This is the download from BFM 89.9, the Business Station. BFM 89.9, the Business Station. My name is Rich Bradbury. This is Matt Splaind, a couple of weeks ago.
We brought you the breaking news that after a mere 100 years of development, the era of the flying car might finally be upon us. This week, we're continuing our journey into the world of machines, as Matt Armitage tells us why science is slick. So you really do yearn for the data machines, we'll replace this document. Richard, you know, I yearn for the data machines will do all of the danger at some dirty jobs.
I mean, that's what they're there for, so that humans don't have to breathe in toxic dust or fumes while they're working, or kill their bodies carrying heavy loads or doing mind-numbing repetitive tasks that end up giving them RSI and all sorts of other disorders. If we looked back to the beginning of 2020 and imagined how quickly automation might escalate over the next few years, we probably never believed that we could have gotten to this point so quickly. We've got robot janitors, sanitising hospitals and malls, we've got all of these stuffless or cashierless supermarkets and convenience stores. And of course, we have the slightly less welcome site of police robots on the streets of Tunisia and robot warning dogs in Singapore that are used as part of their COVID mitigation strategies, or of course, the growing ubiquity of facial recognition systems, which is something we've talked about this year, or that realization by manufacturers that in order to guarantee continuity that their plants, their manufacturing plants, have to be operable with minimal onsite human interaction and intervention.
So this isn't just about automation, it's also about autonomy. Yeah, so our first story is about an autonomous walking excavator, or what my farm friends in the UK would call a digger. Not something that we typically cover on this show, but we should, considering the size and importance of the construction sector. When you look at any construction site, there are huge amounts of complicated machinery, but by and large, the bulk of the work still seems to be done by an army of people.
And while some of that equipment is autonomous or remote operated, for example, some of the enormous dump trucks that you find at mining sites or certain cranes on construction sites, and even some of the bigger harvesting equipment on those ginormous farms that they have in countries like Australia and China. Although, I guess that doesn't really count as construction, but you kind of get the point. What's a walking excavator? Oh, okay, well, a normal excavator has caterpillar tracks or maybe very large wheels and has that crane type thing with a scoop bucket on the front.
Walking excavators were developed for working on hillsides and level ground. So they have hydraulic legs with a smaller wheel at the end, and they can essentially lean into the ground that they're working on. So typically, the kind of work they used for tends to be quite dangerous, because the operator is in a precarious situation, say, on a hillside. So there's a big advantage for this type of machinery to be remote operated at the very least, and autonomous at the very best, just to protect those human workers.
Is autonomous such an advantage here? Well, if you're looking at it from an employment point of view, then obviously remote operated is going to be better. But if you're a business owner and you want to reduce the labor even further, then autonomous may look like the best solution. But there is a shortage of construction workers in some countries, and of course, autonomous machines work much longer hours than human humans.
But that autonomy can also lead to other forms of cooperative working, either with other autonomous machines or human laborers. In any case, the important part is not so much that we need one, but that we have one. And no joke here, it's called the HEAP, H-E-A-P, which is an example of hydraulic excavator for an autonomous purpose. You can see they came up with the name first and kind of...
Yeah, yeah. It was built by a Swiss German team who converted a normal walking excavator and used it to complete a number of tasks autonomously without any human interaction. So it's like the self-driving car of tractors? Well, you're not too far off.
It's a mixture of artificial intelligence and detection systems like LIDO, which allow it to build a 3D picture of the surrounding areas. So they used it to construct a 4-meter high dry stone wall. That's a pretty good accomplishment. They got it to move trees around as though it was doing landscaping.
And they also used it to excavate a trench that was full of Second World War-era abandoned ammunition. And I think it's the roles like that, where it really comes into its own for whether it's nuclear contamination like Fukushima, the polluted wastewater retention ponds that flood from time to time in the United States, or simply things like mining settle, this kind of tool may turn out to have a significant societal impact, which is not something you often get to say about construction equipment. The finished buildings, yes, but not normally the methodology of actually getting that. Now, important question, Matt.
Where can I buy one? Well, it's still proof of concept at this stage, plus I don't know if you're looking for it. But there's nothing to stop you creating your own. Or you need to review LIDAR sensors, a bit of artificial intelligence, and a walking excavator, stuff that most of us have at the back of the garage gathering dust.
