You’re not the customer, you’re the product!

The attention that each one of us pays to an item and the time we spend on a site, article, or application is the most valuable commodity in the world, as witnessed by the fact that the companies that sell it, wholesale, are the largest in the world. Attracting and selling our attention is, indeed, the business of Google and Facebook but also, to a larger extent, of Amazon, Apple, Microsoft, Tencent, or Alibaba. We may believe we are the customers of these companies but, in fact, many of the services provided serve, only, to attract our attention and sell it to the highest bidder, in the form of publicity of personal information. In the words of Richard Serra and Carlota Fay Schoolman, later reused by a number of people including Tom Johnson, if you are not paying “You’re not the customer; you’re the product.“

Attracting and selling attention is an old business, well described in Tim Wu’s book The Attention Merchants. First created by newspapers, then by radios and television, the market of attention came to maturity with the Internet. Although newspapers, radio programs, and television shows have all been designed to attract our attention and use it to sell publicity, none of them had the potential of the Internet, which can attract and retain our attention by tailoring the contents to each and everyone’s content.

The problem is that, with excessive customization, comes a significant and very prevalent problem. As sites, social networks, and content providers fight to attract our attention, they show us exactly the things we want to see, and not the things as they are. Each person lives, nowadays, in a reality that is different from anyone else’s reality. The creation of a separate and different reality, for each person, has a number of negative side effects, that include the creation of paranoia-inducing rabbit holes, the radicalization of opinions, the inability to establish democratic dialogue, and the diffiulty to distinguish reality from fabricated fiction.

Wu’s book addresses, in no light terms, this issue, but the Netflix documentary The Social Dilemma makes an even stronger point that customized content, as shown to us by social networks and other content providers is unraveling society and creating a host of new and serious problems. Social networks are even more worrying than other content providers because they create pressure in children and young adults to conform to a reality that is fabricated and presented to them in order to retain (and resell) their attention.

The Book of Why

Correlation is not causation is a mantra that you may have heard many times, calling attention to the fact that no matter how strong the relations one may find between variables, they are not conclusive evidence for the existence of a cause and effect relationship. In fact, most modern AI and Machine Learning techniques look for relations between variables to infer useful classifiers, regressors, and decision mechanisms. Statistical studies, with either big or small data, have also generally abstained from explicitly inferring causality between phenomena, except when randomized control trials are used, virtually the unique case where causality can be inferred with little or no risk of confounding.

In The Book of Why, Judea Pearl, in collaboration with Dana Mackenzie, ups the ante and argues not only that one should not stay away from reasoning about causes and effects, but also that the decades-old practice of avoiding causal reasoning has been one of the reasons for our limited success in many fields, including Artificial Intelligence.

Pearl’s main point is that causal reasoning is not only essential for higher-level intelligence but is also the natural way we, humans, think about the world. Pearl, a world-renowned researcher for his work in probabilistic reasoning, has made many contributions to AI and statistics, including the well known Bayesian networks, an approach that exposes regularities in joint probability distributions. Still, he thinks that all those contributions pale in comparison with the revolution he speared on the effective use of causal reasoning in statistics.

Pearl argues that statistical-based AI systems are restricted to finding associations between variables, stuck in what he calls rung 1 of the Ladder of Causation: Association. Seeing associations leads to a very superficial understanding of the world since it restricts the actor to the observation of variables and the analysis of relations between them. In rung 2 of the Ladder, Intervention, actors can intervene and change the world, which leads to an understanding of cause and effect. In rung 3, Counterfactuals, actors can imagine different worlds, namely what would have happened if the actor did this instead of that.

This may seem a bit abstract, but that is where the book becomes a very pleasant surprise. Although it is a book written for the general public, the authors go deeply into the questions, getting to the point where they explain the do-calculus, a methodology Pearl and his students developed to calculate, under a set of dependence/independence assumptions, what would happen if a specific variable is changed in a possibly complex network of interconnected variables. In fact, graphic representations of these networks, causal diagrams, are at the root of the methods presented and are extensively used in the book to illustrate many challenges, problems, and paradoxes.

