What do you think?
Rate this book
Outnumbered: From Facebook and Google to Fake News and Filter-bubbles – The Algorithms That Control Our Lives
Algorithms are running our society, and as the Cambridge Analytica story has revealed, we don't really know what they are up to.
Our increasing reliance on technology and the internet has opened a window for mathematicians and data researchers to gaze through into our lives. Using the data they are constantly collecting about where we travel, where we shop, what we buy and what interests us, they can begin to predict our daily habits. But how reliable is this data? Without understanding what mathematics can and can't do, it is impossible to get a handle on how it is changing our lives.
In this book, David Sumpter takes an algorithm-strewn journey to the dark side of mathematics. He investigates the equations that analyse us, influence us and will (maybe) become like us, answering questions such as:
Who are Cambridge Analytica? And what are they doing with our data?
How does Facebook build a 100-dimensional picture of your personality?
Are Google algorithms racist and sexist?
Why do election predictions fail so drastically?
Are algorithms that are designed to find criminals making terrible mistakes?
What does the future hold as we relinquish our decision-making to machines?
Featuring interviews with those working at the cutting edge of algorithm research, including Alex Kogan from the Cambridge Analytica story, along with a healthy dose of mathematical self-experiment, Outnumbered will explain how mathematics and statistics work in the real world, and what we should and shouldn't worry about.
A lot of people feel outnumbered by algorithms – don't be one of them.
Our increasing reliance on technology and the internet has opened a window for mathematicians and data researchers to gaze through into our lives. Using the data they are constantly collecting about where we travel, where we shop, what we buy and what interests us, they can begin to predict our daily habits. But how reliable is this data? Without understanding what mathematics can and can't do, it is impossible to get a handle on how it is changing our lives.
In this book, David Sumpter takes an algorithm-strewn journey to the dark side of mathematics. He investigates the equations that analyse us, influence us and will (maybe) become like us, answering questions such as:
Who are Cambridge Analytica? And what are they doing with our data?
How does Facebook build a 100-dimensional picture of your personality?
Are Google algorithms racist and sexist?
Why do election predictions fail so drastically?
Are algorithms that are designed to find criminals making terrible mistakes?
What does the future hold as we relinquish our decision-making to machines?
Featuring interviews with those working at the cutting edge of algorithm research, including Alex Kogan from the Cambridge Analytica story, along with a healthy dose of mathematical self-experiment, Outnumbered will explain how mathematics and statistics work in the real world, and what we should and shouldn't worry about.
A lot of people feel outnumbered by algorithms – don't be one of them.
271 pages, Kindle Edition
First published April 18, 2018
112 people are currently reading
1403 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 30 of 61 reviews
May 24, 2018
This book contains some impressive and important content - so I struggled initially to understand why I found it difficult to get on with. More on that in a moment.
Applied mathematician David Sumpter takes apart our current obsession with algorithms, information bubbles, AI and fake news, showing that all too often what we read about it is more hype than reality. Whether he is dealing with the impact (or otherwise) of Cambridge Analytica on elections, or the ability of algorithms to out-think humans, he shows that we have too often assumed that sales pitches were a reality: at the moment AI and its algorithms are rarely as good as we are told.
It might seem that this is the work of an academic with an axe to grind about the other mathematicians who are coining it in, but this is no unsubstantiated polemic. In many cases, Sumpter describes constructing a model to simulate the workings of an algorithm and demonstrates how feeble it really is. It was also fascinating to discover the way that an algorithmic presentation of 'also liked' amplifies (mathematical) chaos to bring out near-random winners - responsible, for example, for those YouTube stars where no one can understand their success.
I absolutely loved one section where Sumpter is trying to assess the intelligence levels of current AIs. Clearly they can't match humans. How about dogs? No. Bees, maybe? No. He shows that in reality, current machine learning struggles to match the intelligence level of an advanced bacterium.
Everything about what's in the book (apart from Sumpter's enthusiasm for football) seems a perfect match for someone deeply interested in algorithms and AI. So why did I not find the book particularly compelling? In part it's because it has quite a dry presentation. Unlike Sumpter's previous title Soccermatics, the style here is very measured and near-academic, presumably to add weight to the content, but the result was that some of it proved a dull read.
It's not all like that, I ought to stress. I loved the line when considering what the Cambridge Analytica model promised: 'Democrats... could focus on getting the vote out among Harry Potter fans. Republicans could target people who drank Starbucks coffee and people who go camping. Lady Gaga fans should be treated with caution by both sides.'
I think the other issue was the 'negatives don't engage' syndrome. While it's important to know that algorithms and AI are far less powerful than we are generally given to believe in the news (and some books), it's hard to get too excited when told about something not being the case. It's a bit like the news headline 'War does not break out.'
The last thing I want to do is put people off this book. It really was interesting to learn how relatively ineffective AI is at this stage of its development, given how much news coverage has been given particularly to Cambridge Analytica, but also to the dark power of algorithms. It's an important message. I just wish the way it was delivered had been more engaging.
Applied mathematician David Sumpter takes apart our current obsession with algorithms, information bubbles, AI and fake news, showing that all too often what we read about it is more hype than reality. Whether he is dealing with the impact (or otherwise) of Cambridge Analytica on elections, or the ability of algorithms to out-think humans, he shows that we have too often assumed that sales pitches were a reality: at the moment AI and its algorithms are rarely as good as we are told.
It might seem that this is the work of an academic with an axe to grind about the other mathematicians who are coining it in, but this is no unsubstantiated polemic. In many cases, Sumpter describes constructing a model to simulate the workings of an algorithm and demonstrates how feeble it really is. It was also fascinating to discover the way that an algorithmic presentation of 'also liked' amplifies (mathematical) chaos to bring out near-random winners - responsible, for example, for those YouTube stars where no one can understand their success.
I absolutely loved one section where Sumpter is trying to assess the intelligence levels of current AIs. Clearly they can't match humans. How about dogs? No. Bees, maybe? No. He shows that in reality, current machine learning struggles to match the intelligence level of an advanced bacterium.
Everything about what's in the book (apart from Sumpter's enthusiasm for football) seems a perfect match for someone deeply interested in algorithms and AI. So why did I not find the book particularly compelling? In part it's because it has quite a dry presentation. Unlike Sumpter's previous title Soccermatics, the style here is very measured and near-academic, presumably to add weight to the content, but the result was that some of it proved a dull read.
It's not all like that, I ought to stress. I loved the line when considering what the Cambridge Analytica model promised: 'Democrats... could focus on getting the vote out among Harry Potter fans. Republicans could target people who drank Starbucks coffee and people who go camping. Lady Gaga fans should be treated with caution by both sides.'
I think the other issue was the 'negatives don't engage' syndrome. While it's important to know that algorithms and AI are far less powerful than we are generally given to believe in the news (and some books), it's hard to get too excited when told about something not being the case. It's a bit like the news headline 'War does not break out.'
The last thing I want to do is put people off this book. It really was interesting to learn how relatively ineffective AI is at this stage of its development, given how much news coverage has been given particularly to Cambridge Analytica, but also to the dark power of algorithms. It's an important message. I just wish the way it was delivered had been more engaging.
