Tag: machine learning

John Hopfield, Geoffrey Hinton Awarded Physics Nobel For Research Into Machine Learning

The prize committee noted that machine learning “has long been important for research, including the sorting and analysis of vast amounts of data.”

John Hopfield and Geoffrey Hinton have been awarded the Nobel Prize for Physics for their research into enabling machine learning with artificial neural networks.

It follows yesterday’s awarding of the Nobel Prize in Physiology to Victor Ambros and Gary Ruvkun. The award for chemistry will be announced on Wednesday (October 9).

The Nobel Prize in Physics has been awarded 118 times to 227 recipients from 1901 to 2024. John Bardeen is the only person to win it twice (in 1956 and 1972), so 226 individuals have received the prize.

In announcing the award, the committee led by Hans Ellegren noted that machine learning “has long been important for research, including the sorting and analysis of vast amounts of data.”

Hinton: ‘Machine learning will exceed people in intellectual abilities’

British-Canadian Nobel laureate, computer scientist and cognitive psychologist Geoffrey Hinton spoke to the press shortly after the winners were announced.

“I am flabbergasted, I had no idea this would happen, I am very surprised,” Hinton said, when asked how he felt about being a Nobel laureate. He assures that advancements in neural networks will have a huge influence.

“This will be comparable with the industrial revolution. Machine learning will exceed people in intellectual abilities,” he added.

While he listed the numerous applications, such as in healthcare, AI assistants and increase in work productivity, he also pointed out that Machine Learning poses a threat that things could get out of control.

Hinton admitted to using ChatGPT 4 a lot. “I don’t totally trust it, as sometimes it can hallucinate,” he added.

“I am in a cheap hotel in California without very good internet or phone connection,” he said, as to where he was when he received the news.

AI, machine learning and deep learning – simply explained

Terms like machine learning, AI and deep learning were used heavily at the Nobel Prize announcements. Advancements in computer science have led to extensive research in these fields, Ellegren, the secretary general of the Royal Swedish Academy of Sciences, said.

The tech company IBM describes AI as the umbrella term for machines that mimic human intelligence.

Meanwhile, machine learning is a subset of AI. It focuses on improving AI systems by teaching them to learn from data and make better predictions.

Deep learning, which is the focus of Hopfield’s and Hinton’s research, is a more powerful version of machine learning. Deep learning uses deeper layers of neural networks.

Neural networks are the building blocks of deep learning models, just like neurons are the building blocks in the human nervous system.

Neural networks form the core of deep learning. These are made up of layers of nodes like neurons in the brain. A simple neural network has only a few layers, but a deep learning model must have more than three layers, which gives it the power to solve more complex problems.

‘Godfather of AI’: Who is Geoffrey Hinton?

Hailed as the ‘Godfather of AI’ and a pioneer in that field, Geoffrey Hinton has expressed regret about his role in advancing AI, particularly regarding its potential future impacts.

“If I hadn’t done it, somebody else would have,” he told the New York Times last year.

In 2017, the 76-year-old co-founded the Vector Institute in Toronto and became its chief scientific adviser.

A year later, in 2018, Hinton, along with Yoshua Bengio and Yann LeCun, received the prestigious Turing Award, often called the “Nobel Prize of Computing”, for their ground-breaking work in deep learning.

The trio, dubbed the “Godfathers of Deep Learning”, continued to give public talks together.

In May 2023, Hinton publicly resigned from his position at Google, where he had worked for over a decade. He stepped down to freely express his concerns about the risks associated with AI, including its potential misuse, job displacement and existential threats from advanced AI systems.

He emphasised the need for collaboration among AI developers to establish safety guidelines and prevent harmful outcomes.

This article was originally published on DW.

Alan Turing and the Limits of Computation

Turing wore many hats during his lifetime. But his fundamental contribution to the development of computing as a mathematical discipline is possibly where his significant scientific impact persists to date.

Note: June 23 is Alan Turing’s birth anniversary.

Alan Turing wore many scientific hats in his lifetime: a code-breaker in World War II, a prophetic figure of artificial intelligence (AI), a pioneer of theoretical biology, and a founding figure of theoretical computer science. While the former of his roles continue to catch the fancy of popular culture, his fundamental contribution to the development of computing as a mathematical discipline is possibly where his significant scientific impact persists to date.

Turing’s work emerged from a long legacy in the history of ideas and tools. From the times of the Sumerian abacus, we have a recorded history of attempts to create tools which could simulate at least certain aspects of human reasoning. For the most part of history, they remained rather rudimentary in technology and limited mostly to arithmetical calculations.

Leibniz’s calculator

In 1694, the German polymath Gottfried Wilfred Leibniz invented a mechanical calculator which could perform all four arithmetical operations called the Stepped Reckoner. It may be noteworthy that this invention preceded the industrial revolution and the invention of the steam engine, but was during the early phase of the rise of capitalism in Europe, which brought with it both a desire and demand for technology to aid production.

In the spirit of his times, Leibniz believed that all human reason could be mechanised. He foresaw as a precondition to this mechanisation, a universal formal language in which human thought could be expressed precisely, and unambiguously, which he termed characteristic universalis. Assuming the existence of such a formal language, the validity of any argument can thus be checked by a calculating machine, which treats symbols of the language like a calculator treats arithmetical expressions and arguments like calculations. In Leibniz’s words in The Art of Discovery: 

“The only way to rectify our reasonings is to make them as tangible as those of the Mathematicians, so that we can find our error at a glance, and when there are disputes among persons, we can simply say: Let us calculate [calculemus], without further ado, to see who is right.”

Leibniz’s Stepped reckoner. Photo: Kolossos/Wikimedia Commons, CC BY-SA 3.0

While such a machine, or even such language remained elusive in Leibniz’s lifetime, it sparked off lines of enquiry which persisted for centuries. One of its key contributions was the attention it drew to the fact that any attempt to mechanise reason necessitates a language and set of rules of reasoning which are precise, formal and unambiguous. Nowhere else did the attempt to make the language of expression and methods of reasoning- precise and formal find as much success as it did in the field of Mathematics during the late 19th and the early 20th centuries, following the development of set theory, logic and philosophy of mathematics.

In this intellectual context, David Hilbert and William Ackermann, two mathematicians asked whether there exists an algorithm which takes as input a certain statement and returns ‘True’ or ‘False’, depending on whether the statement holds true in every axiomatic system. This problem was later famously referred to as Hilbert’s Entscheidungsproblem. 

To answer such a question in the affirmative would imply producing an algorithm to achieve the task. But to answer with denial would mean to prove the impossibility of such an algorithm – not merely that such an algorithm does not currently exist, but such an algorithm can never be discovered. The proof of such an impossibility would necessarily have to take the form of mathematical proof by logical deduction.

There was, however, another bridge to be crossed before either attempting such a proof or an algorithm. Answering such a question would need a definition of algorithm, in the language of mathematics and which would make it amenable to mathematical reasoning through the means of proof. An algorithm, informally, can be described as a finite sequence of rigorous instructions to solve a particular problem or a general method precisely stated that leads to a solution of a problem.

Algorithms were long known before any mechanical device on which they can be implemented arrived in the picture. Since the algorithm has to be understood as a general method, rather than a technological entity, it was for the most part a procedural guide for humans to arrive at solutions of problems.

For instance, Euclid’s algorithm from the 3rd Century BC can be used to find the greatest common divisor (GCD) of two numbers, the Persian Mathematician Al-Khwarizmi (referred to as Algorithmus in Latin) defined methods to solve algebraic equations and Indian mathematicians were known for using such methods to solve algebraic questions. Lady Ada Lovelace, wrote the first computer programme a century before an actual computer was invented.

However, in order to reason about the existence or non-existence of an algorithm, the first critical step was to formally define it in a way that encompasses all known understanding of the algorithm and yet is precise enough to lend itself to expression as a purely mathematical object. Such a definition had to be independent of any present or future technology on which the algorithm is implemented. In fact, the definition has to be further independent of the fact whether the algorithm is employed by a human or a machine. Thus lending the definition to all possible algorithms which can be implemented on all possible machines in the present or in future. 

Turing’s contribution

This was Turing’s first major significant contribution in his seminal paper titled ‘On Computable Numbers with an application to the ENTSCHEIDUNGSPROBLEM’: A formal definition of an algorithm, which is now called the Turing Machine. His definition drew its key insights through his observation about how mathematicians perform their calculations: First and foremost, one can realise that any calculation done on a paper with length and breadth, can be done on a single tape divided into cells. Second, every computational task in mathematics involves starting with some terms and rewriting them according to prespecified rules and a state which depends on the current state of calculation, and the states themselves undergo transitions depending on prespecified transition rules. For instance, consider the addition: 13 + 27. It can be carried out on a single tape. The act of adding them involves writing a new number which is obtained by first adding the right-most digits of both the numbers (=10), and subsequently the left-most digits. However adding the leftmost digits of both numbers will involve a carry-over operation, which can be regarded as addition in a state where conventional addition will be replaced with an operation which consists of addition of digits followed by a subsequent increase of the final digit by 1. Such information used when processing the numbers is usually the information encoded in the abstraction termed a state. 