In fact, the chapter on paradoxes is particularly entertaining, covering the Monty Hall, Berkson, and Simpson’s paradoxes, all of them quite puzzling. My favorite instance of Simpson’s paradox is the Berkeley admissions puzzle, the subject of a famous 1975 Science article. The paradox comes from the fact that, at the time, Berkeley admitted 44% of male candidates to graduate studies, but only 35% of female applicants. However, each particular department (departments decide the admissions in Berkeley, as in many other places) made decisions that were more favorable to women than men. As it turns out, this strange state of affairs has a perfectly reasonable explanation, but you will have to read the book to find out.

The book contains many fascinating stories and includes a surprising amount of personal accounts, making for a very entertaining and instructive reading.

Note: the ladder of causation figure is from the book itself.

A conversation with GPT-3 on COVID-19

GPT-3 is the most advanced language model ever created, a product of an effort by OpenAI to create a publicly available system that can be used to advance research and applications in natural language. The model itself published less than three months ago, is an autoregressive language model with 175 billion parameters and was trained with a dataset that includes almost a trillion words.

Impressive as that may be, it is difficult to get some intuition of what such a complex model, trained on billions of human-generated texts, can actually do. Can it be used effectively in translation tasks or in answering questions?

To get some idea of what a sufficiently high-level statistical model of human language can do, I challenge you to have a look at this conversation with GPT-3, published by Kirk Ouimet a few days ago. It relates a dialogue between him and GPT-3 on the topic of COVID-19. The most impressive thing about this conversation with an AI is not that it gets many of the responses right (others not so much). What impressed me is that the model was trained with a dataset created before the existence of COVID-19, which provided GPT-3 no specific knowledge about this pandemic. Whatever answers GPT-3 gives to the questions related to COVID-19 are obtained with knowledge that was already available before the pandemic began.

This certainly raises some questions on whether advanced AI systems should be more widely used to define and implement policies important to the human race.

If you want more information bout GPT-3, it is easy to find in a multitude of sites with tutorials and demonstrations, such as TheNextWeb, MIT Technology Review, and many, many others.

Human Compatible: AI and the Problem of Control

Stuart Russell, one of the better-known researchers in Artificial Intelligence, author of the best selling textbook Artificial Intelligence, A Modern Approach addresses, in his most recent book, what is probably one of the most interesting open questions in science and technology: can we control the artificially intelligent systems that will be created in the decades to come?

In Human Compatible: AI and the Problem of Control Russell formulates and answers the following, very important question: what are the consequences if we succeed in creating a truly intelligent machine?

The question brings, with it, many other questions, of course. Will intelligent machines be dangerous to humanity? Will they take over the world? Could we control machines that are more intelligent than ourselves? Many writers and scientists, like Nick Bostrom, Stephen Hawking, Elon Musk, Sam Harris, and Max Tegmark have raised these questions, several of them claiming that superintelligent machines could be around the corner and become extremely dangerous to the humanity.

However, most AI researchers have dismissed these questions as irrelevant, concentrated as they are in the development of specific techniques and well aware that Artificial General Intelligence is far away, if it is at all achievable.  Andrew Ng, another famous AI researcher, said that worrying about superintelligent machines is like worrying about the overpopulation. of Mars.

There could be a race of killer robots in the far future, but I don’t work on not turning AI evil today for the same reason I don’t worry about the problem of overpopulation on the planet Mars

Another famous Machine Learning researcher, Pedro Domingos, in his bestselling book, The Master Algorithm, about Machine Learning, the driving force behind modern AI, also ignores these issues, concentrating on concrete technologies and applications. In fact, he says often that he is more worried about dumb machines than about superintelligent machines.

Stuart Russell’s book is different, making the point that we may, indeed, lose control of such systems, even though he does not believe they could harm us by malice or with intention. In fact, Russell is quite dismissive of the possibility that machines could one day become truly intelligent and conscious, a position I find, personally, very brave, 70 years after Alan Turing saying exactly the opposite.

Yet, Russell believes we may be in trouble if sufficiently intelligent and powerful machines have objectives that are not well aligned with the real objectives of their designers. His point is that a poorly conceived AI system, which aims at optimizing some function that was badly specified can lead to bad results and even tragedy if such a system controls critical facilities. One well-known example is Bostrom’s paperclip problem, where an AI system designed to maximize the production of paperclips turns the whole planet into a paperclip production factory, eliminating humanity in the process. As in the cases that Russell fears, the problem comes not from a machine which wants to kill all humans, but from a machine that was designed with the wrong objectives in mind and does not stop before achieving them.

