Home
Trending
New Release
Speaker A: This program is brought to you by Resonance 104.4 FM. If you like what you hear and want to support our work, please make a donation at fundraiser.resonance.fm.
Speaker B: Hello, I’m Peter Warren and this is Password, checking out the future of work. The boss is a bot. I’m a self-employed freelancer. My working hours are not dictated by an algorithm, but rather by how much I’ve got to do on any given day and whether I have to take the dog to the vet if he keeps on being sick. But more and more jobs are being defined by gadgets that program workers’ hours and movements according to calculations. They use big data, satnavs, GPS tracking, and project management software to achieve the maximum possible productivity. Indeed, 2 months ago, The Economist magazine pointed out that funds run by computers that follow rules set by humans account for 35% of America’s stock market, 60% of institutional equity assets, and 60% of trading activity. Artificial intelligence programs are writing their own investing rules in ways humans only partly understand. Industries from parcel and pizza delivery to films are being changed by technology. But finance is unique because it can exert voting power over firms, redistribute wealth, and cause mayhem in the economy. They can force companies to close. In a world of online ordering and instant consumer gratification, a bad reputation can get an employee sacked and force a company to lose a delivery contract. It may seem efficient. In practice, it can be cold and deadly. Ruth Lane’s husband Don was a franchised van driver. He died after collapsing in a diabetic coma while he was making deliveries in rural Dorset. Ruth had to make the heartbreaking decision to switch off his life support machine. She blames the automated system that organised his job for causing his untimely death. It tracked his movements, scanned his parcels and logged his deliveries through a handset known as a gun.
Speaker C: Yeah, it sounds terrible. I don’t know where that name came from. It sounds awful. Like, yeah, Don— I mean, Don was found, which sounded terrible, collapsed. He pulled over because he felt unwell. This is when he collapsed and fell into a coma. He had the gun, his gun in his hand. He slumped at the wheel, you know. And I thought, well, that sounds terrible. What else can I call it? Because Don would call it his gun. And what it is, it’s a mini computer. It zaps the parcel, it organises the route, it tells you which parcel to do next. And of course, DPD, hourly slots. It tells him he’s got to do this parcel by this time. One time he missed the hourly slot, and that was just on the last year of his life. Knowing he collapsed into a coma, so he wasn’t, you know, 100% well. They gave him a final breach letter for that because he delivered a parcel that missed his time slot, which is just, just ridiculous. So unfair, isn’t it? Strictness and, and this stupid, stupid fine they had that was in the contract. How on earth can they fine you for not attending, basically not getting us a relief driver. It’s in the contract. It’s stupid, isn’t it? I mean, so many people have said, well, why’d he sign the contract? Well, that’s because he wants to work. He wants a job, and he really believed— because they sell it to you like you can earn this much money, and it’s called a blueprint. And what they say is, we’re going to pay you this much per parcel, We guarantee you, you could be— you’ll be delivering these amount of parcels a day, and that’s how much you can earn. Well, you don’t, do you? You don’t earn it. He wasn’t hitting his blueprint.
Speaker B: There is the amount of money you can earn. I mean, obviously, if you could soup up your, your van and drive completely to the limit, then there is probably a finite amount that they say you can earn. But is it possible to earn that amount? And what amounts are we talking about as well?
Speaker C: Well, Don’s route was— it wasn’t the best route because it was rural. I mean, he had a good, sometimes a 10-minute drive to the next area, whereas some of the couriers, it was all in one street or it was an industrial estate, so they could literally get out the van, boom, boom, boom, that’s 5 or 6 deliveries done to different— but they— so no traveling basically. Because the van was there, or Dom would have to drive to each. And he said some of them, they had such long driveways because they were quite big houses, because we’re talking Dorset, and you know the traffic in Bournemouth and Dorset, it’s just terrible. So you’ve got that against you. And there’s this thing called quantum. I don’t think you’ve ever heard of it if you’ve talked to any other drivers. And what it does, it works out the quickest route that you can do your route this amount of time, you can deliver all these parcels, and this is— and it tells you the time that you’ll be finished. I mean, that’s how much this handset, this gun, dominates your life. And Don had to complain so many times because the actual mileage that the handset, the Quantum, said he’d be doing, say 100 miles a day, well, he was actually doing 110, 120 miles a day.
Speaker D: Day.
Speaker C: So we’d have to point that out to them because it hadn’t long come into force at the depot in Don’s last year of his life, you know. And it would have him crossing farmers’ fields to get to the next route, and it was like, there’s no way I can get there in that amount of time.
Speaker B: So essentially it was making a calculation almost as the pigeon flew, as the crow flies.