A physical Turing machine model. A true Turing machine would have unlimited tape on both sides, however, physical models can only have a finite amount of tape. Photo: Rocky Acosta/Wikimedia Commons, CC BY 3.0

A Turing Machine thus consists of a single tape divided into many cells, a set of states, a set of rules, a head which can move left or right, and can write, rewrite or erase entries in a cell depending on the current entry in the cell and the current state of the machine. While this may seem more like a description of a machine than a mathematical object, the beauty is that this mechanical device can be easily described mathematically using the basic vocabulary of sets and functions. Turing realised that any algorithmic or computable process can be simulated using this abstract model. 

One could argue that this is a specific model of computation, based on certain abstractions and simplifications. As often observed in mathematical modelling, a single phenomenon can lend itself to multiple mathematical models depending on what is retained and what gets abstracted away in the model. What lends credibility to this particular model is that Alonzo Church independently and roughly around the same time arrived at an entirely different model of computation using a more abstract and mathematical notion of computation in terms of lambda calculus. Surprisingly the two distinct definitions of computation inspired by the idea of modelling different notions of computation proved to be equivalent. This led to what is called the Church-Turing thesis, which asserts that our intuitive understanding of computability coincides with the formal definitions laid down by Church and Turing. 

Using the notion of computation, defined by Turing, went on to prove that there exist certain statements which when entered as an input, cannot be answered by any Turing Machine. Several variants of this statement, popularly known as Turing’s halting problem are widely known and several excellent references can be found online. (Martin Davis’s excellent exposition in an essay is highly suggested). The heart of the argument relies on the fact that if we assume that every problem lends itself to an algorithmic, we arrive at a logical contradiction. A simplified version of the statement would be whether there exists an algorithm which takes as input any algorithm (say algorithm A) and an input I, and determine whether algorithm A would halt on input I. Such a statement allows for self-referentiality thus leading to a paradox.  This solved the ENTSCHEIDUNGSPROBLEM, by showing the existence of a statement whose truth value cannot be determined by an algorithm.

While Turing’s approach and solution to the ENTSCHEIDUNGSPROBLEM were remarkable by themselves, its consequences have been equally illuminating. An important consequence is that there are several theoretical limitations to computation. These are limits which do not depend on physical models of computer, but on the very idea of computation. Neither are these empirical facts, which can change with growing evidence but will hold true as long as our basic assertions about what are valid rules of logic remain the same. Equally interesting is that the statement discovered by Turing was no longer the unique statement whose truth value did not lend itself to an algorithmic solution. The problem encoded by such statements is termed an ‘undecidable problem’. Many distinct problems were discovered which were proved out be undecidable, though quite often they can be directly or indirectly reduced to Turing’s halting problem. 

An interesting example of an undecidable problem is the Word Problem in String Rewriting. Consider a game, where the task is to obtain one word from another by replacing a part of the word with a sequence of alphabets at each step, according to a given set of instructions (substitution rules). By words we mean any string of alphabets strung together and not just words in the dictionary. Let us call any such game, an instance of string rewriting. As an example, we consider a particular instance of the game, where our input consists of the following set of instructions:

  1. Replace bit in a word with bot.
  2. Replace arb in a word with v.
  3. Replace tra a word with le.

Let us assume given these instructions, we have to find out if you can derive the word volery from the word arbitrary. As it happens, in this case, we may be able to convert arbitrary to volery. arbitrary -> arbotrary -> votrary -> volery, which employs the rules 3, 1, 2 and 4 respectively. If the two given words are arbitrary and zebra, one can show that it is impossible to derive zebra from arbitrary, since there is no instruction in which the letter z occurs on the right-hand side.  

A general instance of this problem could be stated as: given a set of substitution rules, similar to the ones defined above, and two words, does there exist an algorithm by which we can answer if the one word can be derived from another through the substitutions allowed by the input rules. While prima facie, it does feel like a problem that can be solved easily by a computer programme, the answer is No. This problem was proven (independently) by Emil Post and A.A. Markov in 1947 to belong to the class of undecidable problems, and thus is provably impossible to tackle by means of an algorithm!

The class of undecidable problems, many of which continue to be discovered, which shed light on the limits of computations perhaps proved to be one of Turing’s most lasting legacy in the field of computer science. In a time, when every progress in machine-learning based AI, is greeted with either a sense of euphoria or despair, and the scientific significance gets mystified and eclipsed by a combination of marketing jargon and corporate soothsaying, Turing’s work may be the sobering dose we need to remind ourselves that there are indeed fundamental limits to computation. Perhaps like his work and life, Alan Turing’s eventual death, following the persecution he faced for his sexuality, may also serve as a reminder of all the curious minds lost to science and careers cut short owing to social prejudices, which continue to persist albeit in different forms.   

T.V.H. Prathamesh is an assistant professor of computer science at Krea University, Sri City, Andhra Pradesh.

Elon Musk and Silicon Valley’s Overreliance on Artificial Intelligence

Not knowing or caring how machine learning works, what it can or can’t do, and where its application can be problematic could lead society to significant peril.

When the richest man in the world is being sued by one of the most popular social media companies, it’s news. But while most of the conversation about Elon Musk’s attempt to cancel his $44 billion contract to buy Twitter is focusing on the legal, social, and business components, we need to keep an eye on how the discussion relates to one of tech industry’s most buzzy products: artificial intelligence.

The lawsuit shines a light on one of the most essential issues for the industry to tackle: What can and can’t AI do, and what should and shouldn’t AI do? The Twitter v Musk contretemps reveals a lot about the thinking about AI in tech and startup land – and raises issues about how we understand the deployment of the technology in areas ranging from credit checks to policing.

future tense logo

At the core of Musk’s claim for why he should be allowed out of his contract with Twitter is an allegation that the platform has done a poor job of identifying and removing spam accounts. Twitter has consistently claimed in quarterly filings that less than 5% of its active accounts are spam; Musk thinks it’s much higher than that. From a legal standpoint, it probably doesn’t really matter if Twitter’s spam estimate is off by a few percent, and Twitter’s been clear that its estimate is subjective and that others could come to different estimates with the same data. That’s presumably why Musk’s legal team lost in a hearing on July 19 when they asked for more time to perform detailed discovery on Twitter’s spam-fighting efforts, suggesting that likely isn’t the question on which the trial will turn.

Regardless of the legal merits, it’s important to scrutinise the statistical and technical thinking from Musk and his allies. Musk’s position is best summarised in his filing from July 15, which states: “In a May 6 meeting with Twitter executives, Musk was flabbergasted to learn just how meager Twitter’s process was.” Namely: “Human reviewers randomly sampled 100 accounts per day (less than 0.00005% of daily users) and applied unidentified standards to somehow conclude every quarter for nearly three years that fewer than 5% of Twitter users were false or spam.” The filing goes on to express the flabbergastedness of Musk by adding, “That’s it. No automation, no AI, no machine learning.”

Perhaps the most prominent endorsement of Musk’s argument here came from venture capitalist David Sacks, who quoted it while declaring, “Twitter is toast.” But there’s an irony in Musk’s complaint here: If Twitter were using machine learning for the audit as he seems to think they should, and only labeling spam that was similar to old spam, it would actually produce a lower, less-accurate estimate than it has now.

There are three components to Musk’s assertion that deserve examination: his basic statistical claim about what a representative sample looks like, his claim that the spam-level auditing process should automated or use “AI” or “machine learning,” and an implicit claim about what AI can actually do.

On the statistical question, this is something any professional anywhere near the machine learning space should be able to answer (so can many high school students). Twitter uses a daily sampling of accounts to scrutinise a total of 9,000 accounts per quarter (averaging about 100 per calendar day) to arrive at its under-5% spam estimate. Though that sample of 9,000 users per quarter is, as Musk notes, a very small portion of the 229 million active users the company reported in early 2022, a statistics professor (or student) would tell you that that’s very much not the point. Statistical significance isn’t determined by what percentage of the population is sampled but simply by the actual size of the sample in question. As Facebook whistleblower Sophie Zhang put it, you can make the comparison to soup: It “doesn’t matter if you have a small or giant pot of soup, if it’s evenly mixed you just need a spoonful to taste-test.”

The whole point of statistical sampling is that you can learn most of what you need to know about the variety of a larger population by studying a much-smaller but decently sized portion of it. Whether the person drawing the sample is a scientist studying bacteria, or a factory quality inspector checking canned vegetables, or a pollster asking about political preferences, the question isn’t “what percentage of the overall whole am I checking,” but rather “how much should I expect my sample to look like the overall population for the characteristics I’m studying?” If you had to crack open a large percentage of your cans of tomatoes to check for their quality, you’d have a hard time making a profit, so you want to check the fewest possible to get within a reasonable range of confidence in your findings.

