Why I quit my job to go democratize AI and machine learning
I realized something wasn’t right with the current usability of machine learning when it was taking month after month to integrate a simple feature. We were using a cloud vendor’s Machine Learning as a Service (MLaaS) APIs to perform some simple face recognition. Even though the APIs were more or less straightforward, the actual use of them required some very complicated workflows on our part. And the more we experimented, the larger our cloud bill got.
When that didn’t work, we hunted around for other machine learning models that we might have more control over. The ones we found were barely compiled code that had no deployability or scalability built in. Furthermore, you couldn’t train them or improve them. We would have to do even more work to get them integrated and deliverable than what we were attempting before with the cloud providers.
It became clear to me that AI needs to go through a round of democratization in order for it to really change the world.
As an entrepreneur in the software space, I know the importance of being able to deliver something quickly. When you’re building an app, platform, web service, new feature (anything really), you need to be able to deliver a product or solution that solves a problem in a relatively short amount of time.
What quickly means to each individual company will differ, and it is important to point out that quickly doesn’t mean compromising on quality. All it means is that whatever solution you’re delivering solves a single problem (not every problem).
Developers today are accomplishing extraordinary things, and they’re doing so in lean, mean, agile ways. One of their secrets is to avoid reinventing the wheel whenever possible. And the reason they can do this is that a lot of über powerful technologies like search engines and messaging queues have become democratized. The complex part of the technology has been abstracted away, usually behind some APIs, and the developer can focus on delivering value, rather than trying to build a search engine from scratch.
So I quit my job to join Machine Box to do the same for machine learning.
Why Democratize Machine Learning?
Machine learning can do some incredible things; help fight fake news, provide more convenient ways to authenticate a device, save time searching for things… the list goes on. We all benefit from these capabilities, but machine learning is still really hard to implement.
As a product owner, business strategist, and a hobbyist developer, I’ve tried building, training, integrating and deploying machine learning myself (or lead dev teams doing the same), and the following is what I’ve learned.
In order to successfully integrate AI into a business without using existing tools, you need specialized people who would know the difference between a K means clustering algorithm and a neural net. They’d have to think about precision and recall, know about the latest research into different machine learning algorithms, and be able to use tools like Keras and Tensorflow.
Assuming they knew that, they’d then need to think about the training data, how to gather it, how to clean it, and how to experiment with it.
Training is the hardest part about machine learning. I’ve spent hundreds of hours frustratingly trying and failing to develop meaningful datasets to train models with (all the while racking up big cloud bills).
Once that’s sorted, you then need to pass it all off to a separate group of specialists who can integrate, deploy and scale the models.
This whole process does not make innovation easy. An enterprise that wants to make it easier for doctor’s to find your relevant medical history or a startup trying to weed out Twitter bots will have to spend significant time and money solving all of these fundamentals.
I want developers to be able to jump right in and solve a problem with powerful tools that abstract away all of the complexity so they don’t have to go back to school and get a masters degree in statistics. I want to make the training set gathering process as smooth as possible by allowing you to go through lots of periods of trial and error without breaking the bank. When we want to integrate powerful search, we use ElasticSearch. When we want to scrape news articles and embed specific elements, we use Embedly. When we want to process credit cards from all over the world, we use Stripe. And now, when we want to use machine learning, we use Machine Box.
Standing on the shoulder of giants
The general progress of technology today is powered by abstraction. We give ourselves more and more powerful tools to solve more and more complex problems. When COBOL was invented in 1952, it abstracting machine code so that we could program in words instead of numbers. Operating systemsabstracted command line interfaces, Oracle abstracted storing lots of data, and so on and so forth.
Abstracting technology helps us build the next revolution of technology, which then in turn gets abstracted to enable the next one.
Machine learning is just part of this logical progression. With tools like Microsoft Cognitive Service, Google Vision, and Amazon Rekognition, developers can start to interact with these powerful capabilities. At Machine Box, we take it a step further by letting you manage and train your own models on your own infrastructure, effectively giving you all the power. And the brilliant thing is that you don’t need to know anything about machine learning. You just need to understand the problem you’re trying to solve (and how to POST to an API).
Solving my own problem
Today, I have customers who were struggling to integrate AI because it was either too expensive, too hard to integrate, or both. They’ve told me that Machine Box is saving them tens of thousands of dollars per month, or that they wouldn’t even be able to get off the ground if it weren’t for our tools. That is what makes quitting my job and diving into the masochistic world of startups worthwhile. It is what gets me up every morning, motivates me to stick to what I’m doing and to keep on pushing for the democratization of AI.
This story is published in The Startup, Medium’s largest entrepreneurship publication followed by 308,471+ people.
Subscribe to receive our top stories here.