Speaker C: It was. It was like Google Maps, yeah. It wasn’t a road, it was a farmer’s dirt track, you know, for his private field. God, that poor man. But you know what though, he was so good at his job. I was so proud of him. He’s like, he’s what I call old school, um, you know, because he’s been a courier for so many years. I mean, with Parcel Line/DPD, it was 19 years. And he was the one that, you know, when you— can you remember those red books you just get of all the street, the street maps you get of a certain area. He would memorize those. And you know, he was so amazing. There’s quite a few in the same depot as Don, same knowledge as him. 2 weeks, they’d memorize that whole route because you’ve got to remember the one-way systems, times that they would block off the road where, you know, it’s pedestrianized. And, and oh my goodness, he was good. He was really Good. Yeah.
Speaker B: So did Quantum take into account the things like traffic? Did it take into account issues like roads being closed?
Speaker C: No, no, not at all. That’s why every day when it started, Dom would email his manager. I won’t say his name. I want to name and shame him. He’d email his manager with the correct mileage you know, just to make sure he was going to get paid properly. Because what— there was a sliding rule. I don’t know if you’ve ever seen one of these. It works out how many parcels you’ve done, what your mileage is. So the more mileage you do, you’ll maybe get a bit more per parcel. But obviously, if like some of the drivers, you had quite a condensed route, not much mileage, but you had loads of parcels, you you’d earn less per parcel. So they were very calculating. They made out you’re going to earn so much, and then it went, no, actually, no, the more you do, you’re going to get paid a little bit less now because you’re not doing all the mileage. And it was like, what difference does it make to them? Because the franchisees are paying for their fuel, you know, so it’s not like it’s costing DVD anything, you know. So why they’ve got paid less God knows. Yeah, this handset did everything.
Speaker B: So this was all packed into— he’s essentially— his life is run by computer, he’s answerable to computer, and the computer is informing on him.
Speaker C: Definitely tracks him, so he can be tracked. So if, say, he didn’t quite— because they liked the van to be parked as near to the customer’s door or the property as possible. Say there was one time actually he’d parked up the road because he couldn’t get parked outside the house. And they even questioned that, said, you know, you didn’t actually park outside the house, because I think the customer said they, they never saw the van or something because they weren’t in or something. I kind of— some complaint. And he said, I couldn’t get any nearer, you know, because the car’s parked. You know, that’s a— that’s hard as well, parked cars. But Yeah, so they knew where he parks.
Speaker B: Exactly. So, but surely if he is his own man, if he’s running his own company, if he is the franchisee, he’s the one who can elect where to park and whether to sanction himself. This is— this isn’t that the, the independence that is presented?
Speaker C: Yeah, he’s in his knowledge, his judgment. And also, if he was a genuine self-employed person, Shouldn’t he be charging them, telling them how much he’s gonna charge them for a parcel, not them telling him, “We’re only gonna pay you this.” And also, in the contract, I don’t know if you know this, they’re not allowed to use their vans to work anywhere else. And even if, I mean, Dominic only leased his van, but some of them have bought their vans. And it’s all, Don used to call it livery, livery or something, all signwritten. And you’ve seen DPD vans, the back doors are completely covered, aren’t they, in red. So there is no way they can use those vans to work anywhere else. And if they’re seen working anywhere else, breach of contract and then sacked. So even their own van isn’t their own van. So if, say, they wanted to work a weekend for another courier, they couldn’t. They weren’t allowed to. Well, you couldn’t anyway because it’s all completely covered in advertising DPD. So, but if they were spotted, very strict contract. And it’s so tight as well because I say to Don, how come nobody’s ever taken them to court over this? And he goes, they’ve got the best solicitors who’s written the best contract and it’s tight as anything and nobody dare take them on. Well, guess what? I am.
Speaker B: Ruth Lane is crowdfunding to raise enough money to take the delivery company DPD to court for creating the conditions that meant her husband could not take meal breaks or days off. He was fined for attending a hospital appointment and then missed two further vital medical checkups. In Glasgow, fellow DPD delivery drivers organised a strike on the anniversary of Don Lane’s death in his memory, but were obliged to call it off because of legal action. Drivers for the Deliveroo platform have organised successful strikes in Brighton, and there is now an American union of pizza delivery drivers based in Florida who are campaigning for the state minimum wages to be paid. $6.40 an hour. Don Lane was paid £2 per parcel delivery and had to lease his own van and pay for the fuel, insurance, maintenance, and mobile phone. It’s a case that signposts the possible dystopian future of work. And spookily enough, I’ve been listening to an artificial intelligence that was debating whether AI will have bad consequences for humanity, precisely because of its effect on jobs and working practices. It was at the Cambridge University Union where technology giant IBM was showcasing its Project Debater software for the first time.