September 27, 2018
Entre os melhores livros que li este ano. Sumpter escreveu este livro pós Weapons of Math Destruction, pós o escândalo da Cambridge Analytica e pós eleições do Trump. E faz uma análise bem crítica e atual (em 2018, pelo menos) do que algoritmos têm feito e do que empresas são realmente capazes.
Ele quebra a situação da Cambridge em dois pedaços, o que os donos dizem que eram capazes de fazer e o que podemos realmente fazer. E mostra que tem um exagero dos dois lados, do que prometem e realmente podiam fazer. E da preocupação que temos com este tipo de influência em especial, quando outros pontos são mais preocupantes.
Uma das discussões que mais gostei no livro é do antagonismo entre acurácia e viés/preconceito. Ele dá um exemplo que dominou a discussão de análises, um algoritmo para predição de reincidência de crimes que errava mais frequentemente dizendo que negros podiam voltar a cometer crimes, por ser calibrado para ser preciso. Para em seguida mostrar que, baseado nos dados disponíveis (maior reincidência entre negros), um algoritmo que não errasse dessa forma teria que discriminar entre negros e brancos na hora de fazer a avaliação. E seria menos preciso. Sumpter não deu uma resposta definitiva nem disse o que era o certo nessa questão, só mostrou que o cenário maior precisa ser levado em conta quando fazemos análises.
Outra discussão muito boa foi como alguns tipos de otimização podem criar desigualdades, mesmo sem má intenção por trás. O que cheguei a discutir no IGTV, inclusive. Por exemplo, se alguém otimizar uma propaganda de uma vaga de emprego no Facebook para atingir mais pessoas com um certo gosto (quem assiste Nerdologia, por exemplo), pode acabar apresentando essa propaganda predominantemente para homens. E se estiver otimizando para economizar impressões, isso inclusive pode ser reforçado. Mas a consequência não intencional é que mulheres nem terão a chance de tentar esse emprego porque não verão a vaga.
No fim, achei uma ótima mistura entre conhecimento do assunto e análises de alguém que não está financeiramente envolvido na performance desses programas. Vale para se atualizar sobre o que estamos passando.
Ele quebra a situação da Cambridge em dois pedaços, o que os donos dizem que eram capazes de fazer e o que podemos realmente fazer. E mostra que tem um exagero dos dois lados, do que prometem e realmente podiam fazer. E da preocupação que temos com este tipo de influência em especial, quando outros pontos são mais preocupantes.
Uma das discussões que mais gostei no livro é do antagonismo entre acurácia e viés/preconceito. Ele dá um exemplo que dominou a discussão de análises, um algoritmo para predição de reincidência de crimes que errava mais frequentemente dizendo que negros podiam voltar a cometer crimes, por ser calibrado para ser preciso. Para em seguida mostrar que, baseado nos dados disponíveis (maior reincidência entre negros), um algoritmo que não errasse dessa forma teria que discriminar entre negros e brancos na hora de fazer a avaliação. E seria menos preciso. Sumpter não deu uma resposta definitiva nem disse o que era o certo nessa questão, só mostrou que o cenário maior precisa ser levado em conta quando fazemos análises.
Outra discussão muito boa foi como alguns tipos de otimização podem criar desigualdades, mesmo sem má intenção por trás. O que cheguei a discutir no IGTV, inclusive. Por exemplo, se alguém otimizar uma propaganda de uma vaga de emprego no Facebook para atingir mais pessoas com um certo gosto (quem assiste Nerdologia, por exemplo), pode acabar apresentando essa propaganda predominantemente para homens. E se estiver otimizando para economizar impressões, isso inclusive pode ser reforçado. Mas a consequência não intencional é que mulheres nem terão a chance de tentar esse emprego porque não verão a vaga.
No fim, achei uma ótima mistura entre conhecimento do assunto e análises de alguém que não está financeiramente envolvido na performance desses programas. Vale para se atualizar sobre o que estamos passando.
May 12, 2018
After the recent spat of Cambridge Ananlytica, I was eager to know whether Facebook or internet can really change or reinforce my voting behavior; are we so naive. We all know about filter bubbles that trap us on various social media sites and shopping websites, but this book goes ahead to discuss about how algorithms used by Tech giants go ahead to manipulate human behavior and decisions. No, it does not go on the path of a conspiracy theory or something but rather tries to explain concepts like Principal Componets Analysis, 'also-like' algorithms, personality dimensions created by FB, Black hats and many more. Having done beginner course on data science during my academics I found these concepts quite intriguing.
The book highlights the fact that there has been rising trend to in developed countries to shift from circumstantial evidence to statistical proof and and data providers and analyzers have already started luring security forces (esp in US which has defense expenditure of $610 bn). On the lines of the book Weapons of Math Destruction, the author explains how algorithms of tech giants continue to further inequality in the society. Needless to say that these algorithms are completely unregulated. Citing a small example from the book, white men continue to get ads of high paig jobs whereas women get ads for generic jobs, thus keeping them unaware of the opportunities that would actually suit them. But here algorithm is nowhere to blame as the meta data that has been fed to it propels such process under the guise of optimization. The book discusses about a COMPAS an algorithm used somewhere in US to determine whether a prisoner should be released before end of sentence or does he/she need more correction, deciding his possibility of re-offending; to no surprises the algorithm creates a bias against black males. The moot point here is how these algorithms are slowly creeping into policy making space and there is very little that is being done to regulate them or the decisions based on answers provided by these algorithms.
Coming to Cambridge Analytica, some impact making customized propaganda lines meant to lure people based on where the person stands i Big 5 Personality model are mentioned. Math tools use rankings and probabilities on the big 5 personality model and algorithms are frther tuned to tolerable erors. < Unlike other forecasters Cambridge Analytical does not just rely on demographics but also behavioral conditioning of voters to create informed forecast of future behaviour to influence actions. >
Amongst all this the author also expresses his helplessness that he mentors students who later on go to develop such tech solutions that pay no heed to impact of these solutions on social outcomes.
The author further suggests that neural networks are the way forward. The author gives layman primer of artificial convolutional network and recurrent neural network and their applications, how little have these technologies progressed. Leave alone, emulating a pet by a robot, even emulating C.elegans one of the simplest organisms ( in terms how they interact, predict & optimize) is still a farcry, quoting that the current age is still of advertising intelligence and not artificial intelligence.
I would surely recommend this book to anyone who has little interest in data science.
The book highlights the fact that there has been rising trend to in developed countries to shift from circumstantial evidence to statistical proof and and data providers and analyzers have already started luring security forces (esp in US which has defense expenditure of $610 bn). On the lines of the book Weapons of Math Destruction, the author explains how algorithms of tech giants continue to further inequality in the society. Needless to say that these algorithms are completely unregulated. Citing a small example from the book, white men continue to get ads of high paig jobs whereas women get ads for generic jobs, thus keeping them unaware of the opportunities that would actually suit them. But here algorithm is nowhere to blame as the meta data that has been fed to it propels such process under the guise of optimization. The book discusses about a COMPAS an algorithm used somewhere in US to determine whether a prisoner should be released before end of sentence or does he/she need more correction, deciding his possibility of re-offending; to no surprises the algorithm creates a bias against black males. The moot point here is how these algorithms are slowly creeping into policy making space and there is very little that is being done to regulate them or the decisions based on answers provided by these algorithms.