Also read: Why Understanding This ’60S Sci-Fi Novel Is Key to Understanding Elon Musk

While this thinking does go against the grain of certain impulses (there’s a reason why many people make this mistake), there is also a way to make this approach to sampling more intuitive. Think of the goal in setting sample size as getting a reasonable answer to the question, “If I draw another sample of the same size, how different would I expect it to be?” A classic approach to explaining this problem is to imagine you’ve bought a great mass of marbles, that are supposed to come in a specific ratio: 95% purple marbles and 5% yellow marbles. You want to do a quality inspection to ensure the delivery is good, so you load them into one of those bingo game hoppers, turn the crank, and start counting the marbles you draw, in each color. Let’s say your first sample of 20 marbles has 19 purple and one yellow; should you be confident that you got the right mix from your vendor? You can probably intuitively understand that the next 20 random marbles you draw could end up being very different, with zero yellows or seven. But what if you draw 1,000 marbles, around the same as the typical political poll? What if you draw 9,000 marbles? The more marbles you draw, the more you’d expect the next drawing to look similar, because it’s harder to hide random fluctuations in larger samples.

There are online calculators that can let you run the numbers yourself. If you only draw 20 marbles and get one yellow, you can have 95% confidence that the yellows would be between 0.13% and 24.9% of the total – not very exact. If you draw 1,000 marbles and get 50 yellows, you can have 95% confidence that yellows would be between 3.7% and 6.5% of the total; closer, but perhaps not something you’d sign your name to in a quarterly filing. At 9,000 marbles with 450 yellow, you can have 95% confidence the yellows are between 4.56% and 5.47%; you’re now accurate to within a range of less than half a percent, and at that point Twitter’s lawyers presumably told them they’d done enough for their public disclosure.

Printed Twitter logos are seen in this picture illustration taken April 28, 2022. Photo: Reuters/Dado Ruvic/Illustration/File Photo

This reality – that statistical sampling works to tell us about large populations based on much-smaller samples – underpins every area where statistics is used, from checking the quality of the concrete used to make the building you’re currently sitting in, to ensuring the reliable flow of internet traffic to the screen you’re reading this on.

It’s also what drives all current approaches to artificial intelligence today. Specialists in the field almost never use the term “artificial intelligence” to describe their work, preferring to use “machine learning.” But another common way to describe the entire field as it currently stands is “applied statistics.” Machine learning today isn’t really computers “thinking” in anything like what we assume humans do (to the degree we even understand how humans think, which isn’t a great degree); it’s mostly pattern-matching and -identification, based on statistical optimisation. If you feed a convolutional neural network thousands of images of dogs and cats and then ask the resulting model to determine if the next image is of a dog or a cat, it’ll probably do a good job, but you can’t ask it to explain what makes a cat different from a dog on any broader level; it’s just recognising the patterns in pictures, using a layering of statistical formulas.

Stack up statistical formulas in specific ways, and you can build a machine learning algorithm that, fed enough pictures, will gradually build up a statistical representation of edges, shapes, and larger forms until it recognises a cat, based on the similarity to thousands of other images of cats it was fed. There’s also a way in which statistical sampling plays a role: You don’t need pictures of all the dogs and cats, just enough to get a representative sample, and then your algorithm can infer what it needs to about all the other pictures of dogs and cats in the world. And the same goes for every other machine learning effort, whether it’s an attempt to predict someone’s salary using everything else you know about them, with a boosted random forests algorithm, or to break down a list of customers into distinct groups, in a clustering algorithm like a support vector machine.

You don’t absolutely have to understand statistics as well as a student who’s recently taken a class in order to understand machine learning, but it helps. Which is why the statistical illiteracy paraded by Musk and his acolytes here is at least somewhat surprising.

But more important, in order to have any basis for overseeing the creation of a machine-learning product, or to have a rationale for investing in a machine-learning company, it’s hard to see how one could be successful without a decent grounding in the rudiments of machine learning, and where and how it is best applied to solve a problem. And yet, team Musk here is suggesting they do lack that knowledge.

Once you understand that all machine learning today is essentially pattern-matching, it becomes clear why you wouldn’t rely on it to conduct an audit such as the one Twitter performs to check for the proportion of spam accounts. “They’re hand-validating so that they ensure it’s high-quality data,” explained security professional Leigh Honeywell, who’s been a leader at firms like Slack and Heroku, in an interview. She added, “any data you pull from your machine learning efforts will by necessity be not as validated as those efforts.” If you only rely on patterns of spam you’ve already identified in the past and already engineered into your spam-detection tools, in order to find out how much spam there is on your platform, you’ll only recognise old spam patterns, and fail to uncover new ones.

Also read: India Versus Twitter Versus Elon Musk Versus Society

Where Twitter should be using automation and machine learning to identify and remove spam is outside of this audit function, which the company seems to do. It wouldn’t otherwise be possible to suspend half a million accounts every day and lock millions of accounts each week, as CEO Parag Agrawal claims. In conversations I’ve had with cybersecurity workers in the field, it’s quite clear that large amounts of automation is used at Twitter (though machine learning specifically is actually relatively rare in the field because the results often aren’t as good as other methods, marketing claims by allegedly AI-based security firms to the contrary).

At least in public claims related to this lawsuit, prominent Silicon Valley figures are suggesting they have a different understanding of what machine learning can do, and when it is and isn’t useful. This disconnect between how many nontechnical leaders in that world talk about “AI,” and what it actually is, has significant implications for how we will ultimately come to understand and use the technology.

The general disconnect between the actual work of machine learning and how it’s touted by many company and industry leaders is something data scientists often chalk up to marketing. It’s very common to hear data scientists in conversation among themselves declare that “AI is just a marketing term.” It’s also quite common to have companies using no machine learning at all describe their work as “AI” to investors and customers, who rarely know the difference or even seem to care.

This is a basic reality in the world of tech. In my own experience talking with investors who make investments in “AI” technology, it’s often quite clear that they know almost nothing about these basic aspects of how machine learning works. I’ve even spoken to CEOs of rather large companies that rely at their core on novel machine learning efforts to drive their product, who also clearly have no understanding of how the work actually gets done.

Not knowing or caring how machine learning works, what it can or can’t do, and where its application can be problematic could lead society to significant peril. If we don’t understand the way machine learning actually works – most often by identifying a pattern in some dataset and applying that pattern to new data – we can be led deep down a path in which machine learning wrongly claims, for example, to measure someone’s face for trustworthiness (when this is entirely based on surveys in which people reveal their own prejudices), or that crime can be predicted (when many hyperlocal crime numbers are highly correlated with more police officers being present in a given area, who then make more arrests there), based almost entirely on a set of biased data or wrong-headed claims.

If we’re going to properly manage the influence of machine learning on our society – on our systems and organisations and our government – we need to make sure these distinctions are clear. It starts with establishing a basic level of statistical literacy, and moves on to recognising that machine learning isn’t magic—and that it isn’t, in any traditional sense of the word, “intelligent”– that it works by pattern-matching to data, that the data has various biases, and that the overall project can produce many misleading and/or damaging outcomes.

It’s an understanding one might have expected – or at least hoped – to find among some of those investing most of their life, effort, and money into machine-learning-related projects. If even people that deep aren’t making those efforts to sort fact from fiction, it’s a poor omen for the rest of us, and the regulators and other officials who might be charged with keeping them in check.

This article was originally published on Future Tense, a partnership between Slate magazine, Arizona State University, and New America.

Is the Russian Media Joining China and Iran in Turning on Trump?

In 2016, America’s adversaries seemed to cheer electoral chaos and a withering faith in democracy. Now they seem to be hoping democracy can topple a leader they’ve grown loathe to deal with.

It can be easy to overlook how the rest of the world is making sense of America’s chaotic campaign season.

But in many cases, they’re paying attention just as closely as US voters are. After all, who wins the US presidency has implications for countries around the world.

Since September 22, we’ve been using machine-learning algorithms to identify the predominant themes in foreign media coverage.

How different countries cover the race between Donald Trump and Joe Biden can shed some light on how foreign citizens discern the candidates and the American political process, especially in places that have strict state control of media like China, Russia and Iran.

Unlike in the US, where there is a cacophony of perspectives, by and large the media in these three countries follow very similar narratives.

In 2016, we did the same exercise. Back then, one of the main themes that emerged was the decline of US democracy. With scandal and the disillusionment of voters dominating the headlines, America’s global competitors used the 2016 election to advance their own political narratives about US decline.

Some of these themes have emerged in the coverage of the current race. But the biggest difference is their portrayal of Trump.

The last election cycle, candidate Trump was an unknown. Although foreign nations acknowledged his political inexperience, they were cautiously optimistic about Trump’s deal-making ability. Russian media outlets were particularly bullish on Trump’s potential.

Now, however, the feelings appear to have changed. China, Iran and even Russia seem to crave a return to normalcy – and, to some extent, American leadership in the world.

Dissecting the debate

To assess how America’s competitors make sense of the 2020 campaign, we tracked over 20 prominent news outlets from Chinese, Russian and Iranian native language media. We used automatic clustering algorithms to identify key narrative themes in the coverage and sentiment analysis to track how each country viewed the candidates. We then reviewed this AI-extracted information to validate our findings.

While our results are still preliminary, they shed light on how these countries’ media outlets are portraying the two candidates. Two key moments from the 2020 campaign – the first debate and Trump’s coronavirus diagnosis – are particularly illustrative.