To avoid that risk os misalignment between human and machine objectives, Russell proposes designing provably beneficial AI systems, based on three principles that can be summarized as:

• Aim to maximize the realization of human preferences
• Assume uncertainty about these preferences
• Learn these preferences from human behavior

Although I am not fully aligned with Russell in all the positions he defends in this book, it makes for interesting reading, coming from someone who is a knowledgeable AI researcher and cares about the problems of alignment and control of AI systems.

The mind of a fly

Researchers from the Howard Hughes Medical Institute, Google and other institutions have published the neuron level connectome of a significant part of the brain of the fruit fly, what they called the hemibrain. This may become one of the most significant advances in our understanding of the detailed structure of complex brains, since the 302 neurons connectome of C. elegans was published in 1986, by a team headed by Sydney Brenner, in an famous article with the somewhat whimsical subtitle of The mind of a worm. Both methods used an approach based on the slicing of the brains in very thin slices, followed by the use of scanning electron microscopy and the processing of the resulting images in order to obtain the 3D structure of the brain.

The neuron-level connectome of C. elegans was obtained after a painstaking effort that lasted decades, of manual annotation of the images obtained from the thousands of slices imaged using electron microscopy. As the brain of Drosophila melanogaster, the fruit fly, is thousands of times more complex, such an effort would have required several centuries if done by hand. Therefore, Google’s machine learning algorithms have been trained to identify sections of neurons, including axons, bodies and dendritic trees, as well as synapses and other components. After extensive training, the millions of images that resulted from the serial electron microscopy procedure were automatically annotated by the machine learning algorithms, enabling the team to complete in just a few years the detailed neuron-level connectome of a significant section of the fly brain, which includes roughly 25000 neurons and 20 million synapses.

The results, published in the first of a number of articles, can be freely analyzed by anyone interested in the way a fly thinks. A Google account can be used to log in to the neuPrint explorer and an interactive exploration of the 3D electron microscopy images is also available with neuroglancer. Extensive non-technical coverage by the media is also widely available. See, for instance, the article in The Economist or the piece in The Verge.

Image from the HHMI Janelia Research Campus site.

Machines like me

Ian McEwan´s latest novel, Machines like me does not disappoint if you are looking for a well-written and accurate work of fiction about Artificial Intelligence. The novel takes place in a slightly parallel universe, where Alan Turing did not kill himself and, instead, continued to make important contributions to computer science and to Artificial Intelligence throughout his life. In this world, similar to ours but different in some important respects, AI has evolved much faster and, in the 80s, it became possible to acquire, by a reasonable amount, humanoid robots that could be used as servants, friends or companions.

And, indeed, Adam, the robot, is all of these. From the three characters in the novel (the other two are Charlie and Miranda and yes, there is a sort of love triangle involved) Adam has, no doubt, the more fascinating personality. Without giving away too much, Adam, who starts as something like a sophisticated new laptop, which a 470-page “user manual”, becomes the hero of the story, raising in the mind of the reader many questions about machine intelligence, consciousness, and the rights of intelligent machines. His takes on the events that unfold are sometimes brilliant (e.g., “those who believe in the afterlife will never be disappointed“), other times unexpected,  but never off the mark.

Artificially intelligent or not, Adam is by far the most fascinating character of the lot, and we find ourselves empathizing with him (or it?), in a way that you may not expect

In the process of telling the story, Ian McEwan creates an alternative version of the history of computer science and Artificial Intelligence, which is accurate, thought-provoking, and, ultimately, quite plausible. I strongly recommend this book as an inspiring reading for the summer!

Mastering Starcraft

The researchers at DeepMind keep advancing the state of the art on the utilization of deep learning to master ever more complex games. After recently reporting a system that learns how to play a number of different and very complex board games, including Go and Chess, the company announced a system that is able to beat the best players in the world at a complex strategy game, Startcraft.

AlphaStar, the system designed to learn to play Starcraft, one of the most challenging Real-Time Strategy (RTS) games, by playing against other versions of itself, represents a significant advance in the application of machine learning. In Starcraft, a significant amount of information is hidden from the players, and each player has to balance short term and long term objectives, just like in the real world. Players have to master fast-paced battle techniques and, at the same time, develop their own armies and economies.

This result is important because it shows that deep reinforcement learning, which has already shown remarkable results in all sorts of board games,  can scale up to complex environments with multiple time scales and hidden information. It opens the way to the application of machine learning to real-world problems, until now deemed to difficult to be tackled by machine learning.