Coming to Cambridge Analytica, some impact making customized propaganda lines meant to lure people based on where the person stands i Big 5 Personality model are mentioned. Math tools use rankings and probabilities on the big 5 personality model and algorithms are frther tuned to tolerable erors. < Unlike other forecasters Cambridge Analytical does not just rely on demographics but also behavioral conditioning of voters to create informed forecast of future behaviour to influence actions. >
Amongst all this the author also expresses his helplessness that he mentors students who later on go to develop such tech solutions that pay no heed to impact of these solutions on social outcomes.
The author further suggests that neural networks are the way forward. The author gives layman primer of artificial convolutional network and recurrent neural network and their applications, how little have these technologies progressed. Leave alone, emulating a pet by a robot, even emulating C.elegans one of the simplest organisms ( in terms how they interact, predict & optimize) is still a farcry, quoting that the current age is still of advertising intelligence and not artificial intelligence.
I would surely recommend this book to anyone who has little interest in data science.
August 27, 2018
This is a timely, clear read on filters, connections and capitalism in the internet age.
It does suffer from the academic problem. It's written by someone with actual academic discipline, which means they have no interest in saying something inflammatory or have delusions of grandeur on having discovered the Theory of Everything. The statements are measured, based on evidence and scientifically conservative in their judgements. They're also moderately dull.
If you have a contemporary understanding of statistics and internet corporations, there's little here that's new. I'd be thrilled to give this to an older relative who is anxious or curious about what Twitter is or why Facebook is suddenly in the news all the time.
It's a decent book, but not much of a read.
It does suffer from the academic problem. It's written by someone with actual academic discipline, which means they have no interest in saying something inflammatory or have delusions of grandeur on having discovered the Theory of Everything. The statements are measured, based on evidence and scientifically conservative in their judgements. They're also moderately dull.
If you have a contemporary understanding of statistics and internet corporations, there's little here that's new. I'd be thrilled to give this to an older relative who is anxious or curious about what Twitter is or why Facebook is suddenly in the news all the time.
It's a decent book, but not much of a read.
August 4, 2018
Demystified most of the myths around bad influences of algorithms in our day to day lives. A well researched book on algorithms, AI, and computational mathematics. Gave a new perspective of how things are actually working. Facebook is not influencing our emotions.
Cambridge analytica did not majorly help with the elections.
Google algorithms are not actually racist.
You’d get to know how these truths are unearthed in this book. A great read!!
Cambridge analytica did not majorly help with the elections.
Google algorithms are not actually racist.
You’d get to know how these truths are unearthed in this book. A great read!!
August 9, 2018
#alsoliked
I loved this book, learnt a lot about how FB and Google algorithms rule our online lives 😊
I loved this book, learnt a lot about how FB and Google algorithms rule our online lives 😊
June 25, 2018
Breaks through the hype and hyperbole. A good read. Actually, an important read, to think clearly about the topics of artificial intelligence and algorithms that are increasingly more important in society and in our lives.
November 4, 2018
Um livro muito interessante sobre o papel cada vez mais relevante dos algoritmos em nossas vidas, com uma visão geral sobre a matemática e estatística envolvida. O inglês David Sumpter é professor de Matemática Aplicada na Suécia, e nos apresenta diversos tópicos "polêmicos" com uma boa dose de realismo necessário. Temas que despertaram o interesse público são analisados: bolhas de informação, fake news nas eleições de 2016 nos EUA, algoritmos utilizados para prever reincidência de crimes, etc. Procedimentos matemáticos como PCA, regressões, algoritmos do tipo "also liked", redes neurais, inteligência artificial são apresentadas em essência. O autor também conversou com vários cientistas de dados, matemáticos e estatísticos que trabalham em empresas que estão na linha de frente da análise de dados utilizando as técnicas mais modernas e ambiciosas.
Gostei muito do tom do livro, do rigor do autor e de sua maneira de pensar, combinada com a exposição da realidade atual. Há problemas sérios que podem emergir quando terceirizamos responsabilidades para algoritmos do tipo "caixa preta". É um tema ameaçador, mas se tem algo que o autor não faz é ceder a histerias. Em geral, há muito mais hype do que realidade. Vemos que em muitos aspectos, Cambridge Analytica e Artificial Intelligence poderiam ser substituídos por "Cambridge Hyperbolitica" e "Advertising Intelligence", ou seja, há muita propaganda na área, com um exagero no discurso das empresas envolvidas em comparação com o que de fato podem fazer.
Gostei muito do tom do livro, do rigor do autor e de sua maneira de pensar, combinada com a exposição da realidade atual. Há problemas sérios que podem emergir quando terceirizamos responsabilidades para algoritmos do tipo "caixa preta". É um tema ameaçador, mas se tem algo que o autor não faz é ceder a histerias. Em geral, há muito mais hype do que realidade. Vemos que em muitos aspectos, Cambridge Analytica e Artificial Intelligence poderiam ser substituídos por "Cambridge Hyperbolitica" e "Advertising Intelligence", ou seja, há muita propaganda na área, com um exagero no discurso das empresas envolvidas em comparação com o que de fato podem fazer.
October 10, 2019
Suffers a bit from a mismatch of the material and the marketing. (Which is not Sumpter’s fault) The marketing algorithm said to focus on the ominous and dark side of algorithms. (And to put the title in Red, which is more threatening)
The author seems almost pitted against this tendency to let our imaginations and fear carry us away. MAYBE this title is a sly commentary on this tendency, but it’s more likely .. just marketing.
The actual material of the book is demystifying algorithms and in the process of this, DEFLATING a lot of what makes them sexy. This book is about returning people down to earth, which is admittedly not sexy... and which IS, let’s face it, a less engaging read. But an extra star given for respect for David Sumpter for being sober and honest, because that’s maybe what we needed anyway.
“My journey... has made me not a very interesting person to talk to.”
The author seems almost pitted against this tendency to let our imaginations and fear carry us away. MAYBE this title is a sly commentary on this tendency, but it’s more likely .. just marketing.
The actual material of the book is demystifying algorithms and in the process of this, DEFLATING a lot of what makes them sexy. This book is about returning people down to earth, which is admittedly not sexy... and which IS, let’s face it, a less engaging read. But an extra star given for respect for David Sumpter for being sober and honest, because that’s maybe what we needed anyway.
“My journey... has made me not a very interesting person to talk to.”
September 22, 2019
Outnumbered is a critical and sober overview of the use of algorithms in various areas, including social media, politics, criminal justice, and e-commerce. Using real-world examples, the author clearly explains how these algorithms actually work and he does so in way that doesn't require an advanced understanding of the underlying math (linear algebra, geometry, stats).
Outnumbered identifies what's real and what's hype in artificial intelligence and machine learning while acknowledging the grey areas that no one, including the most advanced technologists, aren't really sure of yet. Outnumbered serves as an excellent primer for smart lay people; those who are much further along in their knowledge will also appreciate this book.
Outnumbered identifies what's real and what's hype in artificial intelligence and machine learning while acknowledging the grey areas that no one, including the most advanced technologists, aren't really sure of yet. Outnumbered serves as an excellent primer for smart lay people; those who are much further along in their knowledge will also appreciate this book.
March 20, 2019
"Outnumbered" is a no-hype-included book about algorithms.