Speaker E: While my job with IBM is secure, at least I hope so, I know this issue is quite pertinent to our discussion today. AI will make lots of people lose their jobs. It will bring more harm than good in that it will displace a lot of workers and cause employment problems. We risk creating a workforce that puts people out of employment. Jobs involving vehicles such as travel is one of the biggest employers, and those jobs will be lost because of AI. Moving on to societies. While AI technology is cool, advanced technology can and will lead to human beings becoming lazy. AI being created to do things for us will leave us in the dust culturally and ruin our future society. Just because a technology is ready doesn’t mean society is ready for it. Development without considering social situations could be detrimental, for just because something can be done doesn’t necessitate that it must be done. AI will take the human element out of almost everything we do and lead to a very disconnected society. Overreliance on AI may prove to be a massive problem in the future, as certain skill sets may be lost from societies entirely. If governments don’t take actions rapidly and impose regulations on ethical and peaceful use of the technology, there will be clearly a danger that somebody will misuse the power of the technology. In the hands of drug dealers, fraudsters, and rogue nations, AI will magnify exponentially their ability to do harm. Since machines can collect, track, and analyze so much about you, it’s very possible for this information to fall into the wrong hands. From criminal organizations to authoritarian states, it will bring more harm than good by making people dependent on it and limiting important decisions Leaving everything in their hands. AI further concentrates production and decision capabilities further in the hands of a few super-rich people. Currently those have no tendency to democracy and sharing of wealth. In the hands of the wrong people, it could be programmed to do something devastating. These were the arguments from the crowd, supporting the idea that AI will bring more harm than good. Thank you for listening.
Speaker B: IBM’s Noam Slonim, based in Tel Aviv, developed the Project Debater as the latest in the company’s stable of thinking machines, building on the success of Watson and Deep Blue. He explains how it works.
Speaker F: So the system has two major sources of information. One of them is a massive collection of newspaper articles, around 400 million 10 billion newspaper articles or 10 billion sentences. When the debate starts, various AI engines come into action and they try to pinpoint short pieces of text that satisfy 3 criteria. They should be relevant to the topic, they should be argumentative in nature, they should argue something about the topic, not just being relevant, and they should hopefully support our side of the debate. And then the system is trying to use other engines to glue them together into a meaningful narrative. The other major source of information of the system is a knowledge graph that we developed over the years trying to model the— to capture the commonalities between the many different debates that humans are having. So in this knowledge graph, we have thousands of more principled argumentative elements. When the debate starts, the system is navigating, so to speak, in this graph searching for the most relevant principled arguments to use them in the right Let me make it more concrete. By a principle argument, let’s say we are debating whether or not to ban the sale of alcohol or to ban organ trade. As you can imagine, in both cases the opposition may argue that if we ban that, we are at the risk of the emergence of a black market. So a black market is a principle argument that can be used in different contexts. One may naively think that this is a keyword matching thing. If you ban something, The opposition will use the black market argument. As I’m sure you can appreciate, this is obviously not the case. Think of a debate of banning breastfeeding in public. Probably there is little risk of a black market in this case. Or a debate of banning internet cookies. Probably we will not see a black market of internet cookies. So really the system needs to develop a more subtle understanding of human language in order to perform well in this task. And finally, there is rebuttal. The system needs somehow to respond to the opponent, and this starts by understanding the words articulated by the opponent. For that, we use Watson speech recognition capabilities out of the box. But of course, we need to go beyond the words. We need to understand the gist of the speech to respond, and we use various techniques. Most of them rely on the same principle. Try to anticipate in advance what kind of arguments the opposition might use, and then listen to determine whether indeed the opposition was making this argument and then respond accordingly. So this is at a high level how the system actually works. Let me give you just a couple of examples why this is so challenging. We’re looking at a debate whether blood donation should be mandatory, and there are two sentences here here, perhaps hard to read, but one of them is talking about the fact that blood donation is good for your health. And the other one is talking about the fact that students are common among blood donors. So we understand that the first one can be used in the debate and the second one is relevant, but is not very persuasive. But how do you teach an AI engine to make this distinction? Here is another example of a highly important debate, whether or not we should abandon Valentine’s And here we have a sentence talking about a survey done in Canada where basically most people answered that Valentine’s Day is a waste of time and money. Well, this is fine and this can be used in the debate. But another survey done in the US telling us that most people, if they are going to break up with someone, they will do that just before Valentine’s Day to save money, which is a very useful piece of information. But not necessarily useful in the context of this debate. So again, how do you teach an AI system to make this distinction? This is quite challenging.
Speaker B: Noam Slonim. And the really uncanny part about it is that the debating machine can put equally strong arguments in favor of AI as a positive benefit in the workplace.