Deepmind presents Artificial General Intelligence for board games

In a paper recently published in the journal Science, researchers from DeepMind describe Alpha Zero, a system that mastered three very complex games, Go, chess, and shogi, using only self-play and reinforcement learning. What is different in this system (a preliminary version was previously referred in this blog), when compared with previous ones, like AlphaGo Zero, is that the same learning architecture and hyperparameters were used to learn different games, without any specific customization for each different game.

Historically, the best programs for each game were heavily customized to use and exploit specific characteristics of that game. AlphaGo Zero, the most impressive previous result, used the spatial symmetries of Go and a number of other specific optimizations. Special purpose chess program like Stockfish took years to develop, use enormous amounts of field-specific knowledge and can, therefore, only play one specific game.

Alpha Zero is the closest thing to a general purpose board game player ever designed. Alpha Zero uses a deep neural network to estimate move probabilities and position values. It performs the search using a Monte Carlo tree search algorithm, which is general-purpose and not specifically tuned to any particular game. Overall, Alpha Zero gets as close as ever to the dream of artificial general intelligence, in this particular domain. As the authors say, in the conclusions, “These results bring us a step closer to fulfilling a longstanding ambition of Artificial Intelligence: a general game-playing system that can master any game.”

While mastering these ancient games, AlphaZero also teaches us a few things we didn’t know about the games. For instance, that, in chess, white has a strong upper hand when playing the Ruy Lopez opening, or when playing against the French and Caro-Kann defenses. Sicilian defense, on the other hand, gives black much better chances. At least, that is what the function learned by the deep neural network obtains…

Actualization: The NY Times just published an interesting piece on this topic, with some additional information.

Kill the baby or the grandma?

What used to be an arcane problem in philosophy and ethics, The Trolley Problem, has been taking center stage in the discussions about the way autonomous vehicles should behave in the case of an accident. As reported previously in this blog, a website created by MIT researchers, The Moral Machine, gave everyone the opportunity to confront him or herself with the dilemmas that an autonomous car may have to face when deciding what action to take in the presence of an unavoidable accident.

The site became so popular that it was possible to gather more than 40 million decisions, from people in 233 countries and territories. The analysis of this massive amount of data was just published in an article in the journal Nature. In the site, you are faced with a simple choice. Drive forward, possibly killing some pedestrians or vehicle occupants, or swerve left, killing a different group of people. From the choices made by millions of persons, it is possible to derive some general rules of how ethics commands people to act, when faced with the difficult choice of who to kill and who to spare.

The results show some clear choices, but also that some decisions vary strongly with the culture of the person in charge. In general, people decide to protect babies, youngsters and pregnant women, as well as doctors (!). At the bottom of the preference scale are old people, animals and criminals. 

Images: from the original article in Nature.

The Second Machine Age

The Second Machine Age, by Erik Brynjolfsson and Andrew McAfee, two MIT professors and researchers, offers mostly an economist’s point of view on the consequences of the technological changes that are remaking civilisation.

Although a fair number of chapters is dedicated to the technological innovations that are shaping the first decades of the 21st century, the book is at its best when the economic issues are presented and discussed.

The book is particularly interesting in its treatment of the bounty vs. spread dilema: will economic growth be fast enough to lift everyone’s standard of living, or will increased concentration of wealth lead to such an increase in inequality that many will be left behind?

The chapter that provides evidence on the steady increase in inequality is specially appealing and convincing. While average income, in the US, has been increasing steadily in the last decades, median income (the income of those who are exactly in the middle of the pay scale) has stagnated for several decades, and may even be decreasing in the last few years. For the ones at the bottom at the scale, the situation is much worst now than decades ago.

Abundant evidence of this trend also comes from the analysis of the shares of GDP that are due to wages and to corporate profits. Although these two fractions of GDP have fluctuated somewhat in the last century, there is mounting evidence that the fraction due to corporate profits is now increasing, while the fraction due to wages is decreasing.

All this evidence, put together, leads to the inevitable conclusion that society has to explicitly address the challenges posed by the fourth industrial revolution.

The last chapters are, indeed, dedicated to this issue. The authors do not advocate a universal basic income, but come out in defence of a negative income tax for those whose earnings are below a given level. The mathematics of the proposal are somewhat unclear but, in the end, one thing remains certain: society will have to address the problem of mounting inequality brought in by technology and globalisation.