That happens for three reasons: 1) the author knows what he's talking about and shows it by testing his explanations in his own code; 2) the author gets baffled at what he sees as a sensationalistic approach in the public debate about it; and 3) the author isn't trying to sell you anything.
He isn't afraid to swim against the current and make very to-the-point criticism to sites I appreciate, like FiveThirtyEight. In fact, his observations made me think deeper about my work as a data journalist.
On top of it, the book is very readable.
That happens for three reasons: 1) the author knows what he's talking about and shows it by testing his explanations in his own code; 2) the author gets baffled at what he sees as a sensationalistic approach in the public debate about it; and 3) the author isn't trying to sell you anything.
He isn't afraid to swim against the current and make very to-the-point criticism to sites I appreciate, like FiveThirtyEight. In fact, his observations made me think deeper about my work as a data journalist.
On top of it, the book is very readable.
August 6, 2023
When being bombarded with "big Tech is spying on us" and "AI is gonna take our jobs" almost daily, it's so refreshing to read a book by someone who actually understands algorithms and AI technology. The book is a few years old, so the chapters on language processing are a bit outdated but all the points made are just as valid.
February 17, 2022
3,75 chyba idealnie trafiłam z tą książka między semestrami z AI i machine learningu
👌🏼
👌🏼
May 23, 2021
Dobra książka, początek wręcz o treści oczekiwanej. Mniej więcej w 3/4 książki trochę robi się za głęboko bardziej dla ludzi zafascynowanych niż czytelnika który tylko chce się dowiedzieć jak to działa albo się odbywa.
August 26, 2023
People who use Facebook or Google Should Read this Book
Modelling human behavior is hard - but it is increasingly important for the lay person to understand how and why algorithms impact our lives. This book is a very readable and comprehensive analysis of how data science is used to influence (or subliminally drive) our behavior - for better or worse. It describes the statistical techniques that are used and provides real life examples of how those techniques work. If you use Facebook or Google you should read this book - I guess that means everyone!
The sub-title of this book really tells the story - “From Facebook and Google to fake news and filter-bubbles - the algorithms that control our lives." Most people understand that today’s technology behemoths use data science techniques to better understand their users and customize their offerings but it generally stops there. This book attempts to explain to the layman more specifically how companies use mathematics and statistics to better understand us and make predictions as to our behavior.
The book is divided into 3 parts.
- Analyzing Us
- This section is mainly about Facebook and Google and how they use different data science techniques to better understand their users and deliver meaningful content. For example:
- Facebook - the importance of understanding how social media companies collect information about our personalities and use this data to predict our future behavior. It starts by explaining principal component analysis - a technique used by Facebook to categorize our personalities, values and socio-economic status. While most people tend to think of themselves as multi-dimensional; psychologists have boiled down human traits to a mere 5 elements - openness, conscientiousness, extroversion, agreeableness and neuroticism. Facebook can then analyze our posts, likes and pictures to better understand us across these 5 dimensions and then pitch products to us that are more likely to appeal.
- Cambridge Analytica - the importance of understanding probabilities and the inherent limitations of algorithmic predictions. This chapter is fascinating and sets the stage by pointing out that the algorithms underpinning data science are probabalistic - they calculate a number that is proportional to the probability of a specific fact being true about a person. He goes on to explain how regression models use data fitted to one group of people to infer the preferences of others. In the old days (1990's) these regression models could use inputs such as age, gender, class etc to determine the probability that a person would vote for a particular political party. However nowadays, modern data analytics companies have access to an almost limitless variety of data.
- Northpointe - the importance of data bias and false positives/negatives as it relates to criminal recidivism
- Spotify -- the importance of "data alchemy”, or the need for intuition and expertise by the human data damodeler to more fully understand our needs.
-
- Sumpter concludes this section with a discussion of Julia Dressel, a student who conducted an experiment that determined that, when all is said and done, human judgment is just as sound as algorithmic predictions, although computers can operate at scale and are faster and cheaper than people. While this may be true when looking at the big picture, I suspect that algorithmic predictions will outperform humans at the margin where a more granular analysis of variables is important.
- Influencing Us
- This section moves the discussion forward by outlining how data science has become more sophisticated in attempting to better understand us as humans and thereby influence our decisions and choices, such as:
- FiveThirtyEight and the booming election forecasting business. There is an inherent challenge in creating probabalistic models to forecast binary events - such as elections or sporting events, although multiple and/or braver predictions (>95% probability) are more meaningful. Further, such predictions must attempt to adjust for the inherent shortcomings in poll data - the primary input in election forecasting, unfortunately it is very difficult to incorporate the wisdom of the crowd into such models. Statistical models need to understand our sometimes subjective and erratic human behavior, otherwise these models typically do not perform much better than human judgment.
- One way of achieving this understanding is to move beyond simply categorizing us to trying to understand and influence us. Amazon recognized this early on, and using a form of collaborative filtering is able to make recommendations as to what we may like. Knowing that people similar to us “also liked” some other product provides the consumer with a level of confidence in a prospective purchase.
- Social media has also become a prevalent way to influence our behavior but it does so by amplifying our pre-existing biases. People tend to click on links that take them to new sites containing similar thoughts and ideas - in effect an Echo chamber forms where our beliefs are reinforced. Likewise, filter bubbles use an algorithmic approach to steer us towards material it believes we will appreciate. That said, Sumpter notes that the media tends to exaggerate and sensationalize the influence these algorithms have on our behavior - our real-life interactions with friends and family are more important determinants of our behavior.
- Becoming Us
- Sumpter notes that the biggest challenge for the current state of algorithms used by the Facebooks and Google’s of the world is that they do not properly understand the meaning of the information we are sharing with each other. Consequently, these very same companies are attempting to solve this problem by chasing the holy grail - artificial intelligence.
- To achieve this, the algorithms must understand context and analogies and to do this they must be able to analyze text, such as the GloVe (Global Vector) algorithm discussed by Sumpter that maps words used together near each other to find meaning and is thereby able to auto-complete our text searches on Google.
- Neural networks, which are designed to mimic the way our brains function, have long been used to recognize patterns. However, more recently developed convolutional neural networks can solve problems without being told which problem they are solving. Computers have been trained to use brute force to play and win games like chess and Jeopardy, but more recently, the folks at Google have been able to learn games from the ground up, through trial and error and understanding the objective function of the game
- The human body contains 37 trillion cells and the brain 86 billion neurons - that represents a lot of catching up for the AI scientists.
- Smart people disagree on the potential perils of AI - for every foreboding Elon Musk or Stephen Hawking there is a far more ambivalent Bill Gates. So where should the average Joe come down on this issue? Maybe they should do as Sumpter suggests and check out Mark Zuckerberg's AI robot (Jarvis) video to see what he can do. IMHO, it's not much more than a slightly smarter home than we have available to us today.
So, is AI’s role in our lives to be relegated to a smart home assistant and a champion game player or will it evolve to a form of general AI? Many smart minds are working on important problems but there is still more hype and hope than results. Sumpter takes a somewhat more skeptical view and I suspect he may be right. Computers and algorithms may be able to do more and more but at what point will the market rebel - I for one am all for AI but I wouldn’t be caught dead in a self-driving car (at least, not yet!) - actually, I suspect I would!
