recentpopularlog in

mcherm : ai   120

« earlier  
A Breakthrough for A.I. Technology: Passing an 8th-Grade Science Test - The New York Times
AI can now pass an 8th grade science test (if you take out all the questions that have pictures or diagrams).
ai  nytimes  via:reddit 
9 days ago by mcherm
Reframing Superintelligence | Slate Star Codex
AI may not be as dangerous as many have feared because it might be just a set of special purpose tools.
ScottAlexander  SlateStarCodex  via:SlateStarCodex  ai 
12 days ago by mcherm
Patent Application Names AI as Inventor
They decided to push the legal system. For obvious reasons, I'm sure that they will get a precedent that an AI cannot be an inventor. Also they'll get publicity.
law  ip-law  patent  ai  via:reddit 
6 weeks ago by mcherm
King -Man +Woman = King ? - Netherlands eScience Center
The examples they give for how awesome word2vec is are mostly lies.
NaturalLanguageProcessing  ai  via:HackerNews 
9 weeks ago by mcherm
Changing my Mind about AI, Universal Basic Income, and the Value of Data – The Art of Research
"The cult thinking is that datasets such as ImageNet, a dataset of over a million labeled images frequently used for AI training, can train AI that will replace more work than the dataset took to make. This seems unlikely for ImageNet, which was labeled by 49,000 workers over 2.5 years"
ai  ViHart  philosophy  basicincome 
june 2019 by mcherm
When algorithms mess up, the nearest human gets the blame - MIT Technology Review
"Moral crumple zone" -- some nearby human we can blame so the automated system doesn't shoulder blame or culpability.
ethics  ai  law  via:reddit 
may 2019 by mcherm
Artificial intelligence, algorithmic pricing, and collusion | VOX, CEPR Policy Portal
Ai pricing models can "collude" on prices without communicating. This doesn't technically violate anti-trust laws.
pricing  ai  ethics  via:reddit 
may 2019 by mcherm
Photos from Crude Sketches: NVIDIA's GauGAN Explained Visually -
The full implementation details on how photo realistic pictures can be built from sketches. Notice how layers of neural nets with particular architectures are used.
machinelearning  ai  programming 
april 2019 by mcherm
GauGAN Turns Doodles into Stunning, Realistic Landscapes | NVIDIA Blog
A use of machine learning that really impressed me. Consider how this might be generalized to have systems making stuff that humans like and expect.
ai  machinelearning  via:HackerNews 
april 2019 by mcherm
Humans Who Are Not Concentrating Are Not General Intelligences | Otium
"I note," the author says, "that when I don't pay attention I can't tell that excellent AI generated essays are drivel."
ai  via:SlateStarCodex  brain 
february 2019 by mcherm
The Rise of the Robot Reporter - The New York Times
The headline is exciting (and 1/3 of Bloomberg News articles are assisted). But the truth behind the article is that the AI can collect data (and shines in areas where the hard part is collecting the data). It doesn't excel at writing.
journalism  ai  nytimes  via:HackerNews 
february 2019 by mcherm
A.I. May Have Written This Article. But Is That Such a Bad Thing?
An air has written some 800 articles in the Washington Post.
ai  journalism 
february 2019 by mcherm
Remove Background from Image –
A tool that processes pictures to automatically recognize the person in the foreground and delete the background behind them.
imageprocessing  pictures  ai  useful  via:reddit 
december 2018 by mcherm
Comparison of the best NSFW Image Moderation APIs 2018
The author tested several commercially available models for detecting porn. BUT, tiny sample size.
machinelearning  ai  via:reddit 
december 2018 by mcherm
20 top lawyers were beaten by legal AI. Here are their surprising responses
An AI beat lawyers at identifying possible issues in a formulaic contract type (an NDA). The lawyers weren't scared of being replaced, they said it's about time we automated this stupid stuff and combined lawyer and machine. Of course, it's all an advertisement.
law  ai  via:reddit 
december 2018 by mcherm
Algorithmic merchandising leaves a bad taste in my mouth. Slowly but surely, it will erode trust for all the tech giants.