Speaker E: About automation, since it allows machines to do jobs that many others would not do due to the repetition or monotony, and since it will open up a lot of opportunities in everything from entertainment to daily tasks, and because it can help advance technology that helps handicapped people be able to live a more normal and independent life, then obviously AI will not bring more harm than good. AI will allow to automate repetitive tasks and eliminate human errors. It will be a great advantage as it will free up more time from having to do mundane and repetitive tasks. Next, a few words about jobs. AI will make the job easier for many people. It will create new jobs in certain sectors and help companies to advance. AI will actually be beneficial because it will cause new types of jobs to arise for people who take care of the AI equipment. It will make our lives easier by taking care of more of the mundane tasks. Artificial intelligences will make jobs more efficient and specific in the future. AI helps us organize our chores better and makes our daily routines simple. AI can bring a lot more efficiency to the workplace. AI capabilities caring for patients or robots teaching school children, there is no longer a demand for humans in those fields either. Let’s move to an issue close to my artificial heart: technology. AI will enable technology to advance and further medical research which will save lives. It will enable us to develop more and more impressive technology. While regulation and serious consideration of the concerns are in place, the benefits of AI technology are enormous and are way beyond the over-exaggerated potential harms. Autonomous vehicles are prime examples of how artificial intelligence is impacting the automotive industry. A large segment of autonomous vehicles are connected and thus able to share the learning with each other. Society has and will continue to benefit from AI based on character facial recognition, digital content analysis, and accuracy in identifying patterns, whether they are used for health sciences, academic researcher technology applications. Early AI technologies such as autonomous cars, voice assistants, and GANs are already enriching human lives. There is no reason to doubt that this will continue into the future.
Speaker B: What we’ve heard from the bad AI and the good AI illustrates perfectly the dilemma that’s facing us as we approach the year 2020. The so-called gig economy is in full swing. App-driven self-employment is already widespread. We have taxi services like Uber, food and parcel deliveries that cannot yet be sent out by drones or robotic vans, and automated checkouts at supermarkets and check-ins at airports. But what does this mean for us humans? This is Password on Resonance FM, debating whether or not AI is a force for good in the workplace. After this on Resonance, you can hear A World in London with DJ Ritu. Of course, it’s easy for elite academics at one of the world’s top universities to argue in the abstract about what it means to work with or work for software robots. Perhaps what they’re missing is the extent to which this is already happening. Professor Neil Lawrence is a DeepMind professor of machine learning at Cambridge.
Speaker G: It’s a very powerful technology that is in a very nice way complementary to our own abilities. But the challenge with it is our own method of computation is, because we’re so limited, is to use our powerful computation in our head to think about about the motivations of all around us and anthropomorphize the things we communicate. And we do that to these machines. That’s why we like to give them names. But in reality, they don’t have names. Now, there’s a danger to this, because that’s the point in the quote, “Lies, damn lies, and big data.” They are a new route to manipulating statistics as presented to us, facts as presented to us. In the past, this danger was perceived, in the 1890s, and the invention of the field of mathematical statistics was designed to deal with that danger. So people like Galton, Pearson, Fisher, they looked at data, the misrepresentation of statistics, and they said, “This is how you represent it so we can draw correct conclusions.” Unfortunately, they also decided that an appropriate use of this new technology was eugenics, because they thought they had some single access to some underlying truth about how you could prove that humanity should move forward. That is an enormous mistake that we continue to make where AI is present, that there is some objective, that there is some truth that we can optimize ourselves towards, that we are anything more than a collective of information-processing individuals who are massively handicapped in our ability to communicate to each other, and we perform this extraordinary an extraordinary cognitive dance in order to do so. We have created entities that are undermining that dance. So when it comes to our interaction with these entities, we place them in roles where they can see who we are. They peer deeply into our soul because of the amount of data we trail on an everyday basis. So these machines know us better than we can know ourselves. How can that be? Because within you, there is a model of who you are that is incorrect. You all think you’re nicer people than you genuinely are. The machine knows who we are. That is limiting our freedoms. It’s limiting our aspirations. Because through knowing who we truly are, the machine can undermine us. And it does that in an emergent relationship where— one way of putting it is dual process cognition theory, which says that there are two systems. One is the higher system, the higher reasoning, slow-thinking Kahneman would call this System 2, and System 1 is the low intuitive fast thinking. The machines, when they provide us information on social media, are plugging into System 1, not System 2. They’re creating something that is addictive, something I think of as like the high-fructose corn syrup of our cognitive diet, making us cognitively obese as we consume this material. As they do that, They are creating an ecosystem that sits under us, something I call System Zero. An ecosystem of shared behavior that draws us into a certain direction, that divides our society, that feeds our worst instincts, that separates us from our higher cognition, our ability to think above ourselves and work together as a community. That’s the challenge we’re facing with the next decade of AI. It is anti-diversity. It is bringing about a form of late eugenics. The rewards of this technology are many and have been elucidated beautifully by Project Beta, and I’m sure they’re going to be even more so by Herwish. But bear this in mind: over the next 10 years, we will be on a perilous journey, a journey where we are going to the heart of who we are and undermining our very selves. Now, my main argument is akin to Pascal’s wager. We should believe that AI will do us harm because it’s the best way to prevent us from falling into those harms. If we say here that AI is some universal good that will take us on a journey of freedom and health, we’ll be in for a very sorry ending as a civilization.
