Change Is Hard

I haven’t made a blog posting in over 2 years because I haven’t really had anything interesting to write about. Or at least nothing I could show people, until now. A few weeks ago, I launched a new site, which is the culmination of many of the different things I have been working on: BoredHumans.com.

It is basically an AI (artificial intelligence) version of my old Bored.com site, which was filled with fun stuff to do when you were bored. I will get into details about the new site in a moment, but first, let me give you some background.

When I first started my internet business in 1995, for the first few years, I did a little of everything. I created web pages for businesses for $50 each. I built an online mall for inventors and another one for network marketers. I registered hundreds of dot com domain names. The whole industry was so new it was very hard to tell if any of it would ever generate significant income. I only made around $1000/month for the first 5 years, and then all of a sudden, some of the things took off. Banner ad networks and affiliate programs became available, and my sites that were earning zero now finally had a way to bring in money. And at the same time, domains that cost me $35/year just to sit there, started getting some offers (I think my first really big sale was Cookies.com for $5,000).

Back then, it was easy to get traffic from the search engines. If you had a good website, they would find it, and people would come to it. One of my sites, Bored.com, was getting so much traffic (70,000 visitors a day) that it became a whole business in itself, and I sold it in 2008. The main reason I did that was I worried my old-style sites would soon be left behind in the coming age of mobile phones, apps, and Web 2.0 (Facebook, YouTube, and other user-generated content).

So I went back to doing more of the same routine. Cranking out new web pages and trying to sell my domains. But I would spend weeks or months creating a great site, and it would get almost no visitors. On top of that, the money I was making from parking my 6,000 domains (having  ads on them) started to dry up, and I was left paying the $60,000/year in renewal fees. I finally made the decision to stop creating new sites and also sold all my unused domains in a bulk deal. I still had several hundred small sites and low-quality domains left, but it wasn’t much of a business any more. I needed to do something new.

I had dabbled in using artificial intelligence on a few of my sites, but it was not something I really understood, I just paid a programmer to do what I needed. Over time though, I started to see the potential in it, and shifted my focus away from my existing business to doing AI projects. The problem was though, none of these projects had much potential to make money, they were more for learning. And, aside from not knowing what I was doing, I was also not a programmer, so I had to pay people to do the programming for me (for these things that probably would never make money). But, at least I was doing something in an industry about to explode, much like the internet did 25 years ago.

Change is hard, though. With the web, even a 7-year-old kid can create a site or buy a domain, even set up an online store. With AI, I was competing with double Ph.D. computer geniuses who were publishing papers in academic journals. But, you have to start somewhere, so over a 3 year period I worked on all sorts of machine learning projects (the computer part of AI, known as ML), such as:

• Entering Kaggle.com contests – You can win $25,000+ if you solve machine learning problems better than other entrants. Examples include detecting fake videos and predicting the sales of items in Walmart. This, at least, is a concrete way to make money from ML without actually getting a job doing it. But the competition is intense. I did manage to finish in the top 10% in one contest, but that was not good enough to get a prize.
• Predicting The Stock Market  – Like Kaggle.com, Numer.ai offers contests where you can win money, specifically for stock predictions. I entered those and also worked on my own automated systems for trading cryptocurrencies (such as Bitcoin) and FX (foreign currencies). A person could spend their whole life just doing this. It is a rabbit hole that will suck all of your time, energy, and money. But fascinating, and the hope of beating the market is something that will never leave you.
• AutoML – Automated Machine Learning. Kind of like how typing something or doing financials was hard before Excel and Word; that is what machine learning is like now. Data scientists spend huge amounts of time manually doing all sorts of things that could be automated but aren’t. One of my goals was to create an AutoML program to use to win Kaggle contests. I did build several such programs, but none were good enough to give me a significant advantage.
• Self-Driving Cars – Much of the technology for this is similar to having a computer play a driving video game, so it is something anybody can mess around with, without the need of using a real car for it. I felt like Elon Musk for a few weeks.
• Natural Language Processing – Anything involving words, such as translating Spanish to English, search engines, speech recognition, and sentiment analysis.
• Reinforcement Learning – This is teaching the computer to do things like control robots, play chess, or navigate a maze. All of this has applications in the real world, in terms of building an AI that can actually do something, such as assembling products in a factory, drug discovery (finding cures for diseases), or flying a drone.
• Obscure Areas of AI – Genetic algorithms, neural architecture search, genetic programming, swarm optimization, and more.

After 3 years of doing this and not making any money, which came after a long period of creating new websites that did not make money, I was under big financial pressure to do something that would actually generate income. And, to make matters worse, I didn’t have anything I could show to people to explain what I had been up to all this time. Unlike websites, which are out there for the whole world to see, machine learning projects are usually in Python on your local PC and don’t work on the web without a lot of special programming to make that happen.

I made a list of possible ventures, including ideas such as starting a company to: offer AI consulting (having other people do the work and I would be manage it), invent AutoML products, or make AI music (this is a hot industry).  But I really did not want to do anything that involved having customers. I instead wanted to do a business that had both huge money-making potential and would be something everybody could use and appreciate. So based on a project I had already been working on, I decided my best hope was to create a search engine on my old FindInfo.com domain, which I was not currently using. I knew I couldn’t do it better than Google, but there are only a few other search engines out there, so maybe there was room for one more. Search is a 90 billion dollar industry, and my hope was that I could eventually build up my site to the level where I could sell it for truckloads of money.

I chose to give it an AI feel by making it like Siri, where you can chat with it and do a search. Just building the chatbot part was a huge project. Then I spent many months trying to have the search engine answer questions directly, with no search results, like if you asked it “What is the capital of France?”, instead of giving you search results, it would give you the answer (Paris). After a lot of trial and error, I got it up to 80% accuracy, but that is not really good enough to be of much use, so I scrapped that plan and instead changed it to use regular search results. That was a massive project too because I had to gather data from hundreds of millions of websites and index them like Google does, to show only the most relevant results. All of this is extremely high-level stuff involving gigantic databases using technologies most data scientists have never even heard of (I didn’t do all of it myself, I figured out all the details and paid other programmers to do much of it).

Next, I had to figure out some ways to get people to go to my FindInfo.com search engine, so I came up with the idea of adding a machine learning showcase to it, as my search engine was powered by AI. There are not a lot of demos of ML online, so I thought that might get some traffic. I started by adding a poetry generator, a song lyrics generator, and a few other things like that. But once I got going with it, I realized there was a lot more I could add, and instead, I should create a separate site to put it all on, and focus all my efforts on this new site and not FindInfo.com. So that is what I did. I spent an hour searching for available domains that would sound fun, yet also make it sound like an AI site. Eventually, I registered BoredHumans.com for it. I then copied the site design from FindInfo.com, paid somebody on Fiverr.com $5 to create a logo for me, and my site was launched. I also removed the chatbot from FindInfo.com and put it on BoredHumans.com instead.

So now I am back where I started 20 years ago, with a site like Bored.com, but with a modern twist. And I finally feel at home again. Trying to beat Google or be a top data scientist are things I wish I could do but are not very realistic. Instead, for example, today I started work on creating a site for BoredHumans.com where you can convert any web page to heavy metal song lyrics. And after that, I plan to make a version that turns any page into a Donald Trump speech.

Everything is different now, yet oddly the same. And I am fine with that.