The Monthly Briefing

No one can guess the future



We all make predictions on a daily basis. We predict how long it will take us to get to the airport, when we will finish a task, or what our savings will be at the end of the month. These predictions may seem more trivial, but we also foresee extremely important scenarios, such as those related to public health or macroeconomic issues. We do this because putting ourselves in the future helps us make decisions in the present. Even though we know that predictions often end up being unfulfilled.

Artificial Intelligence allows us to make far more reliable predictions, and, on many occasions, even accurate ones. Probably the best predictions we have ever been able to make. But as with any prediction, the value offered is never totally certain; the uncertainty involved in any estimate must be taken into account.

Hence, we are actively introducing models for this kind of uncertainty during the development stage of our data-based products, such as the BBVA's app feature that shows the user their estimated balance at the end of the month. In the 2020 edition of the KDD Workshop on Machine Learning in Finance we have presented one of our latest solutions which uses our model with uncertainty to detect unexpected ‘spikes’ on balance series that may point to relevant changes in the customer’s situation or events that require immediate attention.

To do that, our engine offers a prediction of what the customer’s balance could be for each of the following 30 days. However, instead of simply outputting a point estimate of the balance, the uncertainty model allows us to better understand the confidence in the balance prediction and to build other interesting applications. We also may send direct notifications to the customer in order to report relevant changes in their balance.

In order to better explain how the model works, we have followed the predictions of a real case. In the figure below we can see how the engine generates the confidence bounds for each day of the month, that is, the ranges in which the series has a high probability of taking the future values. When the actual value is outside these bounds, we can say that a very unusual variation has occurred. Watch the full video.

In this article our colleagues David Muelas, Luis Peinado and José A. Rodríguez tell more about this particular case and how the model works. Check it out!


US Election: has the result forecast failed (again)?

Jimmy Chan/Via Pexels.

After the failures in polls and forecast models in the 2016 US election, leading public opinion research and media companies, such as FiveThirtyEight, set out to fix the issues for this year's election.

However, although this time it can be said that they have predicted the results better, they still fall into the same errors of several of the country’s states and among some specific demographics. This PewResearch article reviews a number of these major errors, and launches some hypotheses and possible solutions. Among the most important aspects, it highlights the non-participation in the polls observed especially among Republican voters or the "shy Trump" phenomenon, by which their relatives are more reluctant to reveal their voting preferences. Also of concern is the under-representation in the polls of certain demographics, such as non-white college voters or hipanics.

On the other hand, this FiveThirtyEight article is tremendously interesting to better understand how these forecast models work, which take data from polls but also from economic or demographic indicators, among others.

Via FiveThirtyEight.

Nate Silver, founder and editor in chief of FiveThirtyEight and the author of The Signal and the Noise: Why So Many Predictions Fail — But Some Don’t, explains in detail all the variables that are taken into account and what steps they take in building their election forecast, such as analyzing polls, and then combining them with demographic and economic data, or the simulation phase.


Further reading

+A very interesting report on the state of AI in 2020 and the most important advances and trends of this last year. (stateof.ai)
AI investors Nathan Benaich and Ian Hogarth analyze in this comprehensive report the state of AI, from its research and talent implications, to issues related to industry trends and regulation. The authors highlight advances in NLP and AI applied to biology, as large companies continue to make major advances in their AI hardware platforms. Talent continues to be concentrated in North America, although the US AI ecosystem is fuelled by foreign talent. Topics such as ethic and fairness come now into sharp focus. You can also check the 2019 and 2018 reports.

+The Government of Spain presents the first version of the Charter of Digital Rights (El País)
Borja Adsuara, one of the experts involved in the development of this document, says that it is not a normative text and prefers to call it a 'Roadmap'. In this article (in spanish), Adsuara explains the aim of this text: to adapt to the digital environment the rights that are already recognized in the Constitution, and to consider new ones, such as the digital identity right or pseudonymity right, as well as to collect new mechanisms to guarantee existing rights. Here you can check the document for public consultation.

+AI can make bank loans more fair (Harvard Business Review)
According to Sian Townson, "the key to ensure that biases of the past are not baked into algorithms and credit decisions lies in building AI-driven systems designed to encourage less historic accuracy, but greater equity. That means training and testing AI systems not merely on loans or mortgages issued in the past, but instead on how the money should have been lent in a more equitable world."

