Waking Up With AI

Anna Gressel: Well, they're not coming to get us soon, but there are some very exciting developments in this area of thought, and many of them are not new. We've seen developments in the area of robotics since before generative AI came onto the scene in fall of 2022. There have been robotic manufacturing lines across the country for more than a decade.

Katherine Forrest: Exactly. First, let's talk about what robots are. They're a mechanical tool, really a physical way of enabling some form of software to move around in our environment. And when you put AI within a robot, you're enabling that robot, hopefully if it works, to carry out various tasks. So, the whole topic of robots is related to AI, but not necessarily dependent on AI, because you can have sort of robotics that are more in line with sophisticated computer programming than predictive analytical decision making.

Anna Gressel: That's exactly right, Katherine, and let me expand on that just a little bit. So, robotics are physical tools, machinery and hardware, and they rely on software, but they can do that in a bunch of different ways. Some robotics are highly programmed—you know, your Roomba might know exactly where to go in your room. Some are even more highly programmed. They're just used by people to physically manipulate space.

Other robots, though, are not very highly programmed. They're much more autonomous and responsive in the same way that generative AI is much more autonomous and responsive to the data inputs that come in. And those are AI-based robots. So, we're really seeing and talking about now this kind of rise of robotics that are based predominantly around AI and not linear programming as in the past.

Katherine Forrest: And they can come in almost any form such as the kind of robotic little vacuum discs that you can now get. I actually saw a YouTube video yesterday of one that was crossing a highway and the caption was somebody's, vacuum cleaners on the loose. They can come in the form of some children's toys. Those came out, boy, like almost 15 years ago, we first started to see some of the robotic toys or a series of tools on manufacturing lines. And that's been a huge efficiency enhancing set of robotics for assembly of cars, as well as in the medical device area. And we're now most recently, I think, seeing robots that are busing tables in all kinds of restaurants and restaurant lounges. I first saw them in South Korea last year, and now I've started to see them all over the United States.

Anna Gressel: Yeah, and so going back, putting back my sci-fi hat on, I mean, we've seen these dystopian views of robots from pop culture for decades now. Let's start with The Terminator, back in the 80s talking about robots that come from the future into the past and all kinds of epic events that come from wars between robots and humans. But there are also shows like Westworld that really portray this interesting question about what kinds of emotional reasoning capabilities robots might have when they live among us. And those are just two examples. There's so many, many different interesting portrayals of robotics in humans in media.

Katherine Forrest: And so concretely, where are we today with robot development and why are we not in the popular culture hearing more about them?

Anna Gressel: Well, I think there are a few reasons for that. First, the excitement around AI right now is within a computer in the server environment that performs tasks humans previously had to perform, but just much faster and more efficiently. So those tasks might be reading 4,000 contracts and summarizing them or annotating them. You don't really need a physical embodiment of that. You just want AI working in your computer environment.

Katherine Forrest: So much of the public excitement recently around AI has been about use cases that have involved really non-physical embodiments.

Anna Gressel: At least in the use cases that I think are being talked about with generative AI publicly, that is certainly the case. And there's also been just so much work that's been going on for so many decades around robotics and AI within robotics that this doesn't necessarily seem new to folks. It's assembly line work, it's Roomba robotics that we've mentioned, but there's so much R&D happening in that area right now that's not even necessarily public or that is just kind of under the radar at the moment.

Katherine Forrest: And you know, some of the research that's been coming out has related to a lot of companies doing some exciting things. And so, there are a number of articles which people can find really fairly easily on arXiv, which is a terrific website where you can access a lot of academic articles. But there are right now a whole bunch of work that relates to making more consumer products, more research on service robots that can actually help with, for instance, hospital services for human beings, additional robotic medical devices, robots that can actually walk around office environments and analyze things and respond to questions. And you can find a variety of YouTube, really recent YouTube videos on all of that.

Anna Gressel: Yeah, and with aging populations globally, there's also been a big push to develop robots that can serve as human companions, including for folks who are elderly.

Katherine Forrest: And in the form of actually animals and people and also things that look just like robot, know, sort of mechanical devices. these human companions are made to look somewhat animal-like sometimes or somewhat human-like sometimes, android-like, and sometimes just like a robot.

One thing our listeners might want to do is check out a post that is been made on X by DeepMind that actually uses an LLM within a robot to have that robot navigate, and I alluded to this a little earlier, navigate an office environment and respond to a bunch prompts as it's sort of wandering around this 8,000 square foot area and you see it navigating in an incredibly complex way and actually being asked, “take me to someplace to write,” and being able to within a contextual environment, the context of where it is, is it in a machinery area where there wouldn't be any writing implements or is it in an educational area where there might be a whiteboard or a conference room. Within that environment, it's able actually to take humans to where they want to be. The possibilities are really becoming, I think, really quite apparent to us all.

Anna Gressel: That's a really interesting example too because they prompted the robot with a series of questions in really a similar way that you'd prompt a generative AI model. So, something like, “find me somewhere to write.” Then the robot would guide the prompter through the DeepMind offices using visual understanding and surrounding context of what different furniture or areas in the office were used for to guide it.

Katherine Forrest: And so, let's talk about some of the legal issues that can come up with some of these advances in robotics. The first series of risks that come to mind really quickly, and we've already seen litigation in this area, are tort-based risks. That relates to humans being harmed by mechanical devices, for instance, on a manufacturing line that might malfunction or the human might do something wrong. And so there can be an injury and there can be litigation that follows. Or you could have your robotic vacuum take in your pet hamster.

Anna Gressel: Indeed. There are also the same issues companies face today with all kinds of generative AI. So, if the AI is trained with an LLM, is it hallucinating when it's coming up with the appropriate context for a task? Or telling a shopper in the store where to find the towel? So, there are accuracy issues that could lead to some legal concerns depending on how serious those accuracy issues are. And a good example of this is hallucinations that might come up in a self-driving car kind telling the car to go to the wrong place and maybe hitting something on the way.

Katherine Forrest: That's right. And there are also some important indemnification issues for people to look out for to make sure that if you've got a robotic implementation of some AI or semi-autonomous robot that you're really looking hard at your coverage and ensuring that the coverage that you think you've got, that you've got.

All right, folks, that's all we've got time for today. I'm Katherine Forrest.

Anna Gressel: And I'm Anna Gressel. Make sure you to like and share the podcast if you've been enjoying it.