After the first debate, the Chinese media questioned its usefulness to voters and generally portrayed Trump’s performance in a negative light. To them, the “chaotic” back-and-forth was a sobering reflection of America’s political turbulence.

They described Trump as purposely sabotaging the debate by interrupting his opponent and, in the days after the debate, noted that his performance failed to improve his lagging poll numbers. Biden was criticised for being unable to articulate concrete policies, but was nonetheless praised for being able to avoid any major gaffes and – as an article from the Xinhua News Agency put it – responding to Trump with “fierce words”.

Chinese outlets that once relayed cautious optimism over Donald Trump’s deal-making abilities now express exasperation over his chaotic style. Photo: Reuters

Unlike in 2016, where Clinton was portrayed as anti-Russian, corrupt and elitist, Russian media appeared more willing to characterise the Democratic Party nominee in a positive light.

In fact, Russian coverage expressed surprise over Biden’s debate performance. He didn’t come across as feeble; instead, he was, as the daily newspaper Kommersant wrote, a lively opponent who appeared to be “criticising, irritating and humiliating” Trump by calling him a “liar, racist and the worst president”. They did praise Trump’s especially aggressive rhetoric. However, our analysis found that Russian media also repeatedly claimed that, unlike 2016, voters today were tiring of his bombast.

While Trump’s post-debate posturing received some positive coverage, Russian media largely lamented his administration’s failure to deliver substantive progress toward normalising relations between the two countries. They noted the debate neither clarified policies for voters nor for international observers.

Iranian media took the strongest anti-Trump stance. Reports routinely pointed out that Trump has had no foreign policy successes, and has only exacerbated relations with the country’s major rivals. According to Iranian media outlets, Trump’s lack of accomplishments has left him with no choice but to rely on insults and personal attacks.

Biden, however, was said to have kept his calm. As Al Alam News wrote, he used “more credible responses and attacks than Trump.”

The former vice president, in their view, promised some semblance of normalised diplomatic relations.

‘Intransigence’ and ‘ignorance’

The final month of the US presidential race is known for last-minute surprises that can upend the race. This year was no exception, with Trump’s Oct. 2 announcement of his COVID-19 diagnosis quickly shifting media coverage from the debate to Trump’s health.

He received little sympathy from foreign outlets. Across the board, they were quick to note how his personal disregard for public health safety measures symbolised his administration’s failed response to the pandemic.

For example, one Chinese media outlet, The Beijing News, characterised the diagnosis as “hitting” the president “in the face,” given his previous downplaying of the epidemic. Other reports claimed Trump lacked “care about the epidemic,” including disregard for “protective measures such as wearing a mask.”

Also read: America Will Vote to Heal This November. Will India Follow Suit?

Chinese outlets suggested Trump would use the diagnosis to win sympathy from voters, but also noted by being sidelined from holding campaign rallies, he could lose his “self-confessed” ability to attract voters.

Russian media, on the other hand, remained confident that Trump would recover and repeated the White House line of Trump’s good health.

At the same time, Russian outlets tended to chastise Trump’s unwillingness to avoid large gatherings, practice social distancing or wear a mask, all of which violated his administration’s basic health guidelines. Likewise, Russian reports criticised Trump’s post-diagnosis behaviour – like tweeting video messages while at the hospital and violating quarantine with his public appearances – as “publicity stunts” that jeopardised the safety of his Secret Service detail and supporters.

Again, Iranian media most directly criticised Trump. Reports characterised Trump as “determined to continue the same approach,” despite his diagnosis, and remain “without a muzzle,” “irresponsibly” continuing to tweet misinformation falsely comparing COVID-19 to the flu.

Coverage centred on Trump’s inability to, as Al Alam put it, show “any sympathy” for the over 200,000 dead Americans. This death toll, the same article noted, was attributed to Trump’s “mismanagement, intransigence, ignorance and stupidity,” highlighted by his cavalier disregard for safety guidelines such as wearing a mask.

In the bag for Biden?

Russian coverage expressed surprise over Biden’s debate performance. Photo: Reuters

Many of the criticisms of the US found in foreign media outlets in our 2016 study appear in this year’s coverage. But since the 2016 election, geopolitics have changed quite a bit – and, for many of these countries, not necessarily for the better. That might best explain their collective ire toward Trump.

During Trump’s first term, Iranians absorbed the US‘s unilateral withdrawal from the Iran nuclear deal, the reimposition of sanctions and the assassination of one of its top generals.

The Chinese entered into a trade war with the US, while the US government levelled accusations of intellectual property theft, mass murder and blame for the spread of what Trump has called the “China Virus“.

Russians, meanwhile, have seen themselves – fairly or not – bound to Trump’s 2016 election victory and outed as an international provocateur. That Trump has not been able to deliver on normalising US Russian relations despite four years of posturing and political rhetoric has perhaps made Trump more of a political liability than worthwhile ally. Not only has the COVID-19 pandemic sparked unrest in Russia’s backyard, but mounting regional instability is also undermining Putin’s image as a master tactician.

As a result, these countries’ outlets appear to have shifted attention away from a broad critique of US democracy toward exasperation with Trump’s leadership.

The two, of course, aren’t mutually exclusive. And these countries’ relatively positive characterisations of a potential Biden administration likely won’t last.

But even the country’s supposed adversaries seem to be craving a return to stability and predictability from the Oval Office.

Robert Hinck, Assistant Professor, Monmouth College; Robert Utterback, Assistant Professor of Computer Science, Monmouth College, and Skye Cooley, Assistant Professor of Communication, Oklahoma State University

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Why We Must Unshackle AI From the Boundaries of Human Knowledge

Technology alone cannot fight societal biases but we can ensure that our algorithmic offspring don’t inherit our prejudices, and in fact overcome our moral shortcomings.

Artificial intelligence (AI) has made astonishing progress in the last decade. AI can now drive cars, diagnose diseases from medical images, recommend movies, even whom you should date, make investment decisions, and create art that people have sold at auction.

A lot of research today, however, focuses on teaching AI to do things the way we do them. For example, computer vision and natural language processing – two of the hottest research areas in the field – deal with building AI models that can see like humans and use language like humans. But instead of teaching computers to imitate human thought, the time has now come to let them evolve on their own, so instead of becoming like us, they have a chance to become better than us.

Supervised learning has thus far been the most common approach to machine learning, where algorithms learn from datasets containing pairs of samples and labels. For example, consider a dataset of enquiries (not conversions) for an insurance website with information about a person’s age, occupation, city, income, etc. plus a label indicating whether the person eventually purchased the insurance. A supervised learning model trained using this dataset could calculate the probability of a new enquiry converting into a sale. Such models are very good at predicting outcomes but they have one big drawback: their performance is limited by the performance of the system that did the original labelling. And in most cases, the latter is a human being.

This limitation, and how AI can overcome it, is best illustrated by the story of how computers learned to play chess. Through most of the 20th century, AI’s critics argued that computers would never beat humans at chess because playing chess requires imagination, intuition, foresight, planning – what they called real intelligence – and not just computational ability. But in 1997, IBM’s Deep Blue chess program defeated Garry Kasparov.

Chess programs feature sophisticated algorithms that have been fed thousands of recorded games played by grandmasters. The algorithms learn how to play chess by analysing these historical games. However, because the algorithms learn from a dataset of games played by humans, their abilities are limited by the skill humans possess no matter how well they play.

In 2017, AlphaZero, a program developed by a Google subsidiary named DeepMind, entered the world of chess in spectacular fashion. Unlike its predecessors, AlphaZero had been taught only the basic rules of chess, and it taught itself to play by playing against itself. It quickly defeated Stockfish, the champion program at the time. Because AlphaZero did not learn from human games, it had developed a different style of play. In one game, it sacrificed four pawns in a row – something chess players might have found bizarre.

In a technical paper, its developers described how AlphaZero discarded strategies human grandmasters routinely used to invent new ones nobody knew existed. AlphaZero’s success stems from the fact that – unlike previous algorithmic contenders – it didn’t inherit the limitations of human knowledge.

In 1953, Alan Turing wrote that one cannot program a machine to play a game better than one does oneself. Today, what we need more than anything else are computers that know better than their creators. When developing AI models that can diagnose cancer, for example, we want the models to know things we don’t. For that to happen, we must allow the models to teach themselves and free them of the boundaries of our own understanding.

In 2018, researchers found that an AI recruiting tool used by Amazon discriminated against women. The company had created this tool to crawl the web and identify potential candidates and rate them. To train the algorithm to determine a candidate’s suitability, its developers used a database of CVs submitted to the company over a 10-year period. Because Amazon, like most technology companies, employed fewer women than men, the algorithm presumed the gender imbalance was part of Amazon’s formula for success.

For another example, the COMPAS program the US government used to prescribe sentences to convicts based on the probability that they would reoffend was found to have inherited the judicial system’s racial discrimination as well.

To blame AI for the prejudices of human beings is stupid. At the same time, it is also possible to develop algorithms that are better than us at being fair. Technology alone cannot fight societal biases but we can ensure that our algorithmic offspring don’t inherit our prejudices, and in fact overcome our moral shortcomings.

