Allvision’s CTO Ryan Frenz recently participated in the  Pittsburgh Technology Council’s Beyond Big Data AI/ML Summit. In a panel discussion titled “Using Data and Artificial Intelligence (incl. Machine Learning) to Solve Big Problems,” Ryan and the panel covered how society can use data and AI to solve some of the challenges our world is facing now: from worker shortages and climate change to social engineering and deep fakes.

Read on to see how Ryan addressed some of the questions about the enormous potential of big data and artificial intelligence.

Q: Introduction: tell us about who you are and what you’re working on.

Hey everyone, my name is Ryan Frenz, CTO at Allvision. Allvision’s core product is an AI-based software platform that analyzes sensor data to create and maintain 3D ‘Digital Twins’ of transportation infrastructure. There are an enormous number of sensors observing our world all the time – for example autonomous vehicles equipped with multiple high-resolution cameras and LiDAR sensors. Our mission is to harness this noisy, unstructured data and translate it into information that’s useful and actionable for the humans who design, build, own, and maintain our infrastructure.

As a software architect, I spend most of my time thinking about how best to use AI to create systems that are both scalable and reliable. How can we properly assess the performance of a model and build an interface and systems around it to deliver a reliable and consistent product?

Q: What are some big problems worth solving, and what are the challenges?

At Allvision we’re focused on Infrastructure. A few days ago I did that fun search experiment where I typed the first part of a sentence to see what would be suggested based on popular searches. I typed “America’s infrastructure is” and the top suggestions were “crumbling,” “falling apart,” “not ready for climate change,” and “vulnerable to cyber attacks.” Fixing our physical infrastructure is a complex problem. Funding is obviously critical, but there’s also a massive data problem. Most of our infrastructure is old enough to not have digital documentation, and aging and degradation cause as-built information to become invalid almost immediately. We can’t rebuild everything, so to ensure we’re spending money on the right things, we must understand what we have, where it is, and what condition it’s in. This nationwide survey is a problem that can only be solved with remote sensing and automation.

Q: What are the hype vs. accomplishments gaps in AI?

I think any negative perception is mostly due to the assumption that AIs are perfect, or at least can be with enough training. That assumption is invalid; instead we should assume an AI is imperfect and focus on how to make it useful anyway. That’s a challenging problem, because it requires a rigorous characterization of the system to understand where and when it may fail or perform incorrectly. There is a gap here that is very real, and very obvious when exposed. Consider the example of self-driving car features which perform poorly in certain conditions or without proper reinforcement.

Q: Explain any human-machine teaming issues that you’ve experienced and how you’ve dealt with them. 

They are numerous, but mostly rooted in the assumption that an AI can be a direct drop-in replacement for one or more human tasks. Anyone who uses auto-correct or voice texting knows that isn’t quite true. And like those features, a reliable AI-based system requires humans to reinforce the AI (for example, confirming a dictation before sending a text). Applied broadly, this approach requires different system design patterns. This is something we spend a lot of time on at Allvision. The “human in the loop” is a critical role to ensure reliable and consistent results from an AI-based system. Further, if designed properly the remaining human tasks can be used to directly reinforce and retrain the AI.

For more information on the PTC’s Beyond Big Data Summit 2022, click here. Do you have any AI or machine learning questions? Let us know in the comments!