Amazon plays around with how they display products to encourage the products that give them the best margin. This annoys the author (and many others, I am sure).
amazon  ai  marketing  via:HackerNews 
october 2018 by mcherm
Amazon scraps secret AI recruiting tool that showed bias against women | Hacker News
I comment on how bias in AI may not be so bad... human processes are biased also, and at least the AI process can (if we build it this way) be more transparent.
mypostings  ai  discrimination 
october 2018 by mcherm
Amazon scraps secret AI recruiting tool that showed bias against women | Reuters
Amazon trained an AI model to classify resumes based on their previous hiring. Then later they discovered that the model had negative weights on terms like "women" (eg: "women's chess club"). They finally scrapped the whole model.
ai  bias  amazon  hiring  womenintech  via:HackerNews 
october 2018 by mcherm
OpenAI’s Dota 2 defeat is still a win for artificial intelligence  - The Verge
The humans won this time (at a complex video game), but the AI is coming along rapidly.
via:HackerNews  ai 
august 2018 by mcherm
Facebook’s program thinks Declaration of Independence is hate speech - News - Liberty Vindicator
Part of the Declaration of Independence (perhaps the reference to "Indian Savages"?) is deemed "hate speech" by Facebook's algorithms and blocked. Fair cop, actually, but is this the behavior we want? There is no process for appeal.
censorship  zerotolerance  ai 
july 2018 by mcherm
More States Opting To 'Robo-Grade' Student Essays By Computer : NPR
Essays on standardized tests are now being graded by computer algorithms. They are fairly easy to game.
ai  teaching  via:HackerNews 
july 2018 by mcherm
ICE Rigged Its Vetting Tool To Make Sure It Can Always Keep Immigrants Locked Up | Techdirt
The software that recommends "detain" or "release" was re-written so it ONLY says "detain".
law  ai  immigration  DonaldTrump  techdirt  via:Techdirt 
june 2018 by mcherm
A Deep Dive into Monte Carlo Tree Search
A description of the algorithm that AlphaGo/AlphaZero uses, complete with sample code.
ai  programming  go  algorithms  via:HackerNews 
may 2018 by mcherm
Capitalist International — When algorithms surprise us
When genetic algorithms are used to discover a solution, it might be one that breaks the rules. Here are lots of examples, including the "sort algorithm" that deleted all the items (now there's nothing that isn't sorted!).
bugs  ai  via:reddit 
april 2018 by mcherm
The Machine Learning Reproducibility Crisis « Pete Warden's blog
This writer says machine learning folks don't use version control and their experiments aren't reproducible.
versioncontrol  ai  machinelearning  via:boingboing 
march 2018 by mcherm
Stanford, USC, Duke study shows AI reads contracts better than lawyers
AI trounces human lawyers at interpretation of complex contracts.
law  ai  via:reddit 
march 2018 by mcherm
We Are Truly Fucked: Everyone Is Making AI-Generated Fake Porn Now - Motherboard
The technology to stick other people's faces into videos has now arrived and humanity is using it to create fake porn of celebrities.
pornography  ai  video  via:boingboing 
january 2018 by mcherm
Software used in judicial decisions meets its equal in random amateurs | Ars Technica
The software that predicts who will re-offend after prison is wrong 1/3 of the time. Using mechanical turk to ask random people to make predictions was just as accurate. So was just using JUST count of convictions and age.
ai  prison  algorithms  ArsTechnica  via:ArsTechnica  bigdata 
january 2018 by mcherm
A.I. ‘Bias’ Doesn’t Mean What Journalists Say it Means – Jacobite
Many journalists write scare stories about biased AI algorithms when actually the algorithm is race-blind but actual outcomes are not. This may be true, but it also misses some important subtleties.
ai  racism  statistics  via:SlateStarCodex  journalism 
january 2018 by mcherm
AI-Assisted Fake Porn Is Here and We’re All Fucked - Motherboard
An AI algorithm usable by any random person on commodity Hardware is now capable of plausibly pasting someone's face onto another person's body in a porn video.
ai  pornography  video 
december 2017 by mcherm
A Strategist’s Guide to Artificial Intelligence
Part of an assignment at the Darden business center.
ai  via:darden 
october 2017 by mcherm
Using chatbots against voicespam: analyzing Lenny’s effectiveness | the morning paper
Lenny, the chatbot deployed to waste the time of spam callers, has no actual voice recognition. It keeps people engaged (very effectively) because it exploits underlying "turn-taking" mechanisms in human speech.
ai  spam  via:HackerNews 
september 2017 by mcherm
Big Data Is Coming to Take Your Health Insurance - Bloomberg
Better data + AI means insurance companies will be able to better predict health care costs. This will destroy "pooling" (with or without Obamacare's rule on preexisting conditions).
insurance  bigdata  ai  politics  via:boingboing 
august 2017 by mcherm
lcamtuf's blog: Pulling JPEGs out of thin air
They have a fuzzer that monitors the branches taken by the program it is fuzzing input for. Starting from nothing and running it on an image processing program they managed to evolve valid JPEGs. Maybe fuzzers are better than I think!
