The First

State of the Industry

and the amazing story that transpired

Having returned from the Revit Technology conference I was energized. At the conference something amazing had happened. The appropriately named conference centered around Autodesk's Revit (at that time) had a fundamental shift that was obvious to everyone there. That year Dynamo a now trending tool for visual programming Revit had dominated the conference. Jokingly I mentioned it should have been called the Dynamo Technology conference that year.

Upon arriving back in Orlando I set out to join this new way of harnessing the tool I thought I had a good understanding of only to find I was in my infancy of understanding Revit. Having previously Hosted the Orlando Revit users group I knew as I did back when I had little understanding of Revit, I needed a network of like mined professional around me to explore these new tools more deeply. Not looking to derail the existing user group and with a desire to focus on Dynamo I began Design Technology Orlando. The named stemmed from already seeing how naming a group for a software doomed to be supplanted was not a good idea, choose to look for a more macro name that could encompass not just this technology but future ones as well.

dynamo led to the typical initial questions. If i can do this with simple visual programming what if we used the primary programming languages. This intern led to more research and a introduction to the possibilities of machine learning. The topic of machine learning fascinated me. So as anyone at the time would do, You-tubed everything i could on the subject and tried to understand how to do machine learning. It how ever was above my head. Not giving up I looked to Linked-In to find someone willing to tell me how it works in trade for lunch.

Enter Kenneth Stanley of UCF. after a single lunch conversation I had a much better understanding. It did not end there. Kenneth also walked me through his research and the concept of his book. Wow! Stunned I set out to share this amazing information at the new Design Technology Orlando and find other like minded professionals wanting to explore these types of at the time advanced tools.

The first meeting was a State of the Industry so to speak helping everyone understand the possibilities and inspire action. I got both Kenneth Stanley and Anthony Hauck then of Autodesk to present on these amazing new tools and ideas. Kenneth lead the discussion with his research and used the tool picbreeder.org to demonstrate his ideas that programming goals into an application is flawed and that an evolving framework is more likely to provide improvement and original achievements.

Next Anthony showed Autodesk's Project Fractal. Fractal used Dynamo recipes (algorithms) with constrained range based variables to automatically generate a large number of iterations for all the variables and in several combinations of the variables. This results in a many times factor of output possibilities. On the left below the image shows an exploration of cladding panels on a building in many different design iterations generated automatically from the Dynamo recipe. In the image you can see that a single upper right result is selected. just above it on screen it the graph showing the combinations of variables and range values that led to the result. In the image below to the right you see a floor plan exploration of the building's program for space allocation running multiple layouts in a search for the optimized circulation path.

The meeting was hard to measure. I often take attendees silence as a result of disinterest but have been told that sometimes you are so shocked that people need time to process what they have seen before they can express their thoughts. We continued to have meetings about the use of Dynamo. While a small group, Deron Edge and myself we had a new member Stephen Smith with experience in Dynamo and was a pivotal addition to the group probably securing its existence for some time to come.

But i would like to demonstrate that the first sign that the group was really on to something was almost a year later. I was in my truck a red light and a notification for my news feed popped up as I glanced at my phone. Uber acquires UCF AI startup. Intrigued I read the short article and saw Kenneth Stanley's name. What was amazing was the amount the startup he help found was worth thus lending strength to his ideas being worth some investigation. while I could not find the articcle from that day at the stop light, I copied one from UCF below.

For Ken Stanley, an associate professor of computer science at the University of Central Florida, the past two years have been a whirlwind that’s landed him in the international spotlight.

First, in 2015, he co-founded Geometric Intelligence, a startup designed to be a unique, research-oriented, private-sector lab that would focus on the cutting edge of artificial intelligence and machine learning.

Then, last December, Geometric Intelligence was acquired by Uber, the ride-sharing giant whose disruptive technology has made it the highest-valued private tech company in the world, with a valuation around $70 billion.

Uber was drawn to Geometric Intelligence because of its approach of focusing on multiple avenues of artificial intelligence, the chance of adding a ready-made lab to its stable, and the work of Stanley and his co-founders: Gary Marcus, a renowned scientist and researcher from New York University; Zoubin Ghahramani, a leader in the field of machine learning and a professor at the University of Cambridge; and Doug Bemis, a veteran of startups with a Ph.D. in neurolinguistics from NYU.