Modelling human behavior is hard - but it is increasingly important for the lay person to understand how and why algorithms impact our lives. This book is a very readable and comprehensive analysis of how data science is used to influence (or subliminally drive) our behavior - for better or worse. It describes the statistical techniques that are used and provides real life examples of how those techniques work. If you use Facebook or Google you should read this book - I guess that means everyone!
The sub-title of this book really tells the story - “From Facebook and Google to fake news and filter-bubbles - the algorithms that control our lives." Most people understand that today’s technology behemoths use data science techniques to better understand their users and customize their offerings but it generally stops there. This book attempts to explain to the layman more specifically how companies use mathematics and statistics to better understand us and make predictions as to our behavior.
The book is divided into 3 parts.
- Analyzing Us
- This section is mainly about Facebook and Google and how they use different data science techniques to better understand their users and deliver meaningful content. For example:
- Facebook - the importance of understanding how social media companies collect information about our personalities and use this data to predict our future behavior. It starts by explaining principal component analysis - a technique used by Facebook to categorize our personalities, values and socio-economic status. While most people tend to think of themselves as multi-dimensional; psychologists have boiled down human traits to a mere 5 elements - openness, conscientiousness, extroversion, agreeableness and neuroticism. Facebook can then analyze our posts, likes and pictures to better understand us across these 5 dimensions and then pitch products to us that are more likely to appeal.
- Cambridge Analytica - the importance of understanding probabilities and the inherent limitations of algorithmic predictions. This chapter is fascinating and sets the stage by pointing out that the algorithms underpinning data science are probabalistic - they calculate a number that is proportional to the probability of a specific fact being true about a person. He goes on to explain how regression models use data fitted to one group of people to infer the preferences of others. In the old days (1990's) these regression models could use inputs such as age, gender, class etc to determine the probability that a person would vote for a particular political party. However nowadays, modern data analytics companies have access to an almost limitless variety of data.
- Northpointe - the importance of data bias and false positives/negatives as it relates to criminal recidivism
- Spotify -- the importance of "data alchemy”, or the need for intuition and expertise by the human data damodeler to more fully understand our needs.
-
- Sumpter concludes this section with a discussion of Julia Dressel, a student who conducted an experiment that determined that, when all is said and done, human judgment is just as sound as algorithmic predictions, although computers can operate at scale and are faster and cheaper than people. While this may be true when looking at the big picture, I suspect that algorithmic predictions will outperform humans at the margin where a more granular analysis of variables is important.
- Influencing Us
- This section moves the discussion forward by outlining how data science has become more sophisticated in attempting to better understand us as humans and thereby influence our decisions and choices, such as:
- FiveThirtyEight and the booming election forecasting business. There is an inherent challenge in creating probabalistic models to forecast binary events - such as elections or sporting events, although multiple and/or braver predictions (>95% probability) are more meaningful. Further, such predictions must attempt to adjust for the inherent shortcomings in poll data - the primary input in election forecasting, unfortunately it is very difficult to incorporate the wisdom of the crowd into such models. Statistical models need to understand our sometimes subjective and erratic human behavior, otherwise these models typically do not perform much better than human judgment.
- One way of achieving this understanding is to move beyond simply categorizing us to trying to understand and influence us. Amazon recognized this early on, and using a form of collaborative filtering is able to make recommendations as to what we may like. Knowing that people similar to us “also liked” some other product provides the consumer with a level of confidence in a prospective purchase.
- Social media has also become a prevalent way to influence our behavior but it does so by amplifying our pre-existing biases. People tend to click on links that take them to new sites containing similar thoughts and ideas - in effect an Echo chamber forms where our beliefs are reinforced. Likewise, filter bubbles use an algorithmic approach to steer us towards material it believes we will appreciate. That said, Sumpter notes that the media tends to exaggerate and sensationalize the influence these algorithms have on our behavior - our real-life interactions with friends and family are more important determinants of our behavior.
- Becoming Us
- Sumpter notes that the biggest challenge for the current state of algorithms used by the Facebooks and Google’s of the world is that they do not properly understand the meaning of the information we are sharing with each other. Consequently, these very same companies are attempting to solve this problem by chasing the holy grail - artificial intelligence.
- To achieve this, the algorithms must understand context and analogies and to do this they must be able to analyze text, such as the GloVe (Global Vector) algorithm discussed by Sumpter that maps words used together near each other to find meaning and is thereby able to auto-complete our text searches on Google.
- Neural networks, which are designed to mimic the way our brains function, have long been used to recognize patterns. However, more recently developed convolutional neural networks can solve problems without being told which problem they are solving. Computers have been trained to use brute force to play and win games like chess and Jeopardy, but more recently, the folks at Google have been able to learn games from the ground up, through trial and error and understanding the objective function of the game
- The human body contains 37 trillion cells and the brain 86 billion neurons - that represents a lot of catching up for the AI scientists.
- Smart people disagree on the potential perils of AI - for every foreboding Elon Musk or Stephen Hawking there is a far more ambivalent Bill Gates. So where should the average Joe come down on this issue? Maybe they should do as Sumpter suggests and check out Mark Zuckerberg's AI robot (Jarvis) video to see what he can do. IMHO, it's not much more than a slightly smarter home than we have available to us today.
So, is AI’s role in our lives to be relegated to a smart home assistant and a champion game player or will it evolve to a form of general AI? Many smart minds are working on important problems but there is still more hype and hope than results. Sumpter takes a somewhat more skeptical view and I suspect he may be right. Computers and algorithms may be able to do more and more but at what point will the market rebel - I for one am all for AI but I wouldn’t be caught dead in a self-driving car (at least, not yet!) - actually, I suspect I would!
June 21, 2018
On a recent visit to Oxford bookstore in delhi,I stumbled upon this book and began to read it straight out of the store.I read this book in about 5 sittings and I was tired at the end of each one them,trying to comprehend logic and math behind complex algorithms.
Unlike other books on the subject,it does not go on ranting about how Google and Social Media sites have our data,it does for the first two chapters but then it gets back on track.They have our data,it’s general knowledge but the book asks better questions like
1.How that data is being used to analyze and influence us?
2.Are the methods being used reliable?
3.How efficient are these methods and what impact are they having on the real world.
And More importantly,are we being outnumbered?
Unlike other books on the subject,it does not go on ranting about how Google and Social Media sites have our data,it does for the first two chapters but then it gets back on track.They have our data,it’s general knowledge but the book asks better questions like
1.How that data is being used to analyze and influence us?
2.Are the methods being used reliable?
3.How efficient are these methods and what impact are they having on the real world.
And More importantly,are we being outnumbered?
January 2, 2022
Algorithms are used everywhere to help us better understand the world. But do we really want to understand the world better if this means dissecting the things we love and losing our personal integrity?
We love to have heroes and villains, geniuses and idiots, to see things in black and white, and not in the grey reality of probabilities.
Facebook has obtained a series of patents that allows the company to systematically collect, process and analyse our emotional state.
During the Us presidential election fake news stories really took off with a large number of stories originated from a small town in Macedonia, where a group of youngsters were being paid for the adverts shown on the sites.
We love to have heroes and villains, geniuses and idiots, to see things in black and white, and not in the grey reality of probabilities.
Facebook has obtained a series of patents that allows the company to systematically collect, process and analyse our emotional state.