You can say, we're breaking new ground with the story on that, Slinged. I have to apologize to everyone I made Richard say that. And I've mentioned that we often ignore stories like this. So this next story is about something else that I would typically overlook for this show, which is football.
I was banished from the UK for not enjoying football, a bit like you, I think, Richard. Exiled ever to wonder the Earth a succulous wraith. You do like a bit of drama, don't you? Well, if I just said a computer just taught another computer to play football, that sounds a bit boring, really, even though it's actually really very clever.
So researchers at the Google adjacent AI developer DeepMind have used their supercomputers to teach digital people to play the game. See, that's why I say things like human people, because I'm talking about digital people. So if you play soccer video games, then you'll be familiar with the kind of animation and players that are created by this simulation. They were divided into two teams of players.
At the start of the simulation, the players just wobbled and staggered around the pitch. So the first thing they had to learn was to stand up and run. And this was done by training them using motion capture data from actual soccer players. And the system then used reinforcement learning, where successful actions are rewarded to train the players in basics like kicking the ball, dribbling, passing, and shooting at the goal.
Sounds like I could have done with that at school. But how long did you take them to learn? Well, you'd say that I was put off sport because at primary school, I was putting the gloves on a cricket to be a wicket keeper and the teacher bowled down and hit me in the face with the cricket ball. That was it for a hard sport.
But anyway, considering the older aphorism don't run before you can walk, the soccer players, like I said, they had to be taught to stand up and walk. So within about 24 hours, they could run, they could get up if they were knocked down, and they'd learn to keep possession of the ball while they were doing those things. But at this point, they were really only concentrating on the action they were carrying out. They weren't thinking ahead, not even one step or two steps.
So they weren't able to work as a team. If a player had an opportunity to shoot, it would just go ahead and do so, even if its teammate was in a better position to score. So it took the players about three days to learn the basic rules of the game. But it took about 10 days of training before the players really started to think ahead and work as a team.
Look, I know neither was an expert on the game, but I'm pretty sure this was more than four players on a pitch. Well, obviously, this is a limitation of the simulation at this point, and the players would taught a very simplified version of the game. For example, they weren't taught what a foul was. It was the thing my teacher did with a cricket ball.
But I guess because there's a moral component to that action as well as a behavioral one, so it's hard to teach. And they had to essentially create a force field around the pitch for the players to understand the boundaries. They had to understand they couldn't cross those lines. Okay, so the autonomous excavator, I can understand as deeply boring as it might be.
But maybe you could explain why somebody would want to do this. Well, the key here is that the players weren't simply programmed how to play as they would be in a typical video game. They were a blank canvas and they learned how to play using that reinforcement learning approach. In a sense, things like this, it's being done to see how far we can push machines when it comes to this process of self-learning.
More importantly, for those machines to help us in the future, they have to be able to assist in problems that perhaps we can't adequately explain. And football makes a good model because it's complex? Yeah, it involves a lot of incredibly complex calculations and real-time variables. There's a lot of physics involved in chipping a ball over to your colleague to send it into the goal.
So if we go back to that idea of the excavator, ideally, you would want the AI running that machine to be predictive, to know that it was about to rain or storms about to hit, to have an idea how a slope might move, to know what humidity or water might have done to the ammunition in the trench, and to adjust and make decisions on the fly just like a person would. So although they may not seem to be a connection here, we can only make the machines as good as the brains that controlled. All right. On that note, when we come back, machines creating the next generation of machines here on MatSlain on BFM 89.9.
Building future Malaysia, BFM 89.9, the Business Station. BFM 89.9, the Business Station. This is MatSlain. I'm Rich Bredbury.
Now, so far, today's episode has been full of tractors and football. It's beginning to sound like a minute to just childhood. I mean, you say that as a joke, but it is actually true. You know, my brothers used to play on and around farm equipment when I was a kid.
I mean, you probably wouldn't be allowed to do that now, but back then, there weren't as many rules and people didn't have as many fingers. I have to say, I really enjoyed the recent show, Clarkson's Farm on, I think, Amazon, which was all about plowing and seeding and harvesting and all the disasters that go along with it amplified by that Clarkson factor. I think a lot of people seem to have agreed with me. It's apparently the streaming platforms most watched show ever.
That's interesting. Yeah. Something else that was a big part of my childhood, other than farming, was the sci-fi classic Hitchhiker's Guide to the Galaxy. That's an incredible segue.
