We need a new science of measurement to figure out when AI is working and why it fails

Writing on Medium, AI researcher Kate Crawford (previously) and Simply Secure (previously) co-founder Meredith Whittaker make the case for a new scholarly discipline that "measures and assesses the social and economic effects of current AI systems." Read the rest

Blackballed by machine learning: how algorithms can destroy your chances of getting a job

The Guardian's published a long excerpt from Cathy O'Neil's essential new book, Weapons of Math Destruction, in which O'Neil describes the way that shoddy machine-learning companies have come to dominate waged employment hiring, selling their dubious products to giant companies that use them to decide who can and can't work. Read the rest

Sampling bias: how a machine-learning beauty contest awarded nearly all prizes to whites

If you've read Cathy O'Neil's Weapons of Math Destruction (you should, right NOW), then you know that machine learning can be a way to apply a deadly, nearly irrefutable veneer of objectivity to our worst, most biased practices. Read the rest

Weapons of Math Destruction: invisible, ubiquitous algorithms are ruining millions of lives

I've been writing about the work of Cathy "Mathbabe" O'Neil for years: she's a radical data-scientist with a Harvard PhD in mathematics, who coined the term "Weapons of Math Destruction" to describe the ways that sloppy statistical modeling is punishing millions of people every day, and in more and more cases, destroying lives. Today, O'Neil brings her argument to print, with a fantastic, plainspoken, call to arms called (what else?) Weapons of Math Destruction.

The Equation Group's sourcecode is totally fugly

With the leak of exploits developed by The Equation Group, the long-secret, NSA-adjacent super-elite hacking squad -- published by The Shadow Brokers, who have some extremely heterodox theories about auction design -- it's now possible to audit the source code of some of the NSA's crown-jewel cyberweapons. Read the rest

UK/EU security researchers: tax-free stipend to study privacy and authentication

UC London's offering a tax-free stipend for UK/EU students to work on designing and evaluating new approaches for continuous authentication, based on a solid theoretical underpinning so as to give a high degree of confidence that the resulting decisions match expectations and requirements" as well as "ways to preserve user privacy by processing behavioural measurements on the user’s computer such that sensitive information is not sent to the online service." (Image: LordHarris, CC-BY-SA) (Thanks, William!) Read the rest

If the 2016 election is hacked, it's because no one listened to these people

Ever since the Supreme Court ordered the nation's voting authorities to get their act together in 2002 in the wake of Bush v Gore, tech companies have been flogging touchscreen voting machines to willing buyers across the country, while a cadre computer scientists trained in Ed Felten's labs at Princeton have shown again and again and again and again that these machines are absolutely unfit for purpose, are trivial to hack, and endanger the US election system. Read the rest

Forget Skynet: AI is already making things terrible for people who aren't rich white dudes

Kate Crawford (previously) takes to the New York Times's editorial page to ask why rich white guys act like the big risk of machine-learning systems is that they'll evolve into Skynet-like apex-predators that subjugate the human race, when there are already rampant problems with machine learning: algorithmic racist sentencing, algorithmic, racist and sexist discrimination, algorithmic harassment, algorithmic hiring bias, algorithmic terrorist watchlisting, algorithmic racist policing, and a host of other algorithmic cruelties and nonsense, each one imbued with unassailable objectivity thanks to its mathematical underpinnings. Read the rest

100 million VWs can be unlocked with a $40 cracker (and other cars aren't much better)

In Lock It and Still Lose It—On the (In)Security of Automotive Remote Keyless Entry Systems, a paper given at the current Usenix Security conference in Austin, researchers with a proven track record of uncovering serious defects in automotive keyless entry and ignition systems revealed a technique for unlocking over 100,000 million Volkswagen cars, using $40 worth of hardware; they also revealed a technique for hijacking the locking systems of millions of other vehicles from other manufacturers. Read the rest

Trump is an object lesson in the problems of machine learning

Trump's algorithm is to say semi-random things until his crowd roars its approval, then he iteratively modifies those statements, seeking more and more approval, until he maxes out and tries a new tack. Read the rest

How facial recognition works (and how to hack your own in Python)

You could not ask for a clearer, easier-to-read, more informative guide to facial recognition and machine learning thank Adam Geitgey's article, which is the latest in a series of equally clear explainers on machine learning, aimed at non-technical people -- and if you are a programmer, he's got links to Python sample source and projects you can use to develop your own versions. Read the rest

Man builds giant, discrete-component-based computer that can play Tetris

James Newman's "Megaprocessor" is a giant "microprocessor" built on transistors and other discrete components that he soldered onto boards and wired together in frames that stand 2m high and run 10m long. Read the rest

White House contends with AI's big social challenges, July 7/NYC

Meredith from Simply Secure writes, "Artificial Intelligence is already with us, and the White House and New York University’s Information Law Institute are hosting a major public symposium to face what the social and economic impacts might be. AI Now, happening July 7th in New York City, will address the real world impacts of AI systems in the next next 5-10 years." Read the rest

Analyzing all known Metal lyrics with natural language processing

Iain ("an ex-physicist currently working as a data scientist") scraped Dark Lyrics and built a dataset of lyrics to 222,623 songs by 7,364 metal bands, then used traditional natural language processing techniques to analyze them. Read the rest

Even if Moore's Law is "running out," there's still plenty of room at the bottom

A very good piece by Tom Simonite in the MIT Technology Review looks at the implications of Intel's announcement that it will slow the rate at which it increases the density of transistors in microprocessors. Read the rest

Google is restructuring to put machine learning at the core of all it does

Steven Levy is in characteristic excellent form in a long piece on Medium about the internal vogue for machine learning at Google; drawing on the contacts he made with In the Plex, his must-read 2012 biography of the company, Levy paints a picture of a company that's being utterly remade around newly ascendant machine learning techniques. Read the rest

Cataloging the problems facing AI researchers is a cross between a parenting manual and a management book

Concrete Problems in AI Safety, an excellent, eminently readable paper from a group of Google AI researchers and some colleagues, sets out five hard problems facing the field: robots might damage their environments to attain their goals; robots might figure out how to cheat to attain their goals; supervising robots all the time is inefficient; robots that are allowed to try novel strategies might cause disasters; and robots that are good at one task might inappropriately try to apply that expertise to another unrelated task. Read the rest

More posts