Prioritizing Problems and 100 episodes

Talking Machines


You are listening to talking machines. I'm Katherine Gordon Lawrence and from a closet during shutdown of the universe this is. This is how we're talking to each other this the for this episode. Neil and this is our one hundredth episode a very auspicious episode. Eight a marvelous case. I just want to be very clear. I'm not in a closet with Catherine. That would be infringing physical distancing rules. I think as we're now calling it. I am physically distant from Catherine but Katherine is in a closet and I can see the coats hanging on her head. Riches a good. Look actually a hundredth episode. That is exciting. Is it my hundred episode? I think it's just one hundred episode episode. Neil I'm sorry one hundred. I think I'm very healthy. Fifty eight or something like that. I think you're I think I believe you. Actually you're at fifty nine. Wow that was great. Guess Look at me and my capability in a summation very good. Yeah I guess the first question I wanNA ask you really is. How are you is everything? Okay where you are. How are things going before we talk about what we usually talk about? Let's check in with each other as people. Yeah what Crazy Times. So things are fine but obviously naught fine in the UK at the moment. And you know I think is going to be something that plays out over a long period of time but My wife's Italian things a a a bad in Italy And they're a bit further down this road than we are so they could become quite bad hair and I think there's a lot of apprehension about them but I also noticed that different groups seem to have different different levels of concern about this. I don't fully understand those different levels of concern but you certainly see the sudden people out in sort of light touching each other and then the rest of us going. Ron Ole giving them know what we do is we give them dirty looks because we in Britain. And that's how we in fools Social distancing which as I was saying I think perhaps physical distance things the right thing because we need to be social we dole out dirty looks that seems to be how we dealing with corrupt. I'm not sure as a sufficient response but I don't suppose it surprised me but you notice how widespread lack of understanding of exponential say is which I think is natural someone was. I watched a video. Where an Italian guy who named forgets me name forgets me and I forget his name actually in this case. I've forgotten how to speak so it touting I think Delana is his name was talking about how difficult it is to express an exponential in normal times that it's concept that is difficult for our brains to handle and I think perhaps many of us in machine learning much more used to thinking about quite used to it but I was really reflecting on that and thinking. Yeah it is hot and I think the difficulty of handling experts specs potentials combined with uncertainty around. What's going on? Is You know. Really at the heart of the challenges facing for this epidemic. Yeah absolutely is there. Been any interesting work by anyone that you've seen about predicting the impact or the. I think the modeling work is is very interesting. I would not want to be doing it. I was would be but I don't envy those that are having to do it. Because I work. In galveston proceeds occasion. I worked with the Padilla millages and was organizing conference in Sheffield and I remember one of the speakers because we were recording everything for the cameras to be switched off for period. Because they were about to. Present some speculative results on malaria. I think in East Africa and they didn't want those results to be broadcast until they confirmed that modeling because the implications for those people who are funding say mosquito nets and other interventions Quite serious so they didn't want misinformation to get out there and I think that this is a really interesting challenge for the machine learning community where our instinct is to sort of share things very widely quite early and that's a good instinct but this is difficulty the sharing Wash conclusion people are reading and interpreting this and of course that is going on quite a widespread level. So I think it's very interesting around open science like festival you use. It wants to understand the models people using and how they're doing things but secondly you don't want people to get the wrong impression by misusing model models only exists with the context and I think that's one of the biggest challenges of modeling is. Is You have to deeply understand the limitations so tools. I Actually Matthew House from University of Manchester shed a tool that was just stimulating. Effects of changing. All our is the number of expected number of people you expect to transmit the disease to and it was a differential equation model and he shed. It's on python is quite simple model and actually the conclusions you could draw from playing with similar to those of very if people are interested in this for the UK and us. Imperial College has released. A model is being widely looked up and has in theory changed the strategic approach of the UK government. But I'm really uncertain about whether that's true or not. It's whether it changed their approach or whether the the circumstances change their approach. It's so hard to know so people are calling clearly for openness and so the parties in favor of that pardon me then on the stance that some of these models need to be used within a certain context and I think the thing that never fails to impress me as how Cetin so many people are about what the right thing to do is or what the right conclusion is. I just can't be certain. I I feel that you look at these things in the decisions we're making They're gonNA have serious downstream consequences You know for a long time from now. Not just in the here and now the next month's but they have very serious downstream consequences riddled with uncertainty because the major events. Like oh it turns out the drug. Kills it straightaway? Awfully changes thinking. It turns out that you know. Well vaccines I think. The best batches eighteen-month very difficult situation and I almost feel it's dangerous. Even you're talking about in some sense. He's also hard because if you talk about it then your risk of spreading your own misunderstandings or it. It really does make me happy. The we have some understanding of uncertainty a lot of us in the community. Because you study that and I it's always interesting to be to what extent you can try and deploy that in your own decision making people you know Being faced with pretty serious questions like they may have An elderly relative with conditions who live some distance away from them and they've got to make a decision about do they visit or not because the potentially endangering that relative by visiting by increased risk of transmission but simultaneously by being isolated themselves and that relative being isolated at a time. When in some sense you kind of want to be close to family. That's not really a trade off and I think that you know in in machine learning and we all these folks Walker. What's the cost for? That doesn't really work. You know and actually way even when you look at these models they are trying to look at things like so the big trade off I think in these models is the threat to health in the short term versus the long term economic damage by being shot down and some people seem to claim that. That's no trade off that you have to save as many lives now as possible and other people play well. Actually you have to think long term about downstream effects. And I think you can make arguments both ways. So I don't think there's any pure ethical correctness here but I think What actually has happened is the world seems to have gone for the last one and you can't do the former on your own. Because downstream economic consequences will happen at a worldwide scale. So I think that that basically we've the default position has ended up from some. How game theoretic way that we attempt to minimise Nieta impact. And that certainly not a wrong decision as far as we know now but you know events may prove things to take different very challenging. I think what's also very interesting and has a big effect on the way the community will end up looking at things in the future. Is the difference with the capabilities. You get in countries like China where those much will and actually Israel. My understanding is. I don't know the details you will you have the security databases in Israel and in China. You have much more fine grain data collecting capabilities which of the things that we will worry about and it turns out in these situations. They may be quite simple tool. It's actually always interested me that tension between how do you maintain the liberty we expect from not having a data overseen

Coming up next