During the Us presidential election fake news stories really took off with a large number of stories originated from a small town in Macedonia, where a group of youngsters were being paid for the adverts shown on the sites.
January 29, 2019
The book is written by a mathematician, so a good third of the book is spent giving a very high level concept of how various algorithms work (again, VERY high level). For that purpose, I'd recommend this book to anyone who's interested in the technical side, without having the deeper education behind it. I did sometimes feel that the author then extrapolates his own judgement of whether an algorithm is effective or not, with no real evidence. This isn't a book that's meant to consider the ethics of these algorithms, but more of a "is this really something that can be done?"
July 3, 2018
It was very interesting in the first few chapters.Then, it became more technical and I found hard to focus.I read both Outumbered and Freakonomics last week and both these books are brain opening.How data matters more than conventional wisdom is a lesson here. The Author has done applaudable research.I am impressed.
October 29, 2019
A 5-star solid read for one who may not have any clue what's going on with content curation through algorithms, but it's not really a technical read. I did find it informative, and I'd certainly recommend it to someone who wanted to understand the broader concepts.
November 3, 2018
Very informative book, though it is a bit lengthy. It's amazing how many algorithms we encounter in our daily life.
December 26, 2018
Enlightening and insightful, if not always compelling.
February 17, 2019
A fascinating read.
January 4, 2020
Czy przemożny wpływ fake newsów nie jest sam fake newsem?
David Sumpter to matematyk zajmujący się algorytmiką. W książce "Osaczeni przez liczby. O algorytmach, które kontrolują nasze życie. Od Facebooka i Google'a po fake news i bańki filtrujące" w sposób bardzo lekki opisał ścisłe podstawy cyfrowych zjawisk społecznych, które budują internetowe interakcje ludzi z informacjami o innych ludziach. Pyta o faktyczny poziom zagrożeń, manipulacji, anonimowości nas w sieci, gdy korzystamy z uzależniających usług potentatów wirtualnego świata. Oswojenia tego 'sieciowego globalnego stwora' poszukuje profesor poprzez przybliżanie mechanizmów, które korporacje używają do zbierania o nas informacji - zarówno socjotechnik jak i zaszytego kodu programistycznego.
"Osaczeni w liczbach" w istocie jest połączeniem analizy kilku niepokojących i medialnie nagłośnionych ostatnio zjawisk społecznych, które Internet wygenerował (Cambridge Analytica, wybory prezydenckie w USA w 2016, wynik głosowania w sprawie brexitu oraz kilku mniej znanych faktów w kontekście dyskryminacji i wycieku danych) z ciekawym podaniem stojących za nimi algorytmami i statystyką. Cel autora jest prosty - pokazać faktyczne źródła powstających sądów o cyfrowym świecie, by oswoić niepokoje, czasem by ostrzec czy zaprosić czytelnika do zniuansowania przekonań.
Matematyk analizuje bestsellery wydawnicze i sukces pisarski (chętnie zabrałbym głos na ten temat na LC), systemy punktowania publikacji naukowych, sposoby budowania wielowymiarowych baz wiedzy przez portale społecznościowe czy mechanizmy pozycjonujące treści w przeglądarkach. Tu jest sporo ciekawych informacji, z których kilka mnie zaskoczyło. W sumie tak 60% tekstu, to bardzo wartościowy materiał do przemyśleń dla każdego użytkownika Internetu. Są przykłady, trochę grafik i prosty język. Lektura wciąga i pobudza do refleksji.
Trochę narzekania. Sumpter czasem jednak zbyt nieformalnie zwraca się do czytelnika pomijając istotne dla mnie fakty. Unika żargonu (to jest z reguły dobre), ale kilkukrotnie przesadnie skupia się na detalach nieistotnej codzienności i przechodzi wtedy w styl reporterski czy dygresyjny. W samej treści i doborze tematów jest z reguły dobrze, poza przyszywanymi partiami o futbolu czy na siłę doczepionymi rozważaniami o przyszłości sztucznej inteligencji. Zdecydowanie ciekawiej jest w rozdziałach, w których skupia się na kluczowym w narracji styku: algorytm-człowiek-społeczne relacje. W ramach polemiki, nie zgadzam się z diagnozą, że negatywne treści wpuszczane do Internetu są niegroźne, gdy nie dominują nad pozytywnymi (str. 188-190). Moja wiedza o big data sugeruje, że już zaledwie 25% przekazu negatywnego (przy 75% pozytywnego) oznacza, że atrakcyjność negacji rozprzestrzenia się lepiej w sieci. Może jeszcze jedno - trochę za dużo było o amerykańskim świecie, ale ostatecznie Dolina Krzemowa akurat tam rezyduje.
Podsumowując. Bardzo ciekawa pozycja dla użytkowników Internetu. Może dla mnie w sporej części oczywista, ale język autora sprawił, że ostatecznie przyswajanie treści odbywało się sprawnie.
Polecam szczególnie, jeśli nurtują Ciebie pytania:
• Kto lub co odpowiada za dyskryminujące czasem sposoby prezentacji informacji w Internecie?
• Czy FB faktycznie wie o nas więcej, niż my sami?
• W jakim stopniu nasze wybory treści tworzą wokół nas bańki informacyjne?
• Jak działa wyszukiwarka, mechanizm polubień czy dziedziczenie profilujące znajomych?
• Czy fake newsy to dominujący negatywny składnik treści internetowych?
• Jak przechytrzyć portale randkowe, zarobić na często wyświetlanych treściach z YouTube, walczyć z botami?
• No i tak ogólnie - czego należy się bać w kontaktach człowiek-maszyna?
7/10 - DOBRA
David Sumpter to matematyk zajmujący się algorytmiką. W książce "Osaczeni przez liczby. O algorytmach, które kontrolują nasze życie. Od Facebooka i Google'a po fake news i bańki filtrujące" w sposób bardzo lekki opisał ścisłe podstawy cyfrowych zjawisk społecznych, które budują internetowe interakcje ludzi z informacjami o innych ludziach. Pyta o faktyczny poziom zagrożeń, manipulacji, anonimowości nas w sieci, gdy korzystamy z uzależniających usług potentatów wirtualnego świata. Oswojenia tego 'sieciowego globalnego stwora' poszukuje profesor poprzez przybliżanie mechanizmów, które korporacje używają do zbierania o nas informacji - zarówno socjotechnik jak i zaszytego kodu programistycznego.
"Osaczeni w liczbach" w istocie jest połączeniem analizy kilku niepokojących i medialnie nagłośnionych ostatnio zjawisk społecznych, które Internet wygenerował (Cambridge Analytica, wybory prezydenckie w USA w 2016, wynik głosowania w sprawie brexitu oraz kilku mniej znanych faktów w kontekście dyskryminacji i wycieku danych) z ciekawym podaniem stojących za nimi algorytmami i statystyką. Cel autora jest prosty - pokazać faktyczne źródła powstających sądów o cyfrowym świecie, by oswoić niepokoje, czasem by ostrzec czy zaprosić czytelnika do zniuansowania przekonań.