Discrimination is as old as humankind; religious preaching, moral education, processes or legislation may mitigate its consequences but can’t eliminate it altogether. But today, as we increasingly cede decision-making to AI algorithms, we have a unique opportunity. For the first time in history, we have a real shot at building a fair society that is free of human prejudices by building machines that are fair by design.

Of course, this is still only a pipe dream, but our first steps in this direction should be to change our priorities: Instead of obsessing over the performance of supervised learning models on specific problems, we must develop methods that allow AI to learn without labels that humans have created.

Some researchers have already recognised this, and in the last few years have been developing new techniques to support reinforcement learning and unsupervised learning, both of which offer AI more autonomy under less human supervision. AI will soon permeate most aspects of our lives, which in turn means these developments are very important for the future of humanity.

Viraj Kulkarni has a master’s degree in computer science from UC Berkeley and is currently pursuing a PhD in artificial intelligence.

Is This the AI We Should Fear?

We scientists want to understand intelligence as well as create machines that are intelligent. We don’t just want to build machines that interact with humans in a superficial manner while pretending to be deep.

Scientists working on artificial intelligence (AI) have traditionally pursued the goal of constructing intelligent machines, with human intelligence used as a benchmark. However, philosophers like John Haugeland (author of AI: The Very Idea, 1985) have contested that the goal is to build machines with “minds of their own”. Of late, AI has also been conflated with machine learning, which is but one component of AI, as well as with data science and, more imaginatively, even the Internet of Things. But what exactly is AI?

In the middle of the last century, when AI was still in its infancy, scientists were having heated debates on the nature of intelligence and on the possibility of machine intelligence. Alan Turing, English computer scientist, sought to bring the arguments to a close with his ‘imitation game’, known today as the Turing test. Its fundamental principle was that if a machine could respond to text messages like a human could, then it must be intelligent.

Similarly, in this age of machine-learning, we will need to decide when a machine has acquired the ability  to think for itself. To answer this question we need to look within our own heads and think about what makes us intelligent. At first, the list seems endless because intelligence has to be multifaceted. Beating a human champion at chess or Go, diagnosing diseases from images better than humans can and responding to voice commands is smart, but it is not by any stretch of imagination the whole story.

An intelligent agent operates autonomously in its environment. The following diagram depicts a human agent: she senses the world around her, understands what she senses, and deliberates over what she has imbibed.

Image: Deepak Khemani

The diagram depicts the three layers of information processing. The outermost layer processes signals: the light falling on our retinas, sound impinging upon our ears, etc. The middle layer processes the incoming information, recognises patterns and assigns class labels. The innermost layer is concerned with cognition: the process of thought, memory, language, contemplation and imagination.

In this architecture, the precise nature of deliberation remains elusive. We absorb information via our sensory organs – eyes, ears, nose, touch – and we retain only what we want. Our memories have complex structures that can be divided into two, short-term and long-term, and are joined by a deep repository of subconscious knowledge. We digest what we learn in myriad ways and recall relevant bits when required. We let our minds wander when we’re idle and we often allow free rein to the imagination.

Imagination is the soul of our intelligence. We conjure fantastic worlds, imagine how things will work as a result of decisions we are yet to make, and fulfil our goals through deliberated action. Of course, we are not always in control, rather we don’t always have the sense that we are. At the same time, our exploitative behaviour – as evidenced in the crimes of humanity – is a mark of intelligence. No single human is supremely intelligent but every aspect of intelligence does manifest itself in one or the other of us.

As intelligent agents and before the advent of modern technology, we perceived the world through our senses and through oral stories. Then, devices like the postcard, newspaper, the telephone, radio and then television extended the reach of our senses. Soon, through advertisements in some of these media, we began perceiving not just what we wanted to see but also what someone else with commercial or political interests wanted us to see.

Then came the internet and the social media, both of which changed our lives in important ways. Until their advent, information had flowed only one way: from the world outside to us. But now, with each person personally seeking out information from the world wide web, the facilitator began to observe who was consuming what. The flow of information became two-way, and the observer became the observed. Data, as they say, is the new oil.

There is now lots of data available on the internet – as much about our behaviours as everything else – which has in turn triggered work in data science, analytics machine-learning and, nowadays, deep-learning, a kind of machine-learning focusing on ways to organise information instead of performing certain tasks.

The success of deep neural networks in pattern recognition and image-labelling, in particular, has been spectacular. Algorithms trained on multiple and diverse images in which, say, a horse is present, are able to label new images with a horse with considerable accuracy. Similarly, it is possible that algorithms tracking you on the internet can identify what your eyes are looking at as you’re browsing. As a result, the modern picture of our cognitive architecture looks like this:

Image: Deepak Khemani

Today’s AI only scratches the surface of cognition – the core of AI – depicted as the small blue circle. While it is the most significant part of the human cognitive system, it is relatively less important for AI as we understand it today. The machine-learning layer, depicted as the bloated outer shell, is where the action is, and its impact on our lives includes both the good and the bad.

For starters, there have been some big strides in medical diagnostics. It is now possible to combine the experience of thousands of doctors by observing how they diagnose certain conditions and find ways for machines to do that better, such as by homing in better on critical symptoms.

However, each piece data comes from a human patient who may be worried about it falling into the wrong hands. For example, medical information of this kind could be valuable to insurance companies and potential employers. Similarly, social media platforms harvest information about your likes, dislikes, preferences, leanings, inclinations, even beliefs, with the intention of selling the data to advertisers constantly on the lookout for potential customers for their products.

Social media would like you to stay addicted because the more you use the services of a platform, the more the data you generate and the more, and better, marketers are able to target you. This prompted the rise of the influencers: protagonists that thrive on the attention of the members of their social network and whose accounts advertisers use to push their ads. In fact, we’re also tracked when we’re not actually on these platforms: through various apps on our smartphones, some of which purport to be funny camera filters but ask for access to our contacts to work.

When the freebies first began – Hotmail was perhaps the pioneer – it wasn’t clear what was in it for the company. But now, almost everyone knows that when they’re being offered something for free, whether by Google or Facebook, it’s not going to be a one-way transaction. They’re going to know they themselves are the products, that their data is going to be harvested by these companies even as you use them for free.

So AI and machine-learning in their most ubiquitous form today are instruments of the capitalist pursuit of profit. They shouldn’t be confused with automation, another buzzword often uttered in the same breath. While some forms of automation use methods gleaned from AI research, such as self-driving cars and algorithmic trading in stock markets, most of it has little to do with intelligence. It is automation, and not AI as such, that is responsible for the loss of human jobs. AT does make our lives more comfortable but it also requires government regulation to ensure the wealth generated is distributed more equitably.

Nonetheless, these technologies have become quite troubling for their implications for our privacy and data ownership as well. Harvesting and exploiting data, whether for good or bad, is in the realm of data science, analytics and machine-learning. But is it AI? Perhaps not.

We scientists want to understand intelligence as well as create machines that are intelligent. We want to create companion machines in ageing societies, machines that can teach our children math or serve as perceptive personal assistants. We’d like to build a robot that cooks an exotic meal for you with recipes from the internet or even teach you how to play bridge. We don’t just want to build machines that interact with humans in a superficial manner while pretending to be deep.

In sum, the AI which we now see is only the crust of a would-be intelligent entity, but this limited version is what corporate interest lies in. Indeed, this AI is only the tip of the machine-intelligence iceberg, and the corporate world does not seem to be interested in expanding its limits to do more, do better. And it’s likely they won’t until it makes commercial sense for them to do so.

Deepak Khemani is a professor in the department of computer science, IIT Madras.

India Is Falling Down the Facial Recognition Rabbit Hole

Its use as an effective law enforcement tool is overstated, while the underlying technology is deeply flawed.

In a discomfiting reminder of how far technology can be used to intrude on the lives of individuals in the name of security, the Ministry of Home Affairs, through the National Crime Records Bureau, recently put out a tender for a new Automated Facial Recognition System (AFRS). 

The stated objective of this system is to “act as a foundation for a national level searchable platform of facial images,” and to “[improve] outcomes in the area of criminal identification and verification by facilitating easy recording, analysis, retrieval and sharing of Information between different organizations.” 

The system will pull facial image data from CCTV feeds and compare these images with existing records in a number of databases, including (but not limited to) the Crime and Criminal Tracking Networks and Systems (or CCTNS), Interoperable Criminal Justice System (or ICJS), Immigration Visa Foreigner Registration Tracking (or IVFRT), Passport, Prisons, Ministry of Women and Child Development (KhoyaPaya), and state police records. 

Furthermore, this system of facial recognition will be integrated with the yet-to-be-deployed National Automated Fingerprint Identification System (NAFIS) as well as other biometric databases to create what is effectively a multi-faceted system of biometric surveillance.

It is rather unfortunate, then, that the government has called for bids on the AFRS tender without any form of utilitarian calculus that might justify its existence. The tender simply states that this system would be “a great investigation enhancer.” 

