#87 Hannah Fry: The Role of Algorithms

Jul 7, 2020
Overview

Mathematician Hannah Fry discusses the pervasive role of math in modern society, from engaging students to understanding algorithms and human behavior. She explores how mathematical concepts can offer insights into dating, relationships, and critical decision-making, particularly during a pandemic.

At a Glance
13 Insights
52m 1s Duration
14 Topics
3 Concepts

Deep Dive Analysis

Early Interest in Mathematics and School Engagement

The Invisibility of Math in Modern Algorithms

Humanizing Mathematics Through Stories and People

Being Human in the Age of Algorithms

The Dangers of Algorithms Without Human Context

Algorithms in Medicine: Challenges and Ethical Considerations

Transparency and Regulation of Algorithms

When Algorithms or Humans Should Make Decisions

Math's Role in Pandemic Decision-Making

Understanding and Communicating Exponential Growth

Kasparov vs. Deep Blue: Human Weaknesses and AI

Applying Math Concepts to Human Relationships

Optimal Stopping Theory for Dating

Mathematics of Arguments in Relationships

Exponential Growth

This describes a situation where something changes by a fixed fraction in a fixed period. It is often counterintuitive because it leads to outcomes that are far larger than typically imagined, such as a virus doubling every few days, quickly leading to massive numbers.

Optimal Stopping Theory

A mathematical problem for making decisions when faced with a sequence of opportunities, where the goal is to stop at the 'perfect' time. Rules typically include not being able to go back on a rejection or look ahead after making a choice.

Negativity Threshold

In the context of relationship dynamics, this refers to how much annoyance a person can tolerate before reacting to their partner. Research suggests that couples with low negativity thresholds, who address small issues quickly, tend to have better long-term success.

?
How can schools better engage students with math?

Schools can promote better engagement by demonstrating the real-world usefulness and dramatic importance of math in virtually every aspect of the modern world, which can make the subject come alive for students.

?
Why is math often invisible despite its prevalence in modern technology?

For modern technology like mobile phones and algorithms to work effectively for the user, the underlying complex mathematics must be hidden behind the scenes, making it invisible to the average person.

?
How can math be humanized to make it more interesting?

Math can be made more interesting by anchoring its stories to the people involved, showcasing the human element, passion, and struggles of mathematicians throughout history, much like how human stories captivate audiences in other fields.

?
What is the danger of designing algorithms without considering their human integration?

The danger is that algorithms, when built in isolation and then planted into society, can lead to catastrophic consequences, including perpetuating biases (gender, racial) and causing humans to blindly defer to machines, abdicating critical thinking.

?
When should algorithms make decisions versus humans?

Algorithms are ideal for consistent, clear decisions in critical systems like nuclear power stations or flying airplanes, where human inconsistency is a risk. However, humans are crucial for social decisions where algorithms can make catastrophic mistakes, and human oversight is needed to prevent blind deference.

?
How does math help in understanding and responding to pandemics?

Math, particularly through epidemiological models and data analysis, acts as a crucial weapon by predicting future spread and guiding government policies and strategies when pharmaceutical interventions or vaccines are not yet available.

?
Why is exponential growth difficult for people to grasp?

Exponential growth is counterintuitive because it doesn't just mean 'big' but specifically refers to something changing by a fixed fraction in a fixed period, leading to outcomes that are 'beyond imagining' and often underestimated by human intuition.

?
What are the challenges of using algorithms for medical diagnosis?

Algorithms can make 'stupid mistakes' by picking up on irrelevant factors (like the type of scanner or a ruler in a photo) rather than the actual medical condition. Furthermore, being too good at detecting all cancerous cells could lead to unnecessary and invasive treatments for benign conditions that would never become problematic.

?
Should algorithms be open source for transparency?

While transparency is important, open-sourcing proprietary algorithms might be 'too much and too little.' It's too much because understanding the code requires vast technical knowledge, and too little because it could stifle innovation by removing commercial viability. An FDA-like regulatory body might be a better alternative for oversight.

?
How did Deep Blue defeat Garry Kasparov in chess, beyond just computational power?

Deep Blue exploited Kasparov's human weaknesses by being deliberately coded to introduce random delays in its responses, making it appear to be deeply calculating. This tactic 'psyched out' Kasparov, causing him to second-guess the machine and ultimately leading to his defeat.

?
How can math help in dating and relationships?

Math can offer insights into various aspects of love life, such as determining the optimal number of people to date before settling down, understanding which online dating photos work best, designing wedding seating plans, and analyzing the dynamics of arguments in long-term relationships.

1. Low Negativity Threshold in Relationships

For long-term relationship success, address minor annoyances and issues quickly and directly as they emerge, rather than letting them fester. This ’low negativity threshold’ approach allows for continuous repair and prevents bottled-up frustrations from escalating into larger conflicts.

2. Optimal Dating Strategy: 37% Rule

Apply the ‘37% rule’ to dating: spend the first 37% of your dating life exploring without commitment. After this initial period, choose to settle down with the very next person encountered who surpasses all previous partners in suitability.

3. Balance Human and Algorithmic Decisions

Understand that algorithms are superior for consistent, precise tasks, but human oversight remains critical for complex social decisions where machines can make catastrophic errors. Always keep a human in the loop for decisions with significant societal impact.

