slacker, go check out his bio and see just how productive a person can be in this society if they don’t spend their whole day on social media complaining about how they can’t get anywhere in life. Note the 120 patents that Mr. Vigoda has authored, which means he probably has a pretty good handle on securing the intellectual property behind Gamalon’s technology. He’s probably in his late 30s to mid 40s, which means he remembers the era when technology wasn’t very sexy, and neither were the people who dabbled in it:
Fast forward to today and all those nerds are probably feeling pretty good about themselves with loads of cash and their pick of the ladies (or boys as it were). Let’s hope that today’s youth can follow that same trajectory, something that’s a great deal less risky to do today than it was back in the day. While AI might do a better job at coding than today’s youth, at least it won’t drag them behind the gym and beat the isht out of them.
Here at Nanalyze, we invest the majority of our savings in 30 dividend-paying stocks that increase our income every year and outperform during both bull and bear markets. Find out which ones in the Quantigence report freely available to Nanalyze subscribers.