Also read: Humans Can’t Watch All the Surveillance Cameras Out There, so Computers Are

This confidence is misplaced at best. There is significant evidence that not only is a facial recognition system, as has been proposed, ineffective in its application as a crime-fighting tool, but it is a significant threat to the privacy rights and dignity of citizens. Notwithstanding the question of whether such a system would ultimately pass the test of constitutionality – on the grounds that it affects various freedoms and rights guaranteed within the constitution – there are a number of faults in the issued tender. 

Let us first consider the mechanics of a facial recognition system itself. Facial recognition systems chain together a number of algorithms to identify and pick out specific, distinctive details about a person’s face – such as the distance between the eyes, or shape of the chin, along with distinguishable ‘facial landmarks’. These details are then converted into a mathematical representation known as a face template for comparison with similar data on other faces collected in a face recognition database. There are, however, several problems with facial recognition technology that employs such methods. 

Facial recognition technology depends on machine learning – the tender itself mentions that the AFRS is expected to work on neural networks “or similar technology” –  which is far from perfect. At a relatively trivial level, there are several ways to fool facial recognition systems, including wearing eyewear, or specific types of makeup. The training sets for the algorithm itself can be deliberately poisoned to recognise objects incorrectly, as observed by students at MIT. 

More consequentially, these systems often throw up false positives, such as when the face recognition system incorrectly matches a person’s face (say, from CCTV footage) to an image in a database (say, a mugshot), which might result in innocent citizens being identified as criminals. In a real-time experiment set in a train station in Mainz, Germany, facial recognition accuracy ranged from 17-29% – and that too only for faces seen from the front – and was at 60% during the day but 10-20% at night, indicating that environmental conditions play a significant role in this technology.

Also read: Could Super Recognisers Be the Latest Weapon in the War on Terror?

Facial recognition software used by the UK’s Metropolitan Police has returned false positives in more than 98% of match alerts generated.

When the American Civil Liberties Union (ACLU) used Amazon’s face recognition system, Rekognition, to compare images of legislative members of the American Congress with a database of mugshots, the results included 28 incorrect matches.

There is another uncomfortable reason for these inaccuracies – facial recognition systems often reflect the biases of the society they are deployed in, leading to problematic face-matching results. Technological objectivity is largely a myth, and facial recognition offers a stark example of this. 

An MIT study shows that existing facial recognition technology routinely misidentifies people of darker skin tone, women and young people at high rates, performing better on male faces than female faces (8.1% to 20.6% difference in error rate), lighter faces than darker faces (11.8% to 19.2% difference in error rate) and worst on darker female faces (20.8% to 34.7% error rate). In the aforementioned ACLU study, the false matches were disproportionately people of colour, particularly African-Americans. The bias rears its head when the parameters of machine-learning algorithms, derived from labelled data during a “supervised learning” phase, adhere to socially-prejudiced ideas of who might commit crimes. 

The implications for facial recognition are chilling. In an era of pervasive cameras and big data, such prejudice can be applied at unprecedented scale through facial recognition systems. By replacing biased human judgment with a machine learning technique that embeds the same bias, and more reliably, we defeat any claims of technological neutrality. Worse, because humans will assume that the machine’s “judgment” is not only consistently fair on average but independent of their personal biases, they will read agreement of its conclusions with their intuition as independent corroboration. 

In the Indian context, consider that Muslims, Dalits, Adivasis and other SC/STs are disproportionately targeted by law enforcement. The NCRB in its 2015 report on prison statistics in India recorded that over 55% of the undertrials prisoners in India are either Dalits, Adivasis or Muslims, a number grossly disproportionate to the combined population of Dalits, Adivasis and Muslims, which amounts to just 39% of the total population according to the 2011 Census.

If the AFRS is thus trained on these records, it would clearly reinforce socially-held prejudices against these communities, as inaccurately representative as they may be of those who actually carry out crimes. The tender gives no indication that the developed system would need to eliminate or even minimise these biases, nor if the results of the system would be human-verifiable.

This could lead to a runaway effect if subsequent versions of the machine-learning algorithm are trained with criminal convictions in which the algorithm itself played a causal role. Taking such a feedback loop to its logical conclusion, law enforcement may use machine learning to allocate police resources to likely crime spots – which would often be in low income or otherwise vulnerable communities.

Adam Greenfield writes in Radical Machines on the idea of ‘over transparency,’ that combines “bias” of the system’s designers as well of the training sets – based as these systems are on machine learning – and “legibility” of the data from which patterns may be extracted. The “meaningful question,” then, isn’t limited to whether facial recognition technology works in identification – “[i]t’s whether someone believes that they do, and acts on that belief.”

The question thus arises as to why the MHA/NCRB believes this is an effective tool for law enforcement. We’re led, then, to another, larger concern with the AFRS – that it deploys a system of surveillance that oversteps its mandate of law enforcement. The AFRS ostensibly circumvents the fundamental right to privacy, as ratified by the Supreme Court in 2018, through sourcing its facial images from CCTV cameras installed in public locations, where the citizen may expect to be observed. 

Also read: Old Isn’t Always Gold: FaceApp and Its Privacy Policies

The extent of this surveillance is made even clearer when one observes the range of databases mentioned in the tender for the purposes of matching with suspects’ faces extends to “any other image database available with police/other entity” besides the previously mentioned CCTNS, ICJS et al. The choice of these databases makes overreach extremely viable.

This is compounded when we note that the tender expects the system to “[m]atch suspected criminal face[sic] from pre-recorded video feeds obtained from CCTVs deployed in various critical identified locations, or with the video feeds received from private or other public organization’s video feeds.” There further arises a concern with regard to the  process of identification of such “critical […] locations,” and if there would be any mechanisms in place to prevent this from being turned into an unrestrained system of surveillance, particularly with the stated access to private organisations’ feeds.

The Perpetual Lineup report by Georgetown Law’s Center on Privacy & Technology identifies real-time (and historic) video surveillance as posing a very high risk to privacy, civil liberties and civil rights, especially owing to the high-risk factors of the system using real-time dragnet searches that are more or less invisible to the subjects of surveillance.

It is also designated a “Novel Use” system of criminal identification, i.e., with little to no precedent as compared to fingerprint or DNA analysis, the latter of which was responsible for countless wrongful convictions during its nascent application in the science of forensic identification, which have since then been overturned.

In the Handbook of Face Recognition, Andrew W. Senior and Sharathchandra Pankanti identify a more serious threat that may be born out of automated facial recognition, assessing that “these systems also have the potential […] to make judgments about [subjects’] actions and behaviours, as well as aggregating this data across days, or even lifetimes,”  making video surveillance “an efficient, automated system that observes everything in front of any of its cameras, and allows all that data to be reviewed instantly, and mined in new ways” that allow constant tracking of subjects.

Such “blanket, omnivident surveillance networks” are a serious possibility through the proposed AFRS. Ye et al, in their paper on “Anonymous biometric access control”, show how automatically captured location and facial image data obtained from cameras designed to track the same can be used to learn graphs of social networks in groups of people.

Consider those charged with sedition or similar crimes, given that the CCTNS records the details as noted in FIRs across the country. Through correlating the facial image data obtained from CCTVs across the country – the tender itself indicates that the system must be able to match faces obtained from two (or more) CCTVs – this system could easily be used to target the movements of dissidents moving across locations.

Constantly watched

Further, something which has not been touched upon in the tender – and which may ultimately allow for a broader set of images for carrying out facial recognition – is the definition of what exactly constitutes a ‘criminal’. Is it when an FIR is registered against an individual, or when s/he is arrested and a chargesheet is filed? Or is it only when an individual is convicted by a court that they are considered a criminal?

Additionally, does a person cease to be recognised by the tag of a criminal once s/he has served their prison sentence and paid their dues to society? Or are they instead marked as higher-risk individuals who may potentially commit crimes again? It could be argued that such a definition is not warranted in a tender document, however, these are legitimate questions which should be answered prior to commissioning and building a criminal facial recognition system.

Senior and Pankanti note the generalised metaphysical consequences of pervasive video surveillance in the Handbook of Face Recognition: 

“the feeling of disquiet remains [even if one hasn’t committed a major crime], perhaps because everyone has done something “wrong”, whether in the personal or legal sense (speeding, parking, jaywalking…) and few people wish to live in a society where all its laws are enforced absolutely rigidly, never mind arbitrarily, and there is always the possibility that a government to which we give such powers may begin to move towards authoritarianism and apply them towards ends that we do not endorse.”

Such a seemingly apocalyptic scenario isn’t far-fetched. In the section on ‘Mandatory Features of the AFRS’, the system goes a step further and is expected to integrate “with other biometric solution[sic] deployed at police department system like Automatic Fingerprint identification system (AFIS)[sic]” and “Iris.” This form of linking of biometric databases opens up possibilities of a dangerous extent of profiling.

While the Aadhaar Act, 2016, disallows Aadhaar data from being handed over to law enforcement agencies, the AFRS and its linking with biometric systems (such as the NAFIS) effectively bypasses the minimal protections from biometric surveillance the prior unavailability of Aadhaar databases might have afforded. The fact that India does not have a data protection law yet – and the Bill makes no references to protection against surveillance either – deepens the concern with the usage of these integrated databases. 

