AI Is Eating Software
Software will eat the world, but very quickly behind it, Artificial Intelligence will be eating software.
--
“Software is eating the world” is now the tagline of this iconic man’s venture capital firm. And he’s right. Software is set to disrupt everything in our modern lives. Fact is, it already has, in some small ways. Because computers. Because internet. Because mobile. Because…well you get the picture. It’s “Because”, all the way down.
Which brings me to something that I find myself repeating in conversations over and over. To sales people. To the press. To product and dev people. To random people on the street. To my lady, my family. And yes, even to the dog.
AI is eating software.
I know I’ve said this before, but I can’t help but keep repeating it because I’ve only begun to hear the faint drum beat of the academics, researchers, and technologists on the bleeding edge of the future begin to get louder.
At the beginning of 2016, as most folks do, I wrote a Top 10 list of the things that were likely to pick up steam. AI was on it, of course, but it’s already grown so rapidly that I find it pales against everything else in my list. And so I’ve focused on it exclusively to understand all the little minute gears and levers at work, and then how new, larger gears and levers might then attach to that in the future, and of course how people will react to that with their pocket books.
Because someone, somewhere, always have to pay for the product. And it always rolls up to the consumer eventually. So, every business should start with that end in mind, and then work backwards. Because if everyday people don’t pay for it, then a business you do not have.
And so I find myself revisiting my core assumptions over and over again. Checking them. Making sure I haven’t flown off my rocker. I still believe these things hold. Check me on this, though.
- Data is already everywhere in corporations. Internet of Things, more video, and more people getting online globally is only going to compound that problem.
- You need some tool to make sense of it all. That’s really all AI is for at the moment. Instead of Data Scientists combing through millions of spreadsheets worth of data trying to unearth a pattern by luck (“feature engineering”), AI can do that using math to automatically find the things that matter and expose them so you can use them in everyday business (e.g., correlations).
- The cycle time for gaining insights from business operations (and even your own personal health) will turn from months or years to days and eventually seconds, finally resulting in real-time analysis. The holy grail. Continuous optimization.
But we have to get there first.
Most of the stories you read in the media today are about understanding what’s in a picture, or turning your spoken words into text or actions, and then chatting with a bot in a message. Even Tesla and eventually Uber are working on self-driving cars to give us more free time for work and entertainment during our commute.
The end game, as I’ve believed for many years, for our own human species, is to have conversations with people that have since passed on. Much like Superman talks to his father’s hologram in the movies. Of course, there is the philosophical sense of this but also the practical. In that a version of “me” can have simultaneous conversations with outsiders and other versions of themselves to get things done. That is, of course, once I trust my other selves with 99.999% accuracy.
It sounds weird at first, but for very basic tasks, you could see how it would already save you time. Things like, “fill up the car with gas when it gets below a quarter tank” if you haven’t moved to electric yet, doesn’t require a decision. It’s an always yes. So why do you have to still make a decision? Same with groceries. Fact is, you probably buy a similar set of stuff 99% of the time. Meat, veggies, bread, snacks, drinks, some toiletries. Why do you waste hours going to the grocery store or even ordering from Amazon when a machine could do that drudgery for you?
And then we start moving into a bit more complex territory. Small decisions at work. Like, “enter your timesheets on Monday”. Why are we still doing this arcane task across our white collar professionals: law firms, consulting firms, dev shops, repair shops, or anything that requires an hourly rate? Shouldn’t something just track it and bill it automatically for you? And now we’re into accounting and auditing. A machine should be able to read the updates to the tax code every year along with the new FASB rules, then audit your company’s financials and fill out your personal tax return without you making any decisions. Because there isn’t a decision, you’re just trying to understand and go through a process that humans shouldn’t have to go through.
We’re starting to see calendar scheduling Bots that help coordinate a time when we’re both free. This just seems so obvious. (I hate the term bots by the way, they’re so inhuman).
But what if we could go further? What if we could increase the complexity?
For example, a copy of me talks to a copy of you to get things done based on our high level strategic direction. Maybe it’s “Tell me how our Return on Invested Capital compares to our peer group over the last 8 quarters” and then turns into “Figure out a new, novel product that we can develop where there’s market demand without a solution currently? Base it on Google Ad rates and search volume.”