testing  programming  images  via:HackerNews  ai 
may 2017 by mcherm
Google reveals secret test of AI bot to beat top Go players | Hacker News
Google's Go bot played online under an alias and beat a bunch of professional level humans. People on Hacker News speculate on what this means for AI in general.
ai  go  google  via:HackerNews 
january 2017 by mcherm
Superintelligence: The Idea That Eats Smart People
Maciej (operator of Pinboard) writes on why really smart people get taken in by worries over AI superintelligence, but the concerns are overblown.
via:MaciejCegłowski  philosophy  singularity  ai 
december 2016 by mcherm
Facebook fires trending team, and algorithm without humans goes crazy | Technology | The Guardian
Facebook fires news selection team with no notice; AI replacement messes up some of its choices.
ai  facebook  via:reddit 
august 2016 by mcherm
Artificial Intelligence’s White Guy Problem - The New York Times
Lots of existing algorithms seem to be racially biased -- "we don't have to wait for the evil AI takeover, it's already here for those who aren't rich white guys".
ethics  ai  nytimes  via:boingboing 
august 2016 by mcherm
Now this incredible free robot lawyer is fighting homelessness
An AI lawyer which is actually being somewhat effective.
law  ai  via:HackerNews 
august 2016 by mcherm
Google Cuts Its Giant Electricity Bill With DeepMind-Powered AI - Bloomberg
This is a good example of how AI can (already!) optimize companies in ways that mere humans aren't capable of.
google  ai  via:HackerNews 
july 2016 by mcherm
Professor reveals to students that his assistant was an AI all along
This is the first SERIOUS use of AI that I've read that really, truly impressed me.
ai  singularity  via:slashdot 
may 2016 by mcherm
I was an undercover-bot for 2 months. Here is what I learned. — Chatbots Magazine — Medium
He played human stand-in for a website help chatbot. He learned that the way people communicate is vague and incomprehensible, and not even CLOSE to being understood by AI.
ai  customerservice  via:HackerNews 
april 2016 by mcherm
How to Make a Bot That Isn't Racist | Motherboard
Shockingly primitive techniques used to make sure that bots active on Twitter don't say racist things.
ai  racism  censorship  via:boingboing 
april 2016 by mcherm
Did a Computer Bug Help Deep Blue Beat Kasparov? | WIRED
Apparently the move that surprised Kasparov during the match was randomly selected due to a bug.
chess  ai  history  via:CodingHorror 
april 2016 by mcherm
Google Now just blew my mind (again) : Android
Google Now plucks words of reassurance from an old email to provide emotional support.
ai  future  via:reddit 
april 2016 by mcherm
Japanese AI Writes a Novel, Nearly Wins Literary Award | Big Think
I have some doubts but the article claims that any I wrote a book which was very well received by people.
ai  via:reddit  writing 
march 2016 by mcherm
Autocomplete from Stack Overflow
Hmm... with this, can we finally dispense with human developers?
funny  ai  programming  via:reddit 
march 2016 by mcherm
Will Your Job Be Done By A Machine? : Planet Money : NPR
NPR's estimate of the probability of various jobs being eliminated by computers or robots.
ai  future  robots 
march 2016 by mcherm
The NSA’s SKYNET program may be killing thousands of innocent people | Ars Technica UK
The use of AI to decide who to kill is one thing -- maybe a bad thing. The use of BADLY DESIGNED AI that is known to be unreliable because of how it is built is much scarier, and that's what we're doing today.
terrorism  nsa  law  ai  ArsTechnica  via:ArsTechnica  ethics 
february 2016 by mcherm
A Look into Machine Learning's First Cheating Scandal – Subjective functions
A team of Machine Learning scientists "cheated" by submitting their algorithms for testing against the scoring subset too many times.
ai  ScientificMethod  via:HackerNews 
december 2015 by mcherm
Facebook M — The Anti-Turing Test
It claims to be an AR, but is it really human? This author conducts an experiment to find out.
ai  facebook  via:HackerNews 
november 2015 by mcherm
Thinking with the machine | David R. MacIver
Human intelligence is augmented by computers. It works two ways: the human requesting but the computer doing the work (eg: a search) OR the human thinking, but the computer aiding (eg: an outlining tool or a calculator).
ai  intelligence  brain  DavidMacIver  via:reddit 
october 2015 by mcherm
Study shows computers can detect when a Diplomacy player is about to betray you
Their computer program can tell from your choice of words when you are about to betray them in the game "Diplomacy".
games  via:reddit  ai 
september 2015 by mcherm
The Unreasonable Effectiveness of Recurrent Neural Networks
An essay describing the state of the art on a certain kind of neural network that processes bodies of texts and can actually learn grammatical structures such as the rules of writing English sentences or the structure of a markup language.