Did you ever have an original watch that when you were growing up? I did. A towel is about the most massive, useful thing, and it's just a hitchhiker can have. You know that?
Yeah. I take a towel with me everywhere to this day. Sorry if we're giving away any spoilers here, but it's 40 years old already. Yeah.
One of the best bits for me at least is when a computer called Deep Thought is constructed by the Magrathians to solve the ultimate question of life, the universe, and everything. Deep Thought tells its creators that it will take 7.5 million years to solve this commandrum. 42? Yeah.
The answer is 42. Douglas Adams has always been a huge inspiration for me, Douglas Adams and tractors. Deep Thought tells the descendants of its creators that what they really need to know is the question. But Deep Thought isn't powerful enough to quantify that question.
So it sets about designing and building its own successor. And that turns out to be a biocomputer called Earth, which is blown up by the Vogons to build a hyperspace bypass shortly before its calculations were completed. So this has always been one of the fundamental questions in computing and artificial intelligence. Not the 42 part, but should we trust the machines to create a more powerful and next generation version of the machine that they are?
And it's an idea that runs through science fiction, whether it's satire, as much as it runs through the dystopian vision of franchises like Terminator and The Matrix. Someone's brought this dystopian vision to life, haven't they? Well, yes. Or someone is looking for the question that's answered by 42.
Unsurprisingly, it turns out to be Google, which has started using its own AI chips to develop its next generation chips with quite astonishing results. So this is probably something that a lot of people don't know, not because it's a secret, but just because, again, it's boring. Because of the scale of everything that Google operates at, it designs and manufactures its own chips and creates its own software, which makes sense. When you've got data centers and AI and all those search results to serve, you're not going to be buying things off the shelf.
And Google is always hungry for the speed up better chips, bring? Not in a sinister or nefarious sense. I mean, again, you look at the scale that the company works at, any incremental improvements in chip design, whether it's in terms of speed, energy consumption, or even adaptability to certain tasks. That's going to have a massive effect on their overall business.
And because of the complexity of those chips, it has separate teams working on different sections of them. So each team is tweaking variables in the section that they're assigned to. And then they come together and try and piece all of these bits together a bit like a jigsaw. So this is a process that can take months or even years depending on the complexity of the chip they're developing.
How much faster are chips at designing better chips? Well, this is where it's genuinely, I was going to say the word scary, but I think I'll stick with the phrase jaw dropping. The researchers at Google use the latest version of Google's Tensor Processing Unit, which is designed to run neural network algorithms. So in just six hours, the AI was able to design chip schematics that equaled those produced by the human design teams over several months.
So we're potentially looking at shortening the design periods for new chips from years to days. So for Google, a large part of this is about cloud savings, being able to implement new chip designs even a few days early can save them in almost sums of money. But short-cutting that process by months or years could equate to absolutely vast savings. I've got a question.
Is today's team boring but important stories? In a sense, I guess it is. This is something that should be a much bigger story than it is. If companies other than Google can use or can license the use of similar design methods, imagine where that might take us.
Again, if we go back to towels and the hitchhiker's guide to the galaxy, Marvin the paranoid and his refrain of here I am, brain the size of the planet, and then he's dispatched to do some manual task. Could this accelerate deep mind's ability to teach two full teams of footballers how to play an act as a team? Does it mean that teams of autonomous excavators could go out demining in the world's former war zones? Could it lead to better modeling of climate and weather patterns or give advance warning of the kind of building collapse that we've seen recently in Miami?
Without hyping the outcomes, this is a story that really should be all over the news. Let's go from on TV to in TV, video conferencing to be precise. Are you starting to get sick of reading or watching interviews where everyone says, I'm so glad we don't have to zoom anymore. It's so much better to be able to meet in person.
Newsflash to the first world is a lot of us are still locked down and zooming or avoiding zooms in my case. So anything to make that experience more bearable I think is a worthwhile addition. Researchers at the University of California San Diego, one of whom is a computer music specialist who was frustrated by the limitations of video conferencing software when it came to giving his online lessons and most importantly about the lack of non-verbal cues that we can get from other people when we're using the platforms. Such as who's looking at who?