+ Where we live? A series of stunning satellite images of our cities (Planet | Medium)
Planet’s constellation of 13 SkySats offers greater flexibility in showcasing the planet from all its glorious angles.

Doha, Qatdohaar. November 11, 2017 | Via ©2018 Planet Labs, Inc. cc-by-sa

+ More references for the debate on Artificial General Intelligence (AGI). (Togelius | Better playing through algorithms)
Julian Togelius deepens in this topic from the point of view of philosophy, psychology and algorithm development work. Plus, he includes many references that enrich the dissertation. "I think that the discussions about AGI, superintelligence, and the intelligence explosion are mostly an artifact of our confusion about a number of concepts, in particular, intelligence. These discussions are not about AI systems that actually exist".

+ Geoff Hinton: “Deep learning is going to be able to do everything” (MIT Technology Review)
Geoffrey Hinton, winner of the Turing Award in 2018 together with Yoshua Bengio and Yann LeCun for their work in deep learning, talk about his impressions about this technique.

+ Understanding RL Vision, a new Distill paper (Distill)
In this article, Hilton, et al. apply interpretability techniques to a reinforcement learning (RL) model trained to play the video game CoinRun. Using attribution combined with dimensionality reduction, we build an interface for exploring the objects detected by the model, and how they influence its value function and policy.

+ Landmark Papers in Machine Learning, by Dan Turkel (github)
The document attempts to collect the papers which developed important techniques in machine learning. By Dan Turkel.

+ The Data We Do Not See (Nightingale | Medium)
Alyssa Bell interviews Giorgia Lupi, one of the best and award-winning information designer, and she talk about embracing complexity and the future of data.

A dialogue between four hands, a collaboration with Kaki King.

+ The Pope Francis also joins the fever of Artificial Intelligence (The Pope Video)
The Pope asks that progress in robotics and artificial intelligence “be human”. Francis emphasizes the need to orient it “towards respecting the dignity of the person and of Creation.”

Quote

of the

month

"Creative professions are some of the only fields that will withstand the rise of the robots and we need to do much more to ready ourselves, our companies, and our children for the creativity-focused future of work."

Scott Belsky, Chief Product Officer at Adobe and founder of Behance


In an article published in World Economic Forum, titled Creativity will be key to competing against AI in the future workforce - here's how, Belsky thinks about the changes driving AI and automation in the work of the future. The pandemic has accelerated this process and may offer us an opportunity to better understand some elements of this not so distant future.

According to Belsky, the creativity-based jobs will be key in the future and he proposes to apply a new thinking to three critical areas: education, retraining and workplace tools.

BBVA DATA GALAXY 🌌

  • What should be taken into account if Artificial Intelligence is to be regulated?
    In this article, Juan Murillo, Senior Manager of Data Strategy at BBVA, and Jesús Lozano, Manager of Digital Regulation at BBVA, analyze the potential implications of Artificial Intelligence regulations and share their insights into the considerations that should be taken into account to ensure that regulatory aspects support the proper development of this discipline in the future.
  • How do you win the BBVA Hackathon 2020? The winners tell us all
    Last October was held the Hackathon BBVA, a competition in which more than 700 young people from Mexico, Peru, Colombia and Spain took part and used their talents to solve 12 digital transformation challenges in banking. The winning team presented a solution to analyse comments about BBVA on social networks, which they launched in response to the Social Listening challenge.
  • What is GPT-3? The AI that will write for you (in spanish)
    Our colleague César de Pablo tells us about the latest offerings from GPT3, the AI model developed by Open AI that allow you to generate written text.

Keep moving forward

The most important thing about great achievements is the path travelled. Thanks David Alameda for sharing! ❤️

See you folks!

< Read previous issue

For any question or suggestion, you can also write to hello@bbvadata.com

You can enjoy much more content related to data science, innovation, new solutions of financial analytics and how we work in our website: bbvadata.com

Let's talk about it. Join the conversation on Linkedin.

© 2020 All rights reserved. BBVA Data & Analytics. Avenida de Manoteras, 44, 28050. Madrid.