Category Archives: Ai

Making AI models at the University of Southern California (USC) Center for AI in Society does not involve a clean, sorted dataset. Sometimes it means interviewing homeless youth in Los Angeles to map human social networks. Sometimes it involves going to Uganda for better conservation of endangered species.

With AI, we are able to reach 70 percent of the youth population in the pilot, compared to about 25 percent in the standard techniques. So AI algorithms are able to reach far more youth in terms of spreading HIV information compared to traditional methods, saidMilind Tambe, a professor at the USC Viterbi School of Engineering and cofounder of the Center for AI in Society. If I were doing AI normally I might get data from the outside and I would analyze the data, produce algorithms, and so forth, but I wouldnt go to a homeless shelter.

The pilot project will next be expanding to serve 1,000 youth. Other projects currently being taken on by the Center for AI in Society include gang prevention, wildlife conservation with computer vision, and predictive models to improve cybersecurity, prevent suicide, and help homeless youth find housing.

The center has also developed and deployed algorithms for federal agencies such as the U.S. Coast Guard, Air Marshals Service, and Transportation and Security Administration (TSA).

Tambe was one a handful ofauthors of a forward-looking report that examines how AI will evolve and affect business, government, and society between the present and 2030. Commissioned by Stanford University as part of The AI 100 Project, the study found that AI aimed at solving social problems has traditionally lacked investment because it produces no profitable commercial applications. The report prescribes making AI for low resource projects a higher priority and offering AI researchers incentives, but Tambe also believes an entirely new discipline may need to be developed.

[These projects] bring up completely new kinds of AI problems because working with low resource communities, data is sparse, as opposed to being plentiful. When you talk about big data, thats not what were doing here. Whether its wildlife conservation or working with homeless youth, were talking incomplete data and theres no capacity to actually produce that massive clean big data that you can do deep learning on, he said.

Were trying to develop novel AI science as well as novel social science, co-director Eric Rice told VentureBeat in a phone interview. Were not just trying to be data scientists who take advantage of publicly available datasets or social scientists that take advantage of out-of-the-box machine learning tools that are pretty much readily available through canned software packages. What were really trying to build is new science on both sides.

The USC Center for AI in Society is a collaboration between computer science and social science schools at USC, an ambitious initiative created to cross-pollinate ideas between the two disciplines in order to solve some of the worlds biggest problems.

Created in 2013, the program focuses on problems found in the 12 Grand Challenges of social work and the United Nations Sustainable Development Goals.

The 12 Grand Challenges of Social Work was created last year by social workers and espouses goals like ensuring healthy development for all youth, eradicating social isolation, stopping family violence, and ending homelessness.

The Sustainable Development Goals were adopted by U.N. member nations in 2015 and focus on implementing measures to address priorities like access to quality education, gender equity, and the end of poverty and hunger by 2030.

This is the first collaboration as far as we are aware between AI and social work in a center. So were really collaborating across schools in terms of engineering and AI and social work, and its bringing up completely new sets of challenges to the core in terms of problems that the AI community has tackled, Tambe told VentureBeat in a phone interview. Spreading HIV information amongst homeless youth or trying to reduce substance abuse or matching homeless youth to homes, these are challenges that generally have not been tackled within the AI community.

The two schools work together because sometimes an AI data scientist may not understand a social issue if they dont see it emerge in a dataset, and social workers may sometimes fail to understand that an algorithm could significantly impact a social issue.

While there was some initial difficulty in understanding the different vocabularies social scientists and data scientists use, the collaboration leads to completely new kinds of discovery that wouldnt have been possible if either of us were working alone, Tambe said.

Social work tends to be less precise and engineering is very focused, so theres this dance were in, Rice said. Were adding more muddiness to the model and theyre insisting that we are more crisp in our argument, so theres a nice generative aspect to that kind of back and forth.

Read the rest here:

AI is targeting some of the world's biggest problems: homelessness, terrorism, and extinction - VentureBeat

Illustration: iStockphoto Computer algorithms can help magicians create magic tricks that exploit human psychology

You are presented with two decks, one with images and the other with words. The magician shuffles and distributesthe decks into piles of four cards. You get to choose twopiles, one from the word deck and one from the image deck, to make a hand of eight cards. Then youre invited to picka word card and and an imagecardfrom yourhand.Once youve selected a pair, youwatch the magician reveal a previously written prediction about the cards youve chosen. The prediction is correct!