You could see how Google would enjoy something like this. It helps get to knowledge never before available by offering more analytical capabilities. And that’s their mission since their founding. Obviously this is for work and most people would be okay outsourcing that to their own AI and then seeing what comes back. What you’re less likely to see are people outsourcing their social and entertainment lives to an AI. Because people generally enjoy that.
So, could the end game be that we just become hedonists? Living life? What if a dog’s life is the ultimate goal of humanity. Sleep. Give love. Eat. Enjoy the company of others. And sometimes go potty.
But, if we want to understand how we will get from here to there, we need to start with the base foundational layers:
- Energy.
- Computations.
- Data.
- Capital.
- Devices.
- Usage.
Let’s start with energy. You need a place with lots of cheap power, either solar or electric. Google builds its data centers in Iowa because MidAmerican Energy (some of my family used to work there, we were born and raised in Iowa) has very cheap electricity and the state has a very low cost of living and labor (about 85% of the national average, where San Francisco is about 200%). Elon is building his Gigafactory for battery tech in Nevada, where the land is vast and the sun is hot.
Now you need to build massive structures with lots and lots of servers. Go ahead and purchase a bulk order of NVIDIA’s DGX-1, their Deep Learning Supercomputer in a Box for $129,000 a pop. Wire it all together and add a layer of software over the top so any researcher, academic and developer can use it to train their own models, then run them in production. They will do this because it cuts their wait time for these simulations to run from 1 week to 2 hours.
There’s Google’s Machine Learning service on their Cloud Engine, using their custom-built Tensorflow framework. This is much more advanced than AWS’s current machine learning offering. And Facebook doesn’t open their platform up to further development. So those are your options today, but those will expand greatly.
If I’m a software developer or researcher, I’m looking seriously at Google, DeepMind, and Tensorflow of where to start, instead of continuing on with, for example, the Torch framework. So now, we have power, GPUs for computation, frameworks for working with architectures and data and a cloud to run it all via a few API calls. The next hardest step, once you have capital generated by your own business, is getting data. Lots and lots of data.
In order to get to 99.999% accuracy, which is five 9s of AI (compared to five 9s of reliability for software uptime), you need 1 trillion points of data. One trillion. There are a little over 7 billion people on the planet. That means every single person on the planet needs to perform some action 150 times to generate 1 trillion data points. If they do one thing every day (the definition of a Daily Active User in startup parlance), that means you need almost half a year of data before you get to that accuracy.
Lets start with the platform that has the most active users on a daily basis. That’s both Facebook and Google. They have about 1 billion active users performing some core action every day (Apple has 1 billion active devices over 90 days to put that into comparison). So, 1 trillion divided by 1 billion equals 1,000 actions per person. 1,000 actions / 365 days in a year = almost 3 years.
Think about that. It takes 3 years from this very moment before Facebook or Google gets to 99.999% accuracy with their predictions of what you want when you search for something or scroll through your news feed. That’s not even taking into account spoken words, intent, or asking for really complex requests that Siri is likely to get.
3 years ago, AI wasn’t even a blip on the startup radar. It was all about photosharing. Video wasn’t even that hot back then, especially live streaming. That only happened in the last 12 months.
So, by 2020 Facebook and Google will be able to give you what you want based on some really basic queries almost every single time.
And that’s just step 1 in artificial intelligence.
From 2020 to 2030, the rest of the layers will need to be built up. Software will eat the world, but very quickly behind it, Artificial Intelligence will be eating software.
— Sean Everett, VP Business Development, Strategic Systems International
If you enjoyed this read, hit “Recommend” to spread this post to more people. Your comments, suggestions and feedback will be greatly appreciated.
Strategic Systems International (SSI) is an advanced analytics and software engineering firm headquartered in Chicago with 25+ year experience in building applications for enterprises and SAAS companies with an onshore/offshore delivery model. We are a team of data scientists and technologists that seek to solve complex problems through simple technology and data enabled solutions.
Visit Our Website: ssidecisions.com