4. Resist Algorithmic Responsibility Abdication

Actively resist the human tendency to abdicate responsibility by blindly following algorithmic recommendations, as this can lead to cognitive shortcuts and poor decisions. Design systems to prevent this by requiring human engagement and critical review at the decision interface.

5. Integrate Algorithms Thoughtfully

When designing or deploying algorithms, always consider their integration with the human world and society, rather than evaluating them in isolation. This prevents catastrophic consequences by anticipating how people will interact with and be affected by the technology.

6. Design Algorithms for Human Oversight

When designing algorithmic systems, provide users with multiple options and a clear opportunity to review or overrule the machine’s suggestions. This approach prevents blind trust and allows for human sanity checks, mitigating potential errors.

7. Make AI Systems Interrogatable

Design AI and machine learning systems to be transparent and allow human experts to interrogate their decision-making processes. This ‘opens the box’ beyond a simple yes/no answer, enabling professionals to understand and verify the AI’s conclusions.

8. Regulate Algorithms via Independent Boards

Support the establishment of independent regulatory bodies, akin to the FDA for pharmaceuticals, to thoroughly interrogate, stress-test, and approve algorithms. This approach ensures accountability and addresses transparency concerns without hindering innovation.

9. Beware Algorithm Over-Detection

Exercise caution when deploying algorithms for detection, particularly in critical fields like medicine, as being overly sensitive can lead to the identification of benign issues. This can result in unnecessary and invasive treatments for conditions that pose no real threat.

10. Grasp True Exponential Growth

To better comprehend rapidly changing phenomena, understand that exponential growth means a fixed fraction change over a fixed period, not merely a large increase. This counterintuitive concept is crucial for accurately anticipating future developments.

11. Humanize Complex Subjects

To make complex subjects more engaging, connect them to human stories and the people involved in their development or application. This approach leverages our natural inclination towards narratives, making the subject more relatable and alive.

12. Show Subject’s Real-World Utility

When teaching or learning a complex subject, actively seek and demonstrate its practical applications and importance in the modern world. This approach can make the subject more engaging and relevant.

13. Daily Practice Builds Enjoyment

Consistently practice a subject daily, even for a short period (e.g., one page of a textbook), to significantly improve your understanding and skill. This improvement fosters enjoyment, making the subject feel less like hard work.

You can't just build an algorithm, put it on a shelf and decide whether you think it's good or bad completely in isolation. You have to think about how that algorithm actually integrates with the world that you're embedding in.

Hannah Fry

To fly a plane, you need three things, a computer, a pilot, a human and a dog. And the, the computer is there to fly the plane, the human is there to feed the dog, and the dog is there to bite the human if ever it touches the computer.

Hannah Fry

I always think, so I'm a big fan, actually, of Formula One. And the reason why I like it, if I'm honest with you, is because I think of it as a giant maths competition, just with, you know, a bit of glamour on top.

Hannah Fry

The people who have the best chance at long-term success are actually the people who've got really low negativity thresholds.

Hannah Fry

I think that judges are a lot more like them [Japanese tourists who blindly followed a satnav] than we might want them to be.

Hannah Fry

Optimal Dating Strategy (Optimal Stopping Theory)

Hannah Fry
  1. Spend the first 37% of your dating life having a nice time and playing the field, not taking anything too seriously.
  2. After that period, settle down with the next person who comes along that is better than everyone you've seen before.
37%
Optimal percentage of dating life to spend playing the field According to optimal stopping theory, before settling down with the next person better than all previous ones.
5 days
COVID-19 virus doubling time Rate of spread mentioned at the time of recording (March 18th, 2020).
18 quintillion
Grains of rice on a chessboard (exponential growth example) Total grains needed by the end of the chessboard if starting with one and doubling each square.
3 kilometers high
Height of rice stacked over the area of Liverpool Equivalent volume for 18 quintillion grains of rice.
19 years old
Age of Christopher Drew Brooks when arrested Example used to illustrate algorithm bias in judicial sentencing.
18 months
Jail time recommended by algorithm for Christopher Drew Brooks Based on the algorithm assessing him as high risk due to his age and offense.
36 years old
Age at which algorithm would consider Christopher Drew Brooks low risk Despite being 22 years older than the victim, the algorithm would recommend no jail time due to its weighting of age.
2011
Year Hannah Fry finished her PhD and started police project First project was a collaboration with the Metropolitan Police in London after the riots.
4 days
Duration London was on lockdown during 2011 riots Context for Hannah Fry's work with the police on predicting unrest.
300 meters
Distance Japanese tourists drove into the ocean Example of blindly following a satnav without human sanity checks.
150 deaths
Approximate COVID-19 deaths in the UK Number mentioned at the time of recording (March 18th, 2020) to illustrate early stages of pandemic.
2018
Year of BBC project to collect pandemic data Teamed up with epidemiologists to gather data on human movement and contact.
2006
Year of previous UK paper survey on human movement data Prior to the mobile app, this survey of 1000 people was the best available data.
24 hours
Duration people were tracked by BBC mobile app Volunteers allowed tracking to provide detailed data for pandemic models.