That kind of mind-reading magic trick could benefitfrom new AI computer algorithms. These algorithms are designed to exploithuman psychology andhelpmagicians choosethe best card combinations.

Thisassociation magic trickrelies upon making a spectator believe that the magician hasmanaged to predict his or herfree choice from a random combination ofshuffled cards. In reality, the magician has preselected two decks of cards that together containa category of card pairs that triggera particularly powerful mental association for most people. To help pull off this mind-reading illusion, computer scientists created a computer algorithm that can automatically help find compellingword and image combinations.

First and foremost its an entertaining magic trick we have built, but it does potentially allow insight into the processes that humans use to decide associations, saysPeter McOwan, a professor of computer science at Queen Mary University London in the UK.There are a range of mentalism tricks that use associations to accomplish their effects and similar computational frameworks could be applied across that range, he said.

McOwan began practicing magic as a hobby in his teens. He has since used magic tricks to teach computer algorithms and haswrittenfree e-books on the intersection between the two subjects. In recent years, McOwan has teamed up withHoward Williams, another computer scientistat Queen Mary University London, to develop computer algorithms that can help create new magic tricks. Their latest study on the association magic trick was published in the 9 Aug 2017 issue of the journal PLOS One.

The association magic trick takes advantage of how the human subconscious tends to formstrong mental associations between certain concepts. For example, people may quickly make food associations between images of burgers or fruitand related words such asbites,treats,snack andfeast. The human subconscious can quickly recognize and process such associations in a way that appears almost automatic to the conscious mind.

Another key part of the trick involves an appreciation of two psychological systems that underlay our decision making, as described byDaniel Kahneman, a psychologist and Nobel Prize-winner. System 1 covers the swift and seemingly automatic mental processing. System 2 refers to the more active, conscious thinking involved in planning, puzzle solvingor calculations.

The magician wants the spectator participating in the magic show to use the first system and make the automatic association because it makes his or her choice predictableespecially when the decks of cards are organized and shuffled in a way that ensures a matched pair of cards that belongto a certain category will always be among the choices. So the magician adds time pressure by asking the spectator to make a quick decision. That pressure typically ensures the spectator makes the predictable choice rather than making a more idiosyncratic pairingbased on the more conscious thought processes of the second system.

To collect relevant data in making the magic trick, the Queen Mary University London researchersperformed an online psychology experiment by showing human participants various selections of 10 trademarks from a pool of 100 of the most famous trademarks. The researchers then askedparticipantsto write down any words about how the trademarks made them feel, along with any otherassociations they had with each mark.

But theresearchers alsodeveloped an AI to help themfindstrong associations for the magic trick. First, their computer algorithm ran Internet searches on popular trademarks and plucked words from the webpages linked by the top ten search results for each trademark. Second, itused a previously developed search algorithm, called BM25, to organizeand rank the collecteddata according to certain association categories (such as food-related words). Additional AI techniques called word2vec and Wordnet helpedby providingsimilarity scores for certain word pairings.

The AI by itself was not necessarily able to find the strongest or most useful associations for the magic trick without human help. But suchautomated data gathering and organization could prove a handy time-saving tool for complementing data collected from the more time-consuming experimental surveys, according to Williams at Queen Mary University London. He described the tradeoff as follows:

Automated data gathering is useful as it is quick and can gather large sets of data. Experiments take longer to organize, perform, process data, etc., but provide more specific and targeted data. [Its] essentially a tradeoff between quality and quantity. Though quantity provides broadness, and is useful in its own right.

That process led Williams and McOwan to create image and word card decks that contained the food category as the likeliest choice. Theytested out their association magic trick on 143individuals during theBig Bang 2013 science fair in Birmingham, UK, where it succeeded in all but 15 cases. Those more unusual word and image pairings chosen in the unsuccessful cases could potentially be excluded by the computer algorithm or by hand in the future.

Even though there is a fairly clear pathway we have created in the trick for them to follow in the performances, some people just had left field associationsprobably influenced by their life experiences, McOwan says.Its an area worth looking at more.

Magicians could eventually makeuseof popular AI techniques such as machine learning and deep learning that can automatically find and learn from patterns in data. McOwan speculated that such techniques could prove useful in cold reading, which is when a magician uses psychological tricks and a data-driven understanding of population trends to pretend to divine personal details about a stranger.