Speaker B: Professor Neil Lawrence. The DeepMind professor of machine learning. DeepMind, of course, is the research company owned by Alphabet, the parent company of Google. And of course, Google is researching machine learning because it’s a massive generator of data collected from all our online searches and preferences and information from our mobile phones. And it is data more than talent or track record that will define our future career paths. Emil Eifrem, the founder of San Francisco-based Neo4j.com, has built a global business on it.
Speaker D: It really to me comes back to some very deep philosophical things around, you know, what is behavior, what is existence. And it’s not clear to me that if you take exactly the same— we move outside of human beings— if you take exactly the same data and exactly the same history, that you will arrive at exactly the same conclusion all the time. Because this is, again, comes back to the whole fact that fundamentally what AI and machine learning is about is probability. It is not about getting to a certain point, right? So it’s not always that you train the same models with the same data. They don’t always get to the same conclusions.
Speaker B: Psychologists, for example, would say, and a lot of psychology seems to be based upon the fact that we are, have been governed by some of the events that have occurred to us. You know, you have soldiers who have got shell shock. You should theoretically be able to determine the soldiers who are likely to have shell shock. You have people who, if we’re led to believe what psychologists seem to suggest, that if they have traumatic experiences in their early lives, that they are affected by those traumatic experiences. So we try to steer people away from having those traumatic experiences.
Speaker D: Yeah, which, which I think makes sense. That to me is not a— the evidence— like, that is not evidence that our data is our existence though. And that if you have the same exact set of data, that that leads to exactly the same conclusion for all individuals. And I don’t think that’s true for human beings. I don’t think that’s true for artificial intelligence either. Although of course it’s a factor. It’s a huge factor, of course. But that is then when kind of those black swan events come in where all of a sudden, hey, there’s a small anomaly over here and there’s a small anomaly over there, and that then with that compounding effect of that in machine learning ends up with a different conclusion. Now, I think probably the vast majority, like the area on the bell curve here, is probably that it is much more likely that it leads to the same predictions, but not always.
Speaker E: Okay.
Speaker B: Are we run by our data? Is in this new AI world that’s about to emerge, in this new smart city world, are we going to be run by our data? Slaves to our data and the manipulation of it?
Speaker D: Yeah, I don’t think of it as being run by or being slaves to it. I mean, I think maybe this is, it’s just syntax and terminology. I do think that it is, we are being in a sense kind of manipulated at a scale that we’ve never seen before. We’ve certainly been manipulated before. Like if you look at TV ads, right? When the, When the TV came out just the other day, my 4-year-old, I was grocery shopping with her and she was pointing to, “Oh, this laundry over here, this one is perfect for cleaning,” based on some ad that she’d seen somewhere because she’s so susceptible to that. Now, I think that is very much the crawl version of this, whereas with with social media and all this vast amount of data that people have about us and the fast iterations that you can do, we’re moving from the crawl version of that manipulation. And manipulation sounds negative, right?
Speaker B: But the thing is, if that weren’t the case, Emil, if that weren’t the case, then Google wouldn’t have the price valuation it has as a company. Facebook wouldn’t be valued as much as it is as a company unless you can interfere with that desire process or influence it or manipulate it.
Speaker D: I think you’re exactly right. And that’s why I think that that comes back to the fact that, look, I don’t think— I mean, it’s very easy to paint this as kind of evil or nefarious ill intent and whatnot. I don’t think that’s true at all. But I do think that these companies and you and me and all of us live in an economic system that incentivizes things like this, where it’s just very profitable for them to give advertisers an ability to target us in that way. And all of a sudden, those advertisers have a really instantaneous way of measuring the impact on it, whereas, you know, coming back to the TV example, it may take months or years before they actually saw the the impact of this, it’s much more instantaneous now, which just means that the iteration cycles, the feedback loops are much, much shorter, which means that they just become much better so much faster. And without that, of course, like they wouldn’t have these, you know, top 10 biggest companies in the world type valuations, right? Or market cap. And, but I think that’s a, That’s just a consequence of this being valued a lot by a lot of people.
Speaker B: That was AI innovator Emile Eifrem. Philosopher Barry Smith of Edinburgh University believes the advent of AI in our working lives will be different from earlier technology-driven industrial revolutions.
Speaker A: As Dan Dennett, the philosopher, put it, and I think so aptly, he said, robots just don’t have any skin in the game You know, they don’t know what it means to be deprived of something, to be suffering, to, you know, to be hurt. He said, when you play poker just for matchsticks, it’s not the same as playing for money when it really hurts when you lose. That’s what it is to have skin in the game. And he said, so far we can’t think of an analogy to that in robots. And so they don’t have that extra element of human desire. They just have goals.
Speaker B: For the purpose of this program, We’ve been experimenting with a robot vacuum cleaner. It wanted to be given a name, so we’ve called it Bartholomew. Bartholomew does need electricity though. He does. He sends us messages, and sometimes the message will be, I’m stuck and I need to power up.
