How do mathematical models help predict the future? - with Erica Thompson

How do mathematical models help us make predictions about what will happen? And what happens when those models are biased towards a particular view of the world?

This talk was recorded at the Ri on 20 June 2023.

Mathematical models have shaped our world and continue to be at the centre of everything we do. They became headline news as we tackled the COVID-19 pandemic, and are helping us to regulate an increasingly volatile economy and navigate the uncertainties of climate change.

In this talk, policy and models expert Erica Thompson explains the validity of the models we use: what they are, how they work, and the disastrous consequences when the makers and interpreters of models get things wrong.

Erica Thompson is a senior policy fellow at the London School of Economics Data Science Institute, a fellow of the London Mathematical Laboratory and an honorary senior research fellow at UCL's Department of Science, Technology, Engineering and Public Policy. With a PhD from Imperial College, Erica works on the use of mathematical modelling to support real-world decisions, specifically on the ethics of modelling and simulation. She has recently worked on the limitations of models of COVID-19 spread, humanitarian crises, and climate change. Erica's previous work includes the UK Department of for Energy and Climate Change's Global Calculator project, where she provided the climate science information. Erica lives in West Wales, and is reducing her own ecological footprint to a "One Planet" level by not travelling to conferences and finding other ways to reach people.

--
A very special thank you to our Patreon supporters who help make these videos happen, especially:

modsiw, Anton Ragin, Edward Unthank, Robert L Winer, Andy Carpenter, William Hudson
--

Product links on this page may be affiliate links which means it won't cost you any extra but we may earn a small commission if you decide to purchase through the link.

[stiffrichard]

I work as an operational researcher and you have captured very succinctly the challenges around modelling and simulation. Data science is becoming more and more powerful, but modelling chaotic systems will always need assumptions and appropriate simplification that may not be clear from the data alone

[AnimeshSharma]

Thinking of Model as a Compass, giving us a sense of direction, what a beautiful insight! Thanks for sharing 🙏

[collector]

She clearly is not the most engaging speaker but as a scientist myself I found her use of the term ModelLand very insightful. We are living in reality not in Model Land where assumptions dominate your models behaviour. Far too often I see scientists in my field that just do modeling and trust their results without doubting their assumptions. Great talk I will get a book as well

[HaroldTaearaovia-ykvu]

This is a very resourceful Lecture....I'm a Sports Betting enthusiast and I create models to predict scores and outcomes, you've basically covered everything that involves mathematics which can predict the future...🙏🙏🙏

[peterandrew]

Profoundly compelling and insightful presentation by Dr. Thompson! Thank You. Since everything humans know is ultimately a model, representation, or map and not reality itself, the implications of this highly engaging presentation go far beyond any specific model or category of models. Perhaps one of the implications is that as data scale so must also error scale? Speaks to the humility of acknowledging limits.

[NickWestgate]

How could they expend so much effort on making this video, but get the deinterlacing so badly wrong on the computer graphics? It's such an integral part of the talk.

[ccatctc]

The graph at about 40 mins says it all. It is cost-based, so it gives the financial solution. This might involve biomass burning of forests for example. Truly a counter-productive strategy taking us back to the middle ages, but cost-effective apparently. Environmental accounting would not allow this type of modelling.

[toonmoene]

Thank you - very useful analogies (I'm a meteorologist).

[simonstrandgaard]

Interlaced video is hard to watch. Please pick a non-interlaced encoding next time.

Great talk. Lots of effort went into this.

[nandfednu]

what a gifted speaker, i literally realized it was probably a dog the moment before the reveal and not a second sooner

[jynxkizs]

The right goals are the ones that we don't regret asking the evil genie for.

[ccatctc]

Fire models in Australia are based on expertise of fire scientists and not ecologists who understand the ecosystem to some extent- potentlally a disaster waiting to happen, but you can't get through the sociopolitical modelling wall. Another area people like Dr Thompson could spend some time on (other than the damn financial system) is dark energy. Listened to a very interesting talk by the Perimeter Institute and how dark energy moddelling is done and I'm sure they could use extra input.

[ThatisnotHair]

9:00 —·— 23:15 This is what Π is all about

[JM-jvcb]

So many plan examples can you just explain the topic of modeling in detail with a point

[JM-jvcb]

36 min in and I don’t understand if anything has been said. Look at data take model take date don’t use old date use new date then more date then use it. Very bad her not well done

[Discopuss]

I wish she had spent a bit more time in the lecture explaining how if your model doesn’t predict doomsday from climate change, you don't get funding for future models.

[mikets]

She does not understand that math is not a science but an instrument. The discipline of mathematical modelling is a toolset. A well developed one, with ARMAX, Kalman, observability, controllability, etc - but yet it is only a toolset. If you apply a wrong tool, not based on the underlying science, you'll get "results" which may be arbitrarily far from reality. These "results" will look "scientific" and convincing to uninitialized - but their Science-wise value is zero. There is no (and can not be) "generalized" modelling, and any talks about it have no scientific value. Moreover, any pretenses of modelling non-linear loop-backed unstable stochastic systems are fake.

[robicamedia]

Studies say that it's IMPOSSIBLE, that a model can predict the future, or even the weather. As also been seen on this channel in a lecture. I wish you all a nice day.

[gayan]

Well....Mathematics will not predict the future. Mathematics is already entangled in the future. Future will also be shaped according to acts now. Cause and effects. Surely, if used well, Mathematics will tell you all .not just the future... Future is a very important component but a part when the whole existence is considered.!