The researchers have already commercialized magic tricks that were created with the help of computer algorithms. In 2014, they used a computer algorithm to help create a magic jigsaw puzzle that makes certain shapes seem to disappear upon reassembly based on certain geometric principles. That jigsaw puzzlesold out two production runs in a well known London magic shop, McOwan says.

The idea of computer algorithms helping create magic tricks may lack the emotional drama ofChristopher Nolans film The Prestige,where rival magicians vie to perfecttheir magic illusions. But even some of thefictional wizards in the magical world of Harry Potter might appreciate muggle AI technology that can help magicians seem toperform mind reading without wands and spells.

Of course a trick is only as good as the performer and our work is simply giving new tools to create new methods to perform with, McOwan says.The real magic still lies with the magician.

IEEE Spectrums general technology blog, featuring news, analysis, and opinions about engineering, consumer electronics, and technology and society, from the editorial staff and freelance contributors.

Sign up for the Tech Alert newsletter and receive ground-breaking technology and science news from IEEE Spectrum every Thursday.

A laptop-sized system could make it easier to diagnose and study sleep disorders 10Aug

Minor changes to street sign graphics can fool machine learning algorithms into thinking the signs say something completely different 4Aug

Engineers at Knowles bring the hearing aid industry together to fight feedback with simulation.

The DNA-as-malware hackthough difficultpoints to weaknesses in bioinformatics software 10Aug

From consumer audio to ultrasound imaging, the impact of new metamaterial structures for acoustic cloaking is far-reaching.

Deep learning AI can identify individual anthrax spores in seconds within special microscope images 4Aug

An audio technology startup delivers manufacturable transducers for high-end headphones.

In designing acoustical systems, engineers must account for multiple physics and their interactions at multiple scales and frequencies.

Glassdoor's latest research shows software jobs are diffusing beyond traditional geographies and industries 31Jul

A race is on to discover Planet Nine using classical astronomy and new computational techniques 31Jul

Blind opens its tech gossip app to anyone who works in tech, but only some get into closed company rooms 27Jul

Software engineering has highest share of foreign-born workers 25Jul

This body-tracking software could help robots read your emotions 22Jul

Find the programming languages that are most important to you 18Jul

Python jumps to No.1, and Swift enters the Top Ten 18Jul

We analyze the languages that are indemandbyemployers 18Jul

Blind quantum computing in the cloud could keep computation results secret even for remote classical-computer users 14Jul

Personalized learning has to get social. Students learn better through conversation 10Jul

Social and computer scientists parse online bot discourse 6Jul

Hedge funds are testing new quantitative strategies that could supplant traditional fund managers 28Jun

More:

AI Helps Magicians Perform Mind Reading Tricks - IEEE Spectrum

Shantanu Nundy recognized the symptoms of rheumatoid arthritis when his 31-year-old patient suffering from crippling hand pain checked into Marys Center in Washington, D.C. Instead of immediately starting treatment, though, Nundy decided first to double-check his diagnosis using a smartphone app that helps with difficult medical cases by soliciting advice from doctors worldwide. Within a day, Nundys hunch was confirmed. The app had used artificial intelligence (AI) to analyze and filter advice from several medical specialists into an overall ranking of the most likely diagnoses. Created by the Human Diagnosis Project (Human Dx)an organization that Nundy directsthe app is one of the latest examples of growing interest in humanAI collaboration to improve health care.

Human Dx advocates the use of machine learninga popular AI technique that automatically learns from classifying patterns in datato crowdsource and build on the best medical knowledge from thousands of physicians across 70 countries. Physicians at several major medical research centers have shown early interest in the app. Human Dx on Thursday announced a new partnership with top medical profession organizations including the American Medical Association and the Association of American Medical Colleges to promote and scale up Human Dxs system. The goal is to provide timely and affordable specialist advice to general practitioners serving millions of people worldwide, in particular so-called "safety net" hospitals and clinics throughout the U.S. that offer access to care regardless of a patients ability to pay.