Speaker A: So needs and desires are different, right? So electricity is a need, but what is it to want it as opposed to to need it. So need is, you know, don’t have power, can’t work, goal, get power. Here are a number of things to do. One, ask for it. Two, have a pattern action program which shows you how to go and identify sources and plug yourself in. But what is it to want? What is it to have an unfulfilled desire? What is it to yearn for something that isn’t met? That’s what it doesn’t have. There’s a little bit of adjustment as you get into the symbiotic relationship with these devices. Augmentation is good, but then we should decide whether we want to be so augmented. But equally, we have to judge algorithms. So, you know, while it seems to be the machine driving this person, you know, to treat them like a slave and at their mercy, you get, you’ve got to get higher up in the food chain of the programmers and say, hey, you know, there’s a bug in your system, you’re actually not, not doing it right because you failed to take this into account.
Speaker B: But in this gig economy that people are talking about, you— the desire from the supermarket is running the algorithm, which is then dictating the lives.
Speaker A: Sure. So we’re—
Speaker B: we’re all—
Speaker A: again, we’re outsourcing some of our complex decision-making. Uh, we started to do that when we used calculators instead of our our minds and our fingers. And now what we’re doing is complex decisions are made where the algorithm decides buy or sell or close a company or give someone a mortgage or don’t give someone a mortgage. And the thing about that, the interesting thing about that is that they don’t reveal to us the basis on which they’re making that decision. So these deep learning algorithms, and deep learning just means learning by example. Hundreds, thousands, millions of examples. But when they learn by example, it’s not transparent. We can’t quite see inside the workings. If you lift the hood, you can’t see the basis on which it’s making that decision. But we are the ones at the end of the day who have to train the machine by telling them whether it’s getting it right or whether it’s getting it wrong. So, you know, a bank who runs some deep learning algorithm to take people’s complex set of data and decide whether they’re too risky to give a mortgage to or not. It’s gone through thousands and thousands and thousands of examples. But, you know, when it’s, when they’re training it, they’re looking and saying yes or no to whether or not these particular cases are going to be cases where humans would give the mortgage or not give the mortgage. And then they make sure the machine’s trained up on those, and then it extrapolates beyond the cases we’ve come to. So yes, those AIs are making complex decisions, but we’ve told them how to shoot and how to point and shoot, as it were. And they’re the ones that are then using our goals to get better at making those decisions.
Speaker B: Bartholomew has been training me though. I’m one of these chaotic people who has piles of paper and books all over the place. And as he’s been going around, he’s been having problems with those. So I’ve been forced to actually pick things up off the floor.
Speaker A: Well, that’s very accommodating of you. I would just, uh, I would just reprogram him to say, if there’s a pile of papers or books, it must be important enough to leave it there, and you’ll disrupt this man’s life terribly if you rearrange them. As an academic, I know that all too well.
Speaker B: Is this the thing then, Barry, is that we’re in this cooperative relationship with our technology. I mean, it’s, it’s this thing that we don’t seem— we don’t quite seem to have worked this balance out about in whose interest is it. Um, I spoke to somebody from IBM earlier today. They were saying that they think that the relationship should be one of augmentation. Um, some people do seem to be pushed around by these machines. They’re There was a case of a delivery driver in— or delivery bike person in Brighton who was being pushed around by his algorithm. The algorithm wasn’t taking the hills into account, so it was making him work quite hard. To the people who are working in the factory, it’s saying we only need you for a certain amount of time because we only want X amount of pots.
Speaker A: Yes, I mean, I think this is, this is This is the cruelest part of it when the kind of fine computation and calculation about how much we need human beings and how much we don’t need them is driven to an extent that the person at the bottom, not knowing their fate, not understanding why they’re being driven hither and thither, is actually at the mercy of this system. So, but the thing about the gig economy that’s wrong is not necessarily the algorithms that program the routes and the delivery times and all the rest. It’s the clever lawyers who decided to give contracts where there’s no sick pay, holiday pay, no time off, no negotiation, no rights of various kinds that are usually enshrined in employment law because these people are said to be working for themselves and so on. So I think, you know, some of the most malicious uses of technology Look harder and you’ll find a human being using the technology to hide behind.
Speaker B: The philosopher Barry Smith. Now, let’s look at another issue. What happens when we feel poorly but we don’t know whether we should call in sick? In the new world of work, that might not even be an option, as the delivery driver Don Lane found when he tried to take time off work to see doctors about his diabetes. And in the new world of healthcare, it could even be an AI app that diagnoses you. Florian Bundrup is working on the Dossier application.
Speaker H: Okay, so we have fever, whatever, running nose, sore throat, and it actually will now, now ask you what have you additional symptoms. Okay, is it sneeze, a red throat, pain while swallowing, impaired smell, itching of the nose or throat, or none of the above.
Speaker C: Ah, well, I haven’t looked into her throat, but she has been sneezing. Okay, so sneeze, yes, for sure. And pain while swallowing, yes.
