Nanalyze

One Artificial Neural Network to Rule Them All

We’re going to talk about the concept of one artificial neural network being superior to the rest, but before that we need to understand just how saturated the artificial intelligence (AI) space is becoming. The great minds over at CB Insights are putting together a list of the top-100 most promising (AI) companies globally and as of the last update, they’ve had +600 startups nominated. That’s an incredibly large number of startups in a relatively new space. Some estimates place the total number of AI startups at well over 1,500. This is starting to get about as fragmented as the cybersecurity space which actually has a top-500 list. That’s when you know things are getting out of hand.

When we talk about AI, what we’re really talking about is an artificial neural network that can learn things just like humans can. An artificial neural network simulates the way that the human brain works using specialized computer chips and a software framework. Evidence seems to point to one company being in the lead at the moment and that’s who we’re going to talk about. That company is Alphabet (NASDAQ:GOOG) or as we will henceforth refer to it as, Google.

There seems to be this notion held by many mainstream tech investing pundits that you can invest in every single emerging technology out there by simply buying shares of Google. This just simply isn’t true and here’s why:

  • 70.3% – Adwords revenues on Google.com, advertising on YouTube, advertising on Google properties (Gmail, Finance, Maps, Play, etc.) 
  • 20.5% – AdSense, AdExchange, AdMob
  • 9.2% – Sales of apps and media content, hardware sales, service fees

That’s where all of Google’s revenues come from today and nothing up there has anything to do with autonomous cars, virtual reality, or the cloud. Google revenues are driven by the fact that they dominate with almost 90% worldwide desktop market share for searches as seen below:

statista-google-search-market-share

Those above numbers reflect the +100 billion searches that Google processes each month and every single one of those searches now uses an artificial neural network to retrieve results better than humans ever could. This all started rather recently when in 2014 Google acquired an AI company called Deepmind. This British startup was being run with a 20-year roadmap, aimed to “solve intelligence”, and was backed by high-profile visionary investors like Elon Musk and Peter Thiel. As part of the terms of the ~$600 million acquisition, Google had to agree to setup an ethics board to make sure Deepmind didn’t harm anyone.

Since then, Deepmind has pretty much transformed Google in every way. As of early 2015, all Google searches are now run through the Deepmind artificial neural network because it is just far better than humans at finding relevant search results. In March of this year, Deepmind beat the world champion of Go, a game so complex that it is hardly fathomable. Remember how chess is said to have as many moves as atoms in the universe? In Go, there are 10^720 possible games for every atom in the universe. What was most remarkable about Deepmind playing Go was that it made moves that humans had never conceived of before. That is truly indicative of artificial intelligence, and it comes as no surprise that this very same intelligence would prove to be immensely valuable for the 15% of all search queries given to Google that have never been seen before as well.

Fast forward to today and now Google’s artificial neural network is used for image search, speech recognition, spam detection, fraud detection, translation, and more. According to a recent article by TechRepublic, deep learning networks have now replaced 60 “handcrafted rule-based systems” at Google. They even managed to turn it loose on their data centers to realize a 15% reduction in energy usage across the board resulting in 10s of millions of dollars in cost savings and a reduction in their carbon footprint. Perhaps what’s most exciting is now anyone can play around with Google’s fascinating artificial neural network.  Just this year, Google decided to open source their artificial neural network and it’s called Tensorflow. If you want to play around with an artificial neural network yourself, then just click the below image:

tensorflow artificial neural network

If you have a decent understanding of the Python programming language, you can start to build your own deep learning applications. People are starting to do some pretty cool stuff with Tensorflow. Look no further than a Japanese cucumber farmer who used it to identify and sort only the best, most delicious thorny cucumbers from his harvest. He actually built his own cucumber sorting machine using readily available technologies coupled with Google’s Tensorflow as seen below:

tensorflow-cucumber-sorting

Just how cool is that? It’s important to note however that an artificial neural network can be built using any one of multiple frameworks aside from Google’s Tensorflow. Here are some of the competing frameworks that developers appear to searching for:

artificial neural network competitors

We’d venture to say that the actual frameworks you use for artificial neural networks are like programming languages in that they each offer relative performance advantages for various applications, yet all can coexist together peacefully without being direct competitors. Will Google’s artificial neural network rule them all? Well, given the fact that Google is a $535 billion company which controls access to the world’s information and they have $78 billion in cash lying around, it’s probably a safe bet to think that Google will own this space. Will other companies use AI this effectively? Probably, given that there are upwards of 1,500 startups working on it. But you can guarantee that Google has a team with some of the world’s brightest AI subject matter experts looking at the underlying technology of every single startup out there that’s developing an artificial neural network. If Google thinks that there’s something out there that’s a threat to Deepmind, you can bet they are going to known about it and either attack it or gobble it up.

It seems like the Deepmind acquisition was an incredible leap forward for Google and might be seen as the most strategic moves they will ever make. Our biggest collective investment in any single stock at the moment is in Google shares and we’ve been holding those since shortly after they first had an IPO. This may be the first time that we’ve thought that maybe we should accumulate even more on dips. We said before that you can’t invest in artificial intelligence yet. We’ve also said that Google isn’t a stock you should invest in for all technologies. But we’re starting to think that Google looks like a pretty compelling investment for investing in artificial neural networks and AI. We can’t wait to see what cool things they decide to do next with this technology.

If you’re looking for other AI platforms that are open source, be sure to check out our article on “A List of 15 Free AI Software Programs to Download“.

If you pay more than $4.95 a trade, you're paying too much. Ally Invest is one of the lowest-fee brokers around so you spend less money on transaction fees and more on stocks. With more than 30 trades a quarter it drops even lower to $3.95 a tradeOpen an account and begin trading today.

Subscribe to the Nanalyze Weekly Digest

Subscribe to our Nanalyze Weekly Digest to receive a summary of all articles every week.

We’ll never use your email for anything other than sending you great articles about investing in disruptive technologies.

  • This field is for validation purposes and should be left unchanged.

27 Shares
Tweet
Share13
Share9
Reddit
+11
Buffer4