The Perpetual Lineup report warns that the government could use biometric technology “to identify multiple people in a continuous, ongoing manner [..] from afar, in public spaces,” allowing identification “to be done in secret”. Senior and Pankanti warn of “function creep,” where the public grows uneasy as “silos of information, collected for an authorized process […] start being used for purposes not originally intended, especially when several such databases are linked together to enable searches across multiple domains.”

This, as Adam Greenfield points out, could very well erode “the effectiveness of something that has historically furnished an effective brake on power: the permanent possibility that an enraged populace might take to the streets in pursuit of justice.”

What the NCRB’s AFRS amounts to, then, is a system of public surveillance that offers little demonstrable advantage to crime-fighting, especially as compared with its costs to fundamental human rights of privacy and the freedom of assembly and association. This, without even delving into its implications with regard to procedural law. To press on with this system, then, would be indicative of the government’s lackadaisical attitude towards protecting citizens’ freedoms. 

Karan Saini is a security researcher and programme officer with the Centre for Internet and Society. Prem Sylvester is a research intern at the Centre for Internet and Society. The views expressed by the authors in this article are personal.

Scientists Use AI to Help Find Potential Bat Species That Can Carry Nipah Virus

Nipah is a deadly virus that can be transmitted to humans from the bodily fluids of infected bats.

New Delhi:  The recent outbreak of Nipah virus in Kerala, which follows one that occurred in 2018, has turned the focus back on bats, which are known to host the virus. Now, using machine learning, an international group of scientists has identified the bat species with the potential to host the Nipah virus.

Based on the traits of bats to carry the virus, researchers pointed out that more bat species in India may be reservoirs of Nipah than the only one confirmed so far.

“One of our major findings is that until now the Nipah virus presence in India was known from only one species of fruit bats – the Indian flying fox. However, our analysis reveals that at least 11 species of bats in India could be carriers of Nipah,” P.O. Nameer, head of the Centre for Wildlife Studies at the College of Forestry, Kerala Agricultural University, and a member of the research team, told India Science Wire.

The machine learning analysis is based on already published scientific studies from around the world; no samples were actually tested, but that doesn’t come in the way of the findings, which are important from the public health point of view.

“Since Nipah’s presence is now suspected from more species of bats, including some species of insect bats, we need detailed studies by collecting samples. Such studies would help in taking necessary precautions by people and in reducing chances of possible outbreaks in future,” Nameer added.

Their analysis covered 523 bat species and 48 traits, such as foraging methods, diet, migration behaviour, geographic spread, reproduction and environmental conditions. Of the 523 species, 31 are found in India and 11 of them have been found to host the Nipah virus in previous studies. The machine’s algorithm could identify bat species already known to carry Nipah with 83% accuracy. It also identified six other species in Asia, Australia and Oceana with traits that make their bodies conducive to hosting the virus. Four of these species also occur in India, two of which are found in Kerala.

“Our work provides a list of species to guide early surveillance and should not be taken as a definitive list of reservoirs,” the researchers cautioned in their paper describing their results, published on June 27, 2019. “A series of further studies are required to triangulate on the reservoir hosts that pose a risk to humans.”

Nipah is a deadly virus that can be transmitted to humans from the bodily fluids of infected bats. When such bats feed on fruits or date palms, they contaminate the fruits. Once someone gets infected, she can spread the virus to others through contact. Domestic pigs have also been identified as ‘bridging hosts’ in some cases.

The research team included Raina K. Plowright, Daniel E. Crowley and Alex D. Washburne from Montana State University, Montana; Daniel J. Becker of the University of Georgia, Athens; Barbara A. Han and Tao Huang of the Cary Institute of Ecosystem Studies, New York; P.O. Nameer of the Kerala Agricultural University, Thrissur; and Emily S. Gurley of the Johns Hopkins Bloomberg School of Public Health, Maryland.

Will Humans Be Able to Comprehend Art Produced by Machines?

All conscious machines will have embodied experiences of their own, but in bodies that will be entirely alien to us.

Assuming that the emergence of consciousness in artificial minds is possible, those minds will feel the urge to create art. But will we be able to understand it? To answer this question, we need to consider two subquestions: when does the machine become an author of an artwork? And how can we form an understanding of the art that it makes?

Empathy, we argue, is the force behind our capacity to understand works of art. Think of what happens when you are confronted with an artwork. We maintain that, to understand the piece, you use your own conscious experience to ask what could possibly motivate you to make such an artwork yourself – and then you use that first-person perspective to try to come to a plausible explanation that allows you to relate to the artwork.

Your interpretation of the work will be personal and could differ significantly from the artist’s own reasons, but if we share sufficient experiences and cultural references, it might be a plausible one, even for the artist. This is why we can relate so differently to a work of art after learning that it is a forgery or imitation: the artist’s intent to deceive or imitate is very different from the attempt to express something original. Gathering contextual information before jumping to conclusions about other people’s actions – in art, as in life – can enable us to relate better to their intentions.

Also read: Politics Through Art on the Internet

But the artist and you share something far more important than cultural references: you share a similar kind of body and, with it, a similar kind of embodied perspective. Our subjective human experience stems, among many other things, from being born and slowly educated within a society of fellow humans, from fighting the inevitability of our own death, from cherishing memories, from the lonely curiosity of our own mind, from the omnipresence of the needs and quirks of our biological body, and from the way it dictates the space- and time-scales we can grasp. All conscious machines will have embodied experiences of their own, but in bodies that will be entirely alien to us.

We are able to empathise with nonhuman characters or intelligent machines in human-made fiction because they have been conceived by other human beings from the only subjective perspective accessible to us: ‘What would it be like for a human to behave as x?’ In order to understand machinic art as such – and assuming that we stand a chance of even recognising it in the first place – we would need a way to conceive a first-person experience of what it is like to be that machine. That is something we cannot do even for beings that are much closer to us.

It might very well happen that we understand some actions or artifacts created by machines of their own volition as art, but in doing so we will inevitably anthropomorphise the machine’s intentions. Art made by a machine can be meaningfully interpreted in a way that is plausible only from the perspective of that machine, and any coherent anthropomorphised interpretation will be implausibly alien from the machine perspective. As such, it will be a misinterpretation of the artwork.

But what if we grant the machine privileged access to our ways of reasoning, to the peculiarities of our perception apparatus, to endless examples of human culture? Wouldn’t that enable the machine to make art that a human could understand? Our answer is yes, but this would also make the artworks human – not authentically machinic. All examples so far of ‘art made by machines’ are actually just straightforward examples of human art made with computers, with the artists being the computer programmers. It might seem like a strange claim: how can the programmers be the authors of the artwork if, most of the time, they can’t control – or even anticipate – the actual materialisations of the artwork? It turns out that this is a long-standing artistic practice.

Suppose that your local orchestra is playing Beethoven’s Symphony No 7 (1812). Even though Beethoven will not be directly responsible for any of the sounds produced there, you would still say that you are listening to Beethoven. Your experience might depend considerably on the interpretation of the performers, the acoustics of the room, the behaviour of fellow audience members or your state of mind. Those and other aspects are the result of choices made by specific individuals or of accidents happening to them.

Also read: Why Science Students Are Taking to Tribal Art Forms To Visualise Their Work

But the author of the music? Ludwig van Beethoven. Let’s say that, as a somewhat odd choice for the programme, John Cage’s Imaginary Landscape No 4 (March No 2) (1951) is also played, with 24 performers controlling 12 radios according to a musical score. In this case, the responsibility for the sounds being heard should be attributed to unsuspecting radio hosts, or even to electromagnetic fields. Yet, the shaping of sounds over time – the composition – should be credited to Cage. Each performance of this piece will vary immensely in its sonic materialisation, but it will always be a performance of Imaginary Landscape No 4.

Why should we change these principles when artists use computers if, in these respects at least, computer art does not bring anything new to the table? The (human) artists might not be in direct control of the final materialisations, or even be able to predict them but, despite that, they are the authors of the work. Various materialisations of the same idea – in this case formalised as an algorithm – are instantiations of the same work manifesting different contextual conditions.

In fact, a common use of computation in the arts is the production of variations of a process, and artists make extensive use of systems that are sensitive to initial conditions, external inputs or pseudo-randomness to deliberately avoid repetition of outputs. Having a computer executing a procedure to build an artwork, even if using pseudo-random processes or machine-learning algorithms, is no different than throwing dice to arrange a piece of music, or to pursuing innumerable variations of the same formula. After all, the idea of machines that make art has an artistic tradition that long predates the current trend of artworks made by artificial intelligence.

Machinic art is a term that we believe should be reserved for art made by an artificial mind’s own volition, not for that based on (or directed towards) an anthropocentric view of art. From a human point of view, machinic artworks will still be procedural, algorithmic and computational. They will be generative, because they will be autonomous from a human artist. And they might be interactive, with humans or other systems.

But they will not be the result of a human deferring decisions to a machine, because the first of those – the decision to make art – needs to be the result of a machine’s volition, intentions and decisions. Only then will we no longer have human art made with computers, but proper machinic art.

