Machines Who Think

The practical birth of A.I. dates back to the 1950s, when Frank Rosenblatt developed the Perceptron algorithm. Interestingly, while it was initially conceived as software for the IBM 704 computer, the “productized” implementation was a hardware solution called the Mark 1 Perceptron. Even back then, the best experience was a complete one: hardware and software combined for a specific task; in this case, image classification. A version of that algorithm is still around today – in fact, Shelby uses a Multi-Layer Perceptron for its chat interface.

The point is, there are a countable number of steps, over relatively recent history, in the field of machine learning. Most of what is in-use today is derived from what we knew yesterday. If you narrow that field to manufacturing, the milestones are even more sparse. If you narrow it again to algorithmic learning that runs within the operation, Sherlock is virtually peerless. The release of this product is a markedly significant point in the history of artificially intelligence…

The above is an excerpt from the email I sent my team upon completion of our latest 1.0. I included it because I didn’t think I could write anything better to mark this spot in time. I’ve introduced you to Shelby in the past, and while Shelby observes an operation, its newest sibling, Sherlock, actually learns from it.

I didn’t invent Sherlock (nor was I the only inventor on Shelby!) – in this case, the product is deeply indebted to the research and development of folks much, much smarter than me. But I did lead the effort to productize it, and I’m proud that I got a part in bringing it to life. Launching a 1.0 product in manufacturing is act of sheer willpower; once again, I got lucky to have a small core team of people who believed in an idea enough to pour some of themselves into it with me.

This release was step two in a 3-step strategy I helped put together almost 4 years ago: Device -> System -> Enterprise — our plan to make sense of the data in a manufacturing environment in an automatic fashion. I signed up for step 1, committed to step 2 after-the-fact because of the great partnership, and watched someone else make a total mess of step 3.

Most of the people who worked on that strategy with me have given up, or moved on (although one of them recently came back!) and the leadership that originally endorsed it lost focus, or position, or faith… it’s been a long and bruising haul getting to this point — and often a lonely one.

As proud as I am of what we’ve built, I am also very, very tired…

2 thoughts on “Machines Who Think

Leave a Reply

Your email address will not be published. Required fields are marked *