Publish, Share, Monetize Machine Learning APIs
06 Oct 2017
I’ve been playing with Tensor Flow for over a year now, specifically when it comes to working with images and video, but it has been something that has helped me understand what things looks like behind the algorithmic curtain that seems to be part of a growing number of tech marketing strategies right now. Part of this learning is exploring beyond Google’s approach, who is behind Tensor Flow, and understand what is going on at AWS, as well as Azure. I’m stil getting my feet wet learning about what Microsoft is up to with their platform, but I did notice one aspect of the Azure Machine Learning Studio emphasized developers to, “publish, share, monetize” their ML models. While I’m sure there will be a lot of useless vapor ware being sold within this realm, I’m simply seeing it as the next step in API monetization, and specifically the algorithmic evolution of being an API provider.
As the label says in the three ML models for sale in the picture, this is all experimental. Nobody knows what will actually work, or even what the market will bear. However, this is something APIs, and the business of APIs excel at. Making a digital resource available to consumers in a retail, or even wholesale way via marketplaces like Azure and AWS, then playing around with features, pricing, and other elements, until you find the sweet spot. This is how Amazon figured out the whole cloud computing game, and became the leader. It is how Twilio, Stipe and other API as a product companies figured out what developers needed, and what these markets would bear. This will play out in marketplaces like Azure and Google, as well as startup players like Algorithmia–which is where I’ve been cutting my teeth, and learning about ML.
The challenge for ML API entrepreneurs will be helping consumers understand what their models do, or do not do. I see it as an opportunity, because there will be endless amounts of vapor ware, ML voodoo, and smoke and mirrors trying to trick consumers into buying something, as well as endless traps when it comes to keeping them locked in. If you are actually doing something interesting with ML, and it actually provides value in the business world, and you provide clear, concise, no BS language about what it does–you are going to do well. The challenge for you will be getting found in the mountains of crap that is emerging, and differentiating yourself from the smoke and mirrors that we are already seeing so much of. Another challenge you’ll face is navigating the vendor platform course set up by AWS, Google, and Azure as they battle it out for dominance–a game that many of us little guys will have very little power to change or steer.
It is a game that I will keep a close eye on. I’m even pondering publishing a handful of image manipulation models I’ve been working on. IDK. I do not think they are quite ready, and I’m not even entirely sure they are something I want widely used. I’m kind of enjoying using them in my own work, providing me with images I can use in my storytelling. I don’t think the ROI is there yet in the ML API game, and I’ll probably just keep being a bystander, and analyst on the sideline until I see just the right opportunity, or develop just the right model I think will stand out. After seven years of doing API Evangelist I’m pretty good at seeing through the BS, and I’m thinking this experience is going to come in handy in this algorithmic evolution of the API universe, where the magic of AI and ML put so many people under their spell.