Now, Stanley’s days are spent at Uber headquarters in San Francisco integrating what is now known as Uber AI Labs, while also juggling advising his Ph.D. students and overseeing his lab at UCF.

We talked with him about how it happened, the future of AI, and what it means for UCF.

First off, will your work with Uber impact UCF?

I think this is overall a really good outcome for UCF. There are several reasons for that. First, when large technology companies like a Google or a Facebook or a Microsoft move into a new space, usually there are headlines about it and you’ll see the usual suspect universities in those headlines – something like Stanford or MIT. What this has done is put UCF in those headlines and shown that UCF plays in that league. Another impact is that UCF now has new job opportunities in Silicon Valley for Ph.D. students in particular, but also for all students. Here at Uber AI Labs we already employ three UCF alumni [Joel Lehman, Sebastian Risi and Paul Szerlip]. So we have scientists from UCF who are now working in one of the top research labs in the world – and the opportunities will continue because people will see that UCF has produced this kind of expertise. UCF also now has an opportunity for partnerships with Uber. Finally, on a personal note I’d add that my own experience in industry from this acquisition at a company revolutionizing transportation around the world has given me valuable perspective that improves my ability to connect my work as a professor to the opportunities and challenges that students will face in the future.

What makes your lab different?

We brought together people who are from areas that usually don’t intersect with one another. I come from an area called neuroevolution, which is a combination of artificial neural networks and evolutionary computation. Gary Marcus is a prominent psychologist. Zoubin Ghahramani is known for work with Bayesian learning. And Doug Bemis is a neurolinguistics Ph.D. Gary had a vision that by combining areas that don’t normally practice together, something new would come out. Right now, in the field of machine learning the overwhelmingly dominant idea is deep learning, but many of the ideas in the area are beginning to converge to a similar set of research directions. We had a vision that we would try to push current trends in completely different directions than they’re going right now – in effect not as much following the most obvious path that the community seems to be going.

What does Uber gain from acquiring the startup?

Uber saw there was an opportunity to compete with the top tech companies in the world in the space of machine learning and the value of technology we were developing. In the technology industry, artificial intelligence and machine learning are becoming so important to every business that even businesses that are not direct competitors are still a threat if they have superior technology in this area. Uber has very good machine learning practitioners on their staff. But they understood and had the foresight to see that they needed something that will push the cutting edge to go somewhere different and be able to anticipate the future. A company like ours fit well in filling that gap to push in new directions.

Was Uber’s interest solely about self-driving cars or is it more than that?

That is obviously something they’re looking at. But if you think of Uber, they have innumerable AI-type applications. Their entire system – the way that they dispatch vehicles so you can get a ride, the way that they anticipate where things will be when you need them and try to optimize all of that information – it’s a massive machine learning problem. There are opportunities for AI all throughout their business.

What are your days like now?

I’m spending a lot of time helping to create this lab from the ground up. This doesn’t happen that often, especially in industry. And Uber is a huge organization, and the lab itself has resources and funding to expand, which is again very unusual. We’re trying to decide how this whole thing will work – what are our customs, our incentives to inspire people, what are the things we want to accomplish. For example, we want to engage with the academic community, we don’t want to be isolated. We have to come up with all these new ways of doing business that weren’t formerly part of Uber. Then I spend a lot of time advising my UCF students. So my time is really tight.

Humans behind the wheel are able to pick up on cues like a wave from another driver to go first at a stop sign. How far are we away from when self-driving cars will be able to figure that out?

That is a tough question, and there are a lot of really tough problems in self-driving cars. There are a lot of differing opinions from experts and no one knows for sure. But what I would say is that I think you will see it trickle into practice rather than happening all at once. You might see cars doing one type of driving, like daytime highway driving, and then it will grow into other areas. I think you’ll see a gradual expansion that at first you may not even notice. Some of the most complex types of applications you probably won’t see for many years.

Article By Mark Schlueb

Monday, July 3, 2017