Speaker H: Okay, I will select those. But you already see that it will actually really go deep, and well, it obviously knows if you certain symptoms, certain other symptoms might appear or might not appear. It will actually ask about both to include or disclude certain illnesses.
Speaker C: That was a very good demonstration of how, uh, how quick it is.
Speaker H: Yeah, actually.
Speaker C: And I guess if you have all the symptoms of a particular syndrome or illness, then it will tell you so right away.
Speaker H: Yeah, it It will tell you— it typically needs between 10 and 30 questions, really depending on what kind of symptoms you have and what kind of illness that it tries to assess and how severe those might be. And this whole engine is certified as a medical product, Class I, which is important if you want to use it in the public healthcare system, obviously, and to also provide that trust that is needed for patients for an insurance company to really use these kind of technology. You can always improve. Class 1 is not the very, not the highest. You shouldn’t only rely on it, but it can give you a first estimation of what you might or what you shall do or what options you might have.
Speaker C: I can imagine though that some of the most worrying healthcare situations involve children or babies who can’t tell tell you how they’re feeling. Have you factored in any kind of safeguards against parents or childminders assessing things wrongly from people who, for whatever reason, because they’re too young, because they’re too old, maybe because they have communication difficulties or language problems?
Speaker H: Those are the most difficult ones because you can access even less symptoms than you can for yourself, and even those are corrupted because you don’t notice everything about yourself as an external viewer, a doctor, would do. So you always have compromise here, but it will still ask you a couple of questions, still trying to assess the severeness of those symptoms that you at least mentioned. And as you have seen in the application, in the example, you are allowed to say ‘Well, I don’t know.’ And for, for kids, obviously for some situations you will just won’t know and will then anyway determine whether it is severe or not. But if something is really wrong with your child, I guess not all parents will just stay home if an application will tell them, ‘Hey, stay home.’ So it is still allowed and it should be allowed, and that’s actually something we have to worry about.— it still should be allowed to distrust technology and anyway consult with a doctor. And this is something where we have to have a debate about. Once these algorithms get better and maybe are publicly recognized as being valid all the time, if people that trust technology less than the average, are they still allowed to distrust trusted? Do they have to pay extra? It’s something that we don’t speak about. We speak about, oh, it will cost jobs and everything. I mean, in healthcare it won’t cost jobs because you still need, like, the pill. You still need to be treated, and we don’t have enough doctors anyway. So doctors, guys, you’re fine, except you work in radiology. But we have to actually have a debate about how we use this technology and how we deal as a society with it. Can we take it for granted? Do we force others to do so? And imagine a robot that operates you. Scary feeling in the first place, but now let’s think 5 years ahead. This robot, we have a study, an actual, actually independent study that validates this robot is right 98% of the time and the human operators is right 93% of the time. People will change their minds about the robot now. They will ask to be operated by a robot, but some people will choose the human. But what about those 5% errors? So those 5% might have disability afterwards because these mistakes will injure them permanently. Who pays for that? It’s the same discussion like an autonomous car. At the point where we have a lot of autonomous cars, do those people, those humans who drive the car, have to pay extra because they are a risk? These are new questions. Interesting questions, not only about whether AI is reliable or safe. It will be safe at some point. It will get more reliable. Hopefully, in the time— at the time it is reliable and safe, we had a debate about how we use it, because we don’t do it right now.
Speaker B: What Florian said there is quite scary when you think about it. A robotic surgeon doesn’t make mistakes. So if you want to be operated on by a human without robotic assistance, you might have to pay to take out an expensive additional insurance policy. And what happens if the health app can talk to the job app? Will you be given less work because the algorithm knows you’re not feeling 100%? Or no work at all, since it’s not efficient to send delivery jobs to someone who will struggle to accomplish them on time because of illness. The whole debate about robots is beginning to make people feel uneasy about their deployment. That underlies a backlash dubbed the Techlash, something that Dr. Ben Lorica, a top data scientist who works for the AI debate company O’Reilly, says is very real.
Speaker I: I, I definitely share your concerns, but I think there’s some glimmer of hope because basically there’s more— it seems like— I don’t know how you feel, but there seems to be more awareness on the part of consumers. So they are— and also regulators. So as far as the use, collection and retention of data as one example. So there seems to be more awareness. And so I think the more people like you write about these things, the more consumers are able to make informed choices. I think for the most part, until kind of this new— in the US we even have a term for it, techlash— until this recent backlash, people were just using things without even thinking about the implications of what’s happening, photos they’re uploading and things like this. But maybe now there’s more awareness. So that’s the optimistic perspective. On the other hand, there’s so much— people have formed habits as to how they interact with the systems that we may need to mount a massive public education campaign to remind people about what’s happening, right? So, and maybe that’s what this tech clash will do if, as long as the regulators and politicians take advantage of this moment in history instead of grandstanding, right? So that they can probably, in the process of of, of investigating some of these companies, they can actually educate the general public as well.