The problem is not whether machines will or will not develop a sense of self that leads to an eagerness to create art. The problem is that if – or when – they do, they will have such a different Umwelt that we will be completely unable to relate to it from our own subjective, embodied perspective. Machinic art will always lie beyond our ability to understand it because the boundaries of our comprehension – in art, as in life – are those of the human experience.Aeon counter – do not remove

This article was originally published at Aeon and has been republished under Creative Commons.
Aeon counter – do not remove

Infinite in All Directions: Sokal Squared, Sex and Gender, Life as a Jellyfish

It has escaped mainstream attention that the Sokal Squared hoax, perpetrated by three research researchers to expose what they claim is the absurdity of gender studies, chiefly targets research that explores race.

Go where science takes you, even if it is infinite in all directions. Subscribe to the newsletter format here and get this column in your inbox every Saturday morning.

Top news

You can’t have all news items on the homepage all the time even though they might each deserve that place, nor can a single publication cover all the notable news in the world on a given day. But if given the chance, these are the stories I would have liked to showcase on my hypothetical homepage November 2 morning:

* IISc forces senior professor Giridhar Madras into compulsory retirement

* Retraction Watch publishes giant database of retracted papers

* How the global fake medicine industry is putting your life at risk

* World’s oceans could be way warmer than we thought

* India tops global chart of pollution-related child deaths

§

Science and society

Science is frequently understood as an enterprise engaged in the unearthing of new facts, or verification of older, supposed facts, through methods that strive to eliminate biases. Missing in this picture is the key role of interpretation itself: science lies in its facts but also in how those facts are interpreted together in various contexts. In turn, this requires us to view science as a knowledge-building enterprise for all of society, beyond just for a group of specialists.

* Anyone notice the Sokal Squared hoax attempts to dismiss race studies? – “By now we have all read too many summaries, critiques and defenses of the so-called Sokal-squared hoax perpetrated by a trio of self-declared liberal humanists who published some faux journal papers to unmask what they see as the absurdity of ‘grievance studies’. …Concerns with race, however, pervade the trio’s work. In their mainstream reveal piece, they take aim at progressive stacking with puerile pranks, they jokingly rewrite Hitler’s Mein Kampf, they specify as among their targets ‘critical whiteness theory’ and they have specifically mentioned ‘white fragility’ – linking to Robin DiAngelo’s seminal article on the topic.”

(For more about the original Sokal hoax, try this backgrounder + analysis by Steven Weinberg, the celebrated physicist.)

* Report blames humanity for 60% of animal populations loss but forgets to name capitalism – “The latest Living Planet report from the WWF makes for grim reading: a 60% decline in wild animal populations since 1970, collapsing ecosystems, and a distinct possibility that the human species will not be far behind. …[I]n the 148-page report, the word ‘humanity’ appears 14 times, and ‘consumption’ an impressive 54 times. There is one word, however, that fails to make a single appearance: capitalism. … By obscuring capitalism with a term that is merely one of its symptoms – ‘consumption’ – [it risks shifting the] responsibility for species loss onto individual lifestyle choices, while larger and more powerful [institutions] that are compelling individuals to consume are, worryingly, let off the hook.”

* When do you act to fix something you know is broken: when there is sufficient reason or when there is all the reason?

 

§

Oped(s)

Piecing together stories published at disparate times and places but which have a theme or two in common.

Over 1,600 scientists have signed a letter of protest addressed to the White House against its proposed definition of ‘gender’ that purportedly disidentifies transgender and intersex people. According to a press statement issued alongside the letter,

The letter was a grassroots effort. Immediately following the publication of the New York Times article about the administration’s proposal, with its “grounded in science” claim, scientists began voicing their objections on social media. Twenty-two biologists and other scientists in related fields planned and wrote the letter collaboratively.

The letter asks for the administration to withdraw the draft policy and for the petitioners’ “elected representatives to oppose its implementation”. It has been signed by over 1,600 people working as “biologists, geneticists, psychologists, anthropologists, physicians, neuroscientists, social scientists, biochemists, mental health service providers,” and scientists in other fields.

However, subject expertise has little role to play in the context of the letter, and certainly shouldn’t let the Trump administration off the hook simply because it believes only ‘scientific things’ are entitled to legal protection.

If technical expertise were really necessary to disabuse the Trump administration of its misbelief that gender is a biological construct, the experts at the forefront should have included those qualified to comment meaningfully on how people build and negotiate gender. But even this wouldn’t save the letter from its principal problem: it seems to be almost exclusively offended by the Trump administration’s use of the phrase “grounded in science” over anything else, and devotes three paragraphs underlining the lack of empirical knowledge on this count. This is problematic.

In transgender individuals, the existence and validity of a distinct gender identity is supported by a number of neuroanatomical studies. Though scientists are just beginning to understand the biological basis of gender identity, it is clear that many factors, known and unknown, mediate the complex links between identity, genes, and anatomy.

In intersex people, their genitalia, as well as their various secondary sexual characteristics, can differ from what clinicians would predict from their sex chromosomes. In fact, some people will live their entire lives without ever knowing that they are intersex. The proposed policy will force many intersex people to be legally classified in ways that erase their intersex status and identity, as well as lead to more medically unnecessary and risky surgeries at birth. Such non-consensual gender assignment and surgeries result in increased health risks in adulthood and violate intersex people’s right to self-determination.

Millions of Americans identify as transgender or gender non-conforming, or have intersex bodies, and are at increased risk of physical and mental health disorders resulting from discrimination, fear for personal safety, and family and societal rejection. Multiple standards of health care for transgender and intersex people emphasise that recognising an individual’s self-identified gender, not their external genitalia or chromosomes, is the best practice for providing evidence-based, effective, and lifesaving care. Our best available evidence shows that affirmation of gender identity is paramount to the survival, health, and livelihood of transgender and intersex people.

A socio-cultural description of some of the ways in which Americans interpret gender, the challenges they may face and what they believe could be the appropriate way to address them are all conspicuous by absence. People are not rallying to this cause because science doesn’t yet know; that would be disingenuous. Instead, they are speaking up because the cultural experience of gender is missing from the White House’s articulation.

Finally, more than following Trump’s draft policy into its hole of cultural elision, the letter itself seems to fail to distinguish between sex and gender. It says:

The relationship between sex chromosomes, genitalia, and gender identity is complex, and not fully understood. There are no genetic tests that can unambiguously determine gender, or even sex.

The relationship between sex chromosomes and genitalia is much better understood than the relationship between the two and gender identity. Further, sex, being entirely biological, can indeed be determined to a large extent by genetic tests. It is gender that is harder to associate with one’s genes because it is a social/cultural/political construct and genes aren’t its sole determinants.

The following para also notes:

In transgender individuals, the existence and validity of a distinct gender identity is supported by a number of neuroanatomical studies.

It is doubtful if these studies demonstrate causation together with correlation.

Notwithstanding the legal protections afforded to people of non-binary gender and the terms of their provision, the letter would have benefited by calling the policy out for framing it as an insular problem of science, not putting up an equally insular counter-argument and by being more wary of the language it employs to defend its stance. Specialised expertise is important but it is not everything.

§

Unusual stuff

Sciencey things people are trying to do that are out of the ordinary in some way.

* How the immortal jellyfish helps me rewrite my queer childhood – “The immortal jellyfish has no brain or heart. Everything it consumes and extrudes passes through a single orifice (an anus, to be precise.) It does not long for anything. It never feels out of place. It does not understand that it could live forever, just as it has no conception that one day it will probably die. In my eyes, it makes total sense that immortality could only ever exist within the bounds of a brainless life. If anything intelligent was ever given a second chance at adolescence, it may never choose to grow up.”

* Psychology review tries to explain why schadenfreude exists – The “common, yet poorly understood, emotion” of schadenfreude “may provide a valuable window into the darker side of humanity,” according to a review of studies by psychologists at Emory University that draws “upon evidence from three decades of social, developmental, personality and clinical research … The authors propose that schadenfreude comprises three separable but interrelated subforms – aggression, rivalry and justice – which have distinct developmental origins and personality correlates. They also singled out a commonality underlying these subforms”: dehumanisation.

§

See also

Some interesting articles from around the web.

* Scientists point out ‘blatant’ anti-nuclear bias in latest IPCC report – “Some of the scientists most often cited by the Intergovernmental Panel on Climate Change (IPCC) have taken the unusual step of warning leaders of G-20 nations that a recent IPCC report uses a double standard when it comes to its treatment of nuclear as compared to renewables. … The letter signers include leading radiation experts who expressed outrage that the IPCC had claimed a link between nuclear power stations and leukaemia when in reality ‘there is no valid evidentiary support for it and the supposed connection has been thoroughly dismissed in the literature.'”

* Algorithm identifies best phosphor for use in LEDs from giant list in a minute – “A machine learning algorithm that can swiftly identify the most desirable LED phosphor host compound out of a database of almost 120,000 materials has been developed by researchers in the US. Efficient enough to run on a PC, the program was created by Jakoah Brgoch and colleagues at the University of Houston and has predicted the relevant properties of a highly efficient, thermally stable compound in well under a minute. The team’s algorithm could soon be used to speed up the discovery of new materials for use in commercially competitive LEDs.” Paper describing algorithm here.