Matematyk analizuje bestsellery wydawnicze i sukces pisarski (chętnie zabrałbym głos na ten temat na LC), systemy punktowania publikacji naukowych, sposoby budowania wielowymiarowych baz wiedzy przez portale społecznościowe czy mechanizmy pozycjonujące treści w przeglądarkach. Tu jest sporo ciekawych informacji, z których kilka mnie zaskoczyło. W sumie tak 60% tekstu, to bardzo wartościowy materiał do przemyśleń dla każdego użytkownika Internetu. Są przykłady, trochę grafik i prosty język. Lektura wciąga i pobudza do refleksji.
Trochę narzekania. Sumpter czasem jednak zbyt nieformalnie zwraca się do czytelnika pomijając istotne dla mnie fakty. Unika żargonu (to jest z reguły dobre), ale kilkukrotnie przesadnie skupia się na detalach nieistotnej codzienności i przechodzi wtedy w styl reporterski czy dygresyjny. W samej treści i doborze tematów jest z reguły dobrze, poza przyszywanymi partiami o futbolu czy na siłę doczepionymi rozważaniami o przyszłości sztucznej inteligencji. Zdecydowanie ciekawiej jest w rozdziałach, w których skupia się na kluczowym w narracji styku: algorytm-człowiek-społeczne relacje. W ramach polemiki, nie zgadzam się z diagnozą, że negatywne treści wpuszczane do Internetu są niegroźne, gdy nie dominują nad pozytywnymi (str. 188-190). Moja wiedza o big data sugeruje, że już zaledwie 25% przekazu negatywnego (przy 75% pozytywnego) oznacza, że atrakcyjność negacji rozprzestrzenia się lepiej w sieci. Może jeszcze jedno - trochę za dużo było o amerykańskim świecie, ale ostatecznie Dolina Krzemowa akurat tam rezyduje.
Podsumowując. Bardzo ciekawa pozycja dla użytkowników Internetu. Może dla mnie w sporej części oczywista, ale język autora sprawił, że ostatecznie przyswajanie treści odbywało się sprawnie.
Polecam szczególnie, jeśli nurtują Ciebie pytania:
• Kto lub co odpowiada za dyskryminujące czasem sposoby prezentacji informacji w Internecie?
• Czy FB faktycznie wie o nas więcej, niż my sami?
• W jakim stopniu nasze wybory treści tworzą wokół nas bańki informacyjne?
• Jak działa wyszukiwarka, mechanizm polubień czy dziedziczenie profilujące znajomych?
• Czy fake newsy to dominujący negatywny składnik treści internetowych?
• Jak przechytrzyć portale randkowe, zarobić na często wyświetlanych treściach z YouTube, walczyć z botami?
• No i tak ogólnie - czego należy się bać w kontaktach człowiek-maszyna?
7/10 - DOBRA
September 14, 2019
The first part of the book explains why the fantasy of using principal component analysis, regression modelling, and data collection from social media to create a computerised and high dimensional understanding of human personality that can outperform our own current understanding is just hubris. It lacks sufficient power. Modelling by definition simplifies and are inherently probabilistic. Written by humans, algorithms are not better than our collective wisdom in accuracy. However they are efficient when dealing with large volumes and under the right circumstances can be very useful. Crucially, algorithms cannot answer philosophical paradoxes that involve ethical values.
When algorithms involve positive feedback and cross reference in the context of assessing popularity, it can inflate otherwise statistically insignificant initial random fluctuations and differences. If a human behaviour is influenced by these algorithms, apparent but faux meaningfulness will arise from thin air, which is absurd. This does occur in advertising and marketing. However, the author provides ample and solid evidence that echo chambers and filter bubbles have not to date been able to influence political opinions, activities, and outcomes to any significant degree, if at all. We (at least most of us who are otherwise mainstream) are not yet living in a so-called post-truth world.
The final chapters that comprise the last part of the book discuss fascinating topics such as mathematical formalism of lexicology and deep neural networks. The take home message is that AI is not equivalent to intelligence at all as to date it fails to generalise concepts, contextualise problems, or demonstrate anything that resembles understanding.
Very informative and easy read. Four stars.
When algorithms involve positive feedback and cross reference in the context of assessing popularity, it can inflate otherwise statistically insignificant initial random fluctuations and differences. If a human behaviour is influenced by these algorithms, apparent but faux meaningfulness will arise from thin air, which is absurd. This does occur in advertising and marketing. However, the author provides ample and solid evidence that echo chambers and filter bubbles have not to date been able to influence political opinions, activities, and outcomes to any significant degree, if at all. We (at least most of us who are otherwise mainstream) are not yet living in a so-called post-truth world.
The final chapters that comprise the last part of the book discuss fascinating topics such as mathematical formalism of lexicology and deep neural networks. The take home message is that AI is not equivalent to intelligence at all as to date it fails to generalise concepts, contextualise problems, or demonstrate anything that resembles understanding.
Very informative and easy read. Four stars.
May 2, 2021
Great book that shows how algorithms affect (or don't affect) our lives. Sumpter highlights how some of the things people on the left have purported about Cambridge Analytica are false, how companies use algorithms in a way that can put people in bubbles, and how they can control who succeeds and who doesn't - from authors, YouTubers, and scientists. He also talks about where we are at in AI intelligence and where we might go. It's a great combination of math, computers, and social science.
Unlike other readers, I didn't find this book dry at all. Some might be wary of reading it, because, maths. Don't be. I am not a maths major and it wasn't hard to understand. It was actually quite interesting.
Dislikes? Probably has to do with the author. I just don't like when people have to resume, tell their readers they are honest and morally right. (It doesn't instill trust). He brings in the popularly trending, hot-button social topics of racism and sexism and seems to advocate for "fair".
(If fairness is a human construct, when algorithms are applied equally to everyone but produce "unfair" outcomes, who exactly determines that and gets to decide what is "fair"?).
I appreciate that he tries to remain impartial to left and right sides of the political aisle, but you might pick up on the little trailing barbs towards people on the right...that's where the racists hang out, the ones who fall for fake news, and believe conspiracy theories because they don't watch main stream news... (Ignore the fact that shortly before he was talking about how main stream news exaggerates and distorts truth).
As I said though, he tries. The problems I have with it are relatively minor.
Overall, I recommend it. (An AI may not understand that last sentence, but you probably do. :))
Unlike other readers, I didn't find this book dry at all. Some might be wary of reading it, because, maths. Don't be. I am not a maths major and it wasn't hard to understand. It was actually quite interesting.
Dislikes? Probably has to do with the author. I just don't like when people have to resume, tell their readers they are honest and morally right. (It doesn't instill trust). He brings in the popularly trending, hot-button social topics of racism and sexism and seems to advocate for "fair".
(If fairness is a human construct, when algorithms are applied equally to everyone but produce "unfair" outcomes, who exactly determines that and gets to decide what is "fair"?).
I appreciate that he tries to remain impartial to left and right sides of the political aisle, but you might pick up on the little trailing barbs towards people on the right...that's where the racists hang out, the ones who fall for fake news, and believe conspiracy theories because they don't watch main stream news... (Ignore the fact that shortly before he was talking about how main stream news exaggerates and distorts truth).
As I said though, he tries. The problems I have with it are relatively minor.