ai  neuralnet  via:reddit  machinelearning 
may 2015 by mcherm
What you wanted to know about AI - FastML
Some examples of the state of AI today. Not as good as the hype would lead you to believe.
ai  via:HackerNews 
march 2015 by mcherm
Robots are starting to break the law and nobody knows what to do about it -- Fusion
The program made random purchases off a "darknet" site until it purchased some stuff that was illegal.
personal_net  via:LoweringTheBar  ai  law 
january 2015 by mcherm
Mind of a Worm Uploaded to a LEGO Robot to Make the Weirdest Cyborg Ever | eTeknix
They build a software simulator for the neural net of an actual worm and put it in a robot -- and succeeded in getting a few simple behaviors out of it (but not more complex behaviors yet).
ai  robots  via:reddit  science  singularity 
january 2015 by mcherm
extrapolated art - Cambridge Machine Learning Group | Yarin Gal
An AI that looks at a painting and "extends" it in the same style. Works surprisingly well in a few cases, but it's no great artist.
ai  art  via:boingboing 
december 2014 by mcherm
Can an Artificial Intelligence win 2048? - Ranjay Krishna
An AI that plays 2048. May actually do better than my algorithm.
onlinegames  ai  2048  via:HackerNews 
december 2014 by mcherm
Deep learning for… chess | Erik Bernhardsson
A simple chess ai that was just trained by watching a corpus of games manages to do decently.
ai  chess  via:HackerNews 
december 2014 by mcherm
MIT computer scientists can predict the price of Bitcoin | MIT News
They claim that they can predict short-term movement of bitcoin prices, and they claim to have proved it by doubling their money in 2 months. But they didn't ACTUALLY trade, so I don't think it counts.
investing  bitcoin  ai 
october 2014 by mcherm
Google self-driving car: It may never actually happen.
The google self-driving car isn't as good as I have been led to believe. In particular, it needs an accurate map of the area (detailed enough to have each traffic light and stationary object near the road) before driving it. Today it can only drive over a few thousand miles of road that have been meticulously mapped.
slate  google  cars  ai  via:boingboing 
october 2014 by mcherm
It’s Official: AIs are now re-writing history | Robert Elliott Smith
I don't even know what to make of this. Algorithms automatically altering photographs without the user even realizing.
photography  ai  socialcomputing  via:HackerNews 
october 2014 by mcherm
"Humans Need Not Apply" on YouTube
15 min video that does a good job of showing how robots and AI will transform the economy.
basicincome  video  ai  robots  via:reddit 
august 2014 by mcherm
Exercise to detect Algorithmically Generated Domain Names.
An interesting exercise where they slap together an AI filter to distinguish human-made domain names from algorithm-made ones. We get to see the data analysis, the use of Python for plotting, and the creation of the AI filter.
python  mathematica  via:HackerNews  ai 
july 2014 by mcherm
Marion Tinsley: Human Perfection at Checkers?
A human checkers champion who lost only 7 games in 45 years(!!!) as undisputed world champion. He went up against a computer and won, repeatedly.
games  ai  via:Wikipedia 
july 2014 by mcherm
The Singularity Is Further Than It Appears - Charlie's Diary
The first smarter-than-human AI won't immediately create an even-smarter AI leading to an exponential increase in speed because the problem of getting smarter is probably a non-linear problem.
singularity  via:CharlesStross  RamezNaam  ai 
april 2014 by mcherm
NSA robots are 'collecting' your data, too, and they're getting away with it | Comment is free |
Re-defining "collect data" to mean "only if a human sees it" and "computers looking doesn't count" is hogwash. Computers looking DOES matter, because they run algorithms and then DO something with it (like show it to humans, but other pernicious things are also possible).
nsa  privacy  surveillance  BruceSchneier  via:reddit  TheGuardian  ai 
february 2014 by mcherm
Massive Open Lab gets volunteers to beat computers in designing RNA | Ars Technica
They turned protein folding into a game, and the players quickly became better than the best computer programs. They even came up with heuristics that no one hand known previously and which significantly improved the performance of the algorithms.
gamedesign  intelligence  ai  science  via:ArsTechnica  ArsTechnica 
january 2014 by mcherm
Magnus Carlsen's Win in Chess Championship Shows Powerful Role of Computers -
When computers beat humans at chess, it didn't cause everyone to quit playing chess, instead they got MORE interested and started getting better by learning from the computers.
chess  ai  via:DavidMacIver 
november 2013 by mcherm
« earlier      
per page:    204080120160

Copy this bookmark:

to read