Speaker B: It is interesting, isn’t it? Some research has come out, um, that research seems to indicate that young people in their teens seem to be suffering from this backlash. People for 50 years plus, maybe 40 years plus, are also sort of now, because it’s actually informed their prejudices because they didn’t like technology anyway. Whereas that millennial group, as you’ve said, who habitually become used to using technology, they’re not affected. But I do find that that’s interesting. But to go to another one of your points, the regulators didn’t take any notice of this until they thought that the Russians had been interfering with the American election.
Speaker I: But it’s still a good moment in history because now they’re maybe more aware. So hopefully that this will lead to more awareness in the public, right? So I’m not sure about what will happen in terms of regulations and legislation, but maybe in the course of this— because you’re— at least I’m reading more articles about it.
Speaker B: Lorica says that the solution has to be to include more ethicists, writers, and philosophers in AI’s development?
Speaker I: Probably, yeah. ‘Cause I mean, I think that, so there’s also a lot of researchers looking at fairness and things like this, but then, you know, you can have like a set of guidelines that you have to follow in order to make sure that your machine learning learning model is fair and not biased. But the larger question of ethics in AI, I mean, there’s no checklist. It’s ethics.
Speaker B: So at last, a career path opens to all of those philosophers.
Speaker I: You know what I mean? So you gotta— you have to have people who are knowledgeable on your team in order to— it’s not, it’s not just pure technology, right?
Speaker B: So interdisciplinary. Extraordinary. So really, all of those philosophers who are studying Kant should be looking—
Speaker I: So speaking of career paths, if you talk to people in the chatbot industry, they’ve come to really appreciate English majors because it turns out that, you know, these conversations, in order for you to have really engaging conversations, you have to have good writing, right? Yeah, yeah, yeah. So, so there’s some— I’ve talked to some people in the chatbot industry, they’re employing English majors.
Speaker B: Ironically, it was a writer who originated the idea of the robot in the first place. It comes from a Czech language word for a worker, and believe it or not, that first appearance of a robot in the world’s mind was almost exactly 100 years ago when The book Rossum’s Universal Robots by Karel Capek was published on January 21st, 1920. Karel Capek, an anti-fascist whom the Nazis named as their public enemy number 2, turned the book into a theatre play, a fictional invention. It’s eerie to hear how much of it has come true., and if you listen to the start, you will hear similarities to Ridley Scott’s Blade Runner, itself based on Philip K. Dick’s Do Androids Dream of Electric Sheep? So very likely an homage by Dick.
Speaker J: Anyone who has looked into human anatomy will have seen at once that man is too complicated, and that a good engineer could make him more simply. So young Rossum began to overhaul anatomy and tried to see what could be left out or simplified. Young Rossum said to himself, a man is something that feels happy, plays the piano, likes going for a walk, and in fact wants to do a whole lot of things that are really unnecessary. Oh, that are unnecessary when he wants, let us say, to weave or count. Do you play the piano? Yes. That’s good, but a working machine must not play the piano. Must not feel happy, must not do a whole lot of other things. A gasoline motor must not have tassels or ornaments, Miss Glory. And to manufacture artificial workers is the same thing as to manufacture gasoline motors. The process must be of the simplest and the product of the best from a practical point of view. What sort of worker do you think is the best from a practical point of view? What? What sort of worker do you think is the best from a practical point of view?
Speaker C: Perhaps the one who is most honest and hardworking?
Speaker J: No, the one that is the cheapest. The one whose requirements are the smallest. Young Rossum invented a worker with the minimum amount of requirements. He had to simplify him. He rejected everything that did not contribute directly to the progress of work, everything that makes man more expensive. In fact, he rejected man and made The Robot.
Speaker B: That was an extract from Rossum’s Universal Robots performed by volunteer readers for LibriVox. It’s in the public domain and you can hear the full version on YouTube. Tragically, Karel Čapek died of pneumonia, but the inventor of the word and the concept of the robot was Karel Čapek’s brother Joseph, who died in the Nazi concentration camp Belsen So we’ll never know what he thought of the embodiment of what he imagined. Robota, a Czech word for drudgery and the work of serfs or slaves. It’s something I’ve been watching firsthand with a robot vacuum cleaner I was lent by a company called Neato, and the experience has been a little uncanny. You can give your Neato cleaner a name. Why? I don’t know, but I did. It’s now Bartholomew and is very good. The rooms it is in are now very clean, and somehow or other it’s uncannily got me going around cleaning up piles of papers from the floor. It’s tidying me up. We’ll keep wrestling with the robots for better or for worse, and you can read more of our deliberations at our new improved website www.futureintelligence.co.uk. Password was written by Jane Wyatt and Peter Warren and produced by Blue Buffery, and Peter Warren also presented. We’ll be back next month with a new edition produced entirely by human hands. Thanks for listening.
Speaker A: Goodbye. This program has been brought to you by Resonance 104.4 FM. If you liked what you heard and want to support our work, please make a donation at fundraiser.resonance.fm.