Overall, I recommend it. (An AI may not understand that last sentence, but you probably do. :))
October 26, 2018
The book is something of an antidote to the fear of algorithms taking over everything. Sumpter clearly explains many of the mathematical concepts underlying algorithms used by Facebook or Google, then relates his conversations with other experts on those algorithms. The explanations are great and the conversations are interesting, but two issues prevent this book from getting five stars:
At one point, Sumpter discusses Facebook's experiment of showing users more negative news. When this was first publicized, there was a lot of criticism of Facebook, but Sumpter claims that the criticism was overblown, pointing to the results of the study showing that the filtering had minimal effect. This might be true, but it's not OK to use the results as a post hoc justification - they didn't know, when running the study, if the effect would be minimal, therefore they should not have done it.
Also, the book has two chapters discussing filter bubbles which seem to disagree. When first introducing the concept, Sumpter illustrates how filtering algorithms can act on even people who are equally likely to share liberal or conservative content. Then, he talks about networks of friends and followers on Facebook and Twitter and argues that both liberals and conservatives are exposed to content from both sides. I think that even if there are friends or followers who share other content, the filtering would tend to show preferred content, but the chapter doesn't really distinguish between friends liking/sharing content and me actually seeing it.
At one point, Sumpter discusses Facebook's experiment of showing users more negative news. When this was first publicized, there was a lot of criticism of Facebook, but Sumpter claims that the criticism was overblown, pointing to the results of the study showing that the filtering had minimal effect. This might be true, but it's not OK to use the results as a post hoc justification - they didn't know, when running the study, if the effect would be minimal, therefore they should not have done it.
Also, the book has two chapters discussing filter bubbles which seem to disagree. When first introducing the concept, Sumpter illustrates how filtering algorithms can act on even people who are equally likely to share liberal or conservative content. Then, he talks about networks of friends and followers on Facebook and Twitter and argues that both liberals and conservatives are exposed to content from both sides. I think that even if there are friends or followers who share other content, the filtering would tend to show preferred content, but the chapter doesn't really distinguish between friends liking/sharing content and me actually seeing it.
January 18, 2019
Excelente leitura para quem gosta de entender um pouco mais sobre o que é o mundo dos dados e o impacto em nossas vidas.
Sumpter traz uma leitura que chega a incomodar, pois foge muito da forma como se é discutido e veiculado nas mídias. Aqui, o autor derruba algumas crenças e mitos ( o que mais me impressionou foi a ótica das bolhas da internet), sempre trazendo dados e pesquisa que ele mesmo fez.
O que mais me chamou atenção foi a explicação de algoritmos enviesados - racistas, no caso. Sumpter mostra, de forma leve e muito bem explicada, como é impossível (ao menos atualmente) ter um algoritmo que tenha boa acurácia - calibrado- mas que não tenha um viés significativo.
Também foi interessante a forma como ele tratou o caso da Cambridge Analytica (o que mais me motivou a comprar o livro), mostrando que a equipe envolvida no caso, e a mídia também, exageraram, e muito, nos resultados. Não há como afirmar, categoricamente e sem erros, que as eleições foram decididas por este caso.
Enfim...o livro é extenso mas muito bom, com uma leitura leve, bem explicada (ás vezes ahco que ele enrola demais heheh) e com muito bom humor - o autor faz diversas piadas e até de si mesmo.
Previsões, análise de dados, interpretações de dados, manipulações, tudo isso em diversos territórios como música (entrevistas com cientistas de dados do spotify), esportes (especialidade do autor), redes sociais, política e muito mais.
Mais que recomendado, necessário nos dias atuais.
Link Amazon -->> Outnumbered
Sumpter traz uma leitura que chega a incomodar, pois foge muito da forma como se é discutido e veiculado nas mídias. Aqui, o autor derruba algumas crenças e mitos ( o que mais me impressionou foi a ótica das bolhas da internet), sempre trazendo dados e pesquisa que ele mesmo fez.
O que mais me chamou atenção foi a explicação de algoritmos enviesados - racistas, no caso. Sumpter mostra, de forma leve e muito bem explicada, como é impossível (ao menos atualmente) ter um algoritmo que tenha boa acurácia - calibrado- mas que não tenha um viés significativo.
Também foi interessante a forma como ele tratou o caso da Cambridge Analytica (o que mais me motivou a comprar o livro), mostrando que a equipe envolvida no caso, e a mídia também, exageraram, e muito, nos resultados. Não há como afirmar, categoricamente e sem erros, que as eleições foram decididas por este caso.
Enfim...o livro é extenso mas muito bom, com uma leitura leve, bem explicada (ás vezes ahco que ele enrola demais heheh) e com muito bom humor - o autor faz diversas piadas e até de si mesmo.
Previsões, análise de dados, interpretações de dados, manipulações, tudo isso em diversos territórios como música (entrevistas com cientistas de dados do spotify), esportes (especialidade do autor), redes sociais, política e muito mais.
Mais que recomendado, necessário nos dias atuais.
Link Amazon -->> Outnumbered
November 23, 2018
An apt read for these times of fears over 'fake' news and tall claims made by digital companies. Sumpter takes a close look at technological claims with his magnifying glass and finds that the world is not yet as dangerous as some people claim, e.g. so-called 'fake' news might not be actually wielding as much influence as we have been told, and on the flip-side, technologies might not have yet become as cool as some companies would like us to believe, e.g. algorithms are still not better than humans at profiling voters, as far as accuracy is concerned.
Having said that, Sumpter encourages us to keep moving forward towards those lofty goals, such as identifying and weeding out fake news, and being able to be better at, say, predicting who will be a repeat offender, with the gentle caveat that we keep ourselves grounded while not losing sight of those lofty goals.
A bit of statistics background might be required to understand things better, but that should not be a deal-breaker given Sumpter's lucid language. If I were to point something out given Sumpter's skepticism of current state of affairs, it is that he is very eager to believe the counter-points from the people that are going against the grain. But that in itself is not a bad thing - because at least you get to see two points of view.
A fluid, fast read. I would have given 4.5 stars, but given the importance of the subject for current times, I give it 5.
Having said that, Sumpter encourages us to keep moving forward towards those lofty goals, such as identifying and weeding out fake news, and being able to be better at, say, predicting who will be a repeat offender, with the gentle caveat that we keep ourselves grounded while not losing sight of those lofty goals.
A bit of statistics background might be required to understand things better, but that should not be a deal-breaker given Sumpter's lucid language. If I were to point something out given Sumpter's skepticism of current state of affairs, it is that he is very eager to believe the counter-points from the people that are going against the grain. But that in itself is not a bad thing - because at least you get to see two points of view.
A fluid, fast read. I would have given 4.5 stars, but given the importance of the subject for current times, I give it 5.
November 22, 2018
Strongly recommended to those who are interested on the development of algorithms, information bubbles, fake news, and AI (especially neural network & unsupervised learning). There are also talks about Trump's use of the political consultants of Cambridge Analytica (or better to call it Cambridge Hyperbolytica? :D), and the failure of statisticians to predict the UK's Brexit vote.
Briefly, it can be concluded that every algorithm has its limitations. And we should be aware of that. It's getting better and better. But, we're still very long way from true (general) AI or human-like intelligence. Don't be outnumbered!
Briefly, it can be concluded that every algorithm has its limitations. And we should be aware of that. It's getting better and better. But, we're still very long way from true (general) AI or human-like intelligence. Don't be outnumbered!
Displaying 1 - 30 of 61 reviews