Yeah, essentially when you're in a meeting and someone's talking, you don't really know who they're directing their attention towards. So you get those awkward moments when people ask questions and nobody responds because nobody really knows who the question was addressed to. And to fill that silence, a bunch of people jumping at once and of course a lot of the video platforms will only prioritize one voice at a time. So online meetings are full of these behavioral knots that have to be untied and that can be very frustrating.
Boring. Yeah, we're going back to that theme for today. I don't know how many times we've said boring a lot. So the guys at San Diego used machine learning to create an algorithm that tracks the eye movement of the speaker.
So depending on how we have our screens or our view set up, different participants are in different places on different screens. So the algorithm works to identify where on the screen the presenter is looking and which person that correlates to, which is a fancy way of saying who the speaker is looking at and hoping that one of your colleagues hasn't decided to take a shower with the bathroom door open as I think we saw from a Spanish council meeting because then it can inform those people that the presenters eye of Mordor is suddenly upon them. And I know that it sounds like such a small and insignificant thing, but the thing about nonverbal cues is that while they are small, they're anything but insignificant. They are our feedback mechanism, which is why we find it so weird when there are people that we struggle to read.
And to give you an idea of how effective the makers hope this will be, the computer music specialist Shlomo Dugnav hopes that it will allow a conductor to direct an entire orchestra during remote practices. So we're talking about that level of accuracy to be able to deal with over a hundred people on the screen. Fantastic. Now, one last story.
Can we have a machine that's actually about a machine rather than software? Okay, this is a cool one about pain management. Again, not a phrase you often get to say. We have a lot of impending health crises.
Some of them are anticipated like the effects of long COVID and the lifelong damage that some of the coronavirus sufferers are likely to experience. Other factors like longer life expectancies mean that people will be living for longer periods with persistent conditions. So we're expecting to see the incidence of chronic pain increase over coming years. Chronic pain can be hard to treat with medication.
A lot of pain medications do have long-term side effects, some like opioids can be effective in the short term, but the recipient quickly builds both a tolerance and dependence. So we've already seen some non-medication based interventions, and there have been successes with treatments like electrical nerve stimulation. You're using implanted electrodes. Why do your stories always come back to putting sensors and stuff into people?
Hang on. Yes. Okay. Usually they have to be implanted in the spinal cord under general anesthetic.
I mean, and there are usually clusters of the electrodes, so often up to 32, that send those electrical signals along the spinal cord to the brain. And these signals tell the brain to ignore or stop processing that pain. The surgery is complicated. Like anything that happens to your spine, it comes with risks.
So they may even have to remove part of the spinal cord to fit the implant. So it's not a perfect solution. Scientists at the University of Cambridge have come up with a tiny implantable device made of plastic and gold sheets that when it's rolled up has a thickness of less than two millimetres. Does that mean it can be injected?
Yeah, and that's the really cool thing about this. It's still at that testing stage. If successful, it means that patients could have a relatively simple procedure under local anesthetic. It's implanted in the epidural space along your spine.
And it can then be unfurled like a balloon by filling it with a tiny amount of air. Once in place, it then starts sending those signals that deliver pain relief because it actually wraps itself around that part of the spine. So how's it powered? Well, I think these are the kinks that they're still trying to work out at the moment.
The most likely solution is that they'll use an implanted battery from which the implant would draw power by induction. So similar to those charging mats that some smartphones use, the team has successfully implanted the device into cadavers to prove that they can implant them without damaging the spinal cord, not to try and reanimate the body. But there's a lot of work to be done before it can be trialled on living patients. If it does get to that stage, the hope and the expectation is that it should carry no more risk than receiving an epidural injection during the labour part of pregnancy, which could result in thousands more people having access to medication-free pain relief treatment.
And that's a pretty good outcome for such a tiny little machine. It certainly is. Thanks for that, Matt. That's my pleasure.
So, so long and thanks for all the fish. Ladies and gents, you have, of course, been tuned in to Matt's Blaine here on BFM 89.9. If you missed any part of the show, don't forget you can download the podcast wherever you know and listen to your podcast from. I recommend downloading the BFM app.
It's available from the Apple App Store or Google Play and don't forget you can find Matt on Instagram and Twitter at culture mat. You can also head over to culturepop.com for transcripts of these shows and information about culture pop and its consulting services. BFM 89.9, the Business Station. Thank you for listening to this podcast.
To find more great interviews, go to bfm.9 or find us on iTunes, BFM 89.9, the Business Station.