We need to find solutions that scale the capacity of existing doctors to serve more patients at the same or cheaper cost, says Jay Komarneni, founder and chair of Human Dx. Roughly 30 million uninsured Americans rely on safety net facilities, which generally have limited or no access to medical specialists. Those patients often face the stark choice of either paying out of pocket for an expensive in-person consultation or waiting for months to be seen by the few specialists working at public hospitals, which receive government funding to help pay for patient care, Komarneni says. Meanwhile studies have shown that between 25 percent and 30 percent (pdf) of such expensive specialist visits could be conducted by online consultations between physicians while sparing patients the additional costs or long wait times.

Komarneni envisions augmenting or extending physician capacity with AI to close this specialist gap. Within five years Human Dx aims to become available to all 1,300 safety net community health centers and free clinics in the U.S. The same remote consultation services could also be made available to millions of people around the world who lack access to medical specialists, Komarneni says.

When a physican needs help diagnosing or treating a patient they open the Human Dx smartphone app or visit the projects Web page and type in their clinical question as well as their working diagnosis. The physician can also upload images and test results related to the case and add details such as any medication the patient takes regularly. The physician then requests help, either from specific colleagues or the network of doctors who have joined the Human Dx community. Over the next day or so Human Dxs AI program aggregates all of the responses into a single report. It is the new digital equivalent of a curbside consult where a physician might ask a friend or colleague for quick input on a medical case without setting up a formal, expensive consultation, says Ateev Mehrotra, an associate professor of health care policy and medicine at Harvard Medical School and a physician at Beth Israel Deaconess Medical Center. It makes intuitive sense that [crowdsourced advice] would be better advice, he says, but how much better is an open scientific question. Still, he adds, I think its also important to acknowledge that physician diagnostic errors are fairly common. One of Mehrotra's Harvard colleagues has been studying how the AI-boosted Human Dx system performs in comparison with individual medical specialists, but has yet to publish the results.

Mehrotra's cautionary note comes from research that he and Nundy published last year in JAMA Internal Medicine. That study used the Human Dx service as a neutral platform to compare the diagnostic accuracy of human physicians with third-party symptom checker Web sites and apps used by patients for self-diagnosis. In this case, the humans handily outperformed the symptom checkers computer algorithms. But even physicians provided incorrect diagnoses about 15 percent of the time, which is comparable with past estimates of physician diagnostic error.

Human Dx could eventually help improve the medical education and training of human physicians, says Sanjay Desai, a physician and director of the Osler Medical Training Program at Johns Hopkins University. As a first step in checking the service's capabilities, he and his colleagues ran a study where the preliminary results showed the app could tell the difference between the diagnostic abilities of medical residents and fully trained physicians. Desai wants to see the service become a system that could track the clinical performance of individual physicians and provide targeted recommendations for improving specific skills. Such objective assessments could be an improvement over the current method of human physicians qualitatively judging their less experienced colleagues. The open question, Desai says, is whether the algorithms can be created to provide finer insights into an [individual] doctors strengths and weaknesses in clinical reasoning.

Human Dx is one of many AI systems being tested in health care. The IBM Watson Health unit is perhaps the most prominent, with the company for the past several years claiming that its AI is assisting major medical centers and hospitals in tasks such as genetically sequencing brain tumors and matching cancer patients to clinical trials. Studies have shown AI can help predict which patients will suffer from heart attacks or strokes in 10 years or even forecast which will die within five. Tech giants such as Google have joined start-ups in developing AI that can diagnose cancer from medical images. Still, AI in medicine is in its early days and its true value remains to be seen. Watson appears to have been a success at Memorial Sloan Kettering Cancer Center, yet it floundered at The University of Texas M. D. Anderson Cancer Center, although it is unclear whether the problems resulted from the technology or its implementation and management.

The Human Dx Project also faces questions in achieving widespread adoption, according to Mehrotra and Desai. One prominent challenge involves getting enough physicians to volunteer their time and free labor to meet the potential rise in demand for remote consultations. Another possible issue is how Human Dx's AI quality control will address users who consistently deliver wildly incorrect diagnoses. The service will also require a sizable user base of medical specialists to help solve those trickier cases where general physicians may be at a loss.

In any case, the Human Dx leaders and the physicians helping to validate the platform's usefulness seem to agree that AI alone will not take over medical care in the near future. Instead, Human Dx seeks to harness both machine learning and the crowdsourced wisdom of human physicians to make the most of limited medical resources, even as the demands for medical care continue to rise. The complexity of practicing medicine in real life will require both humans and machines to solve problems, Komarneni says, as opposed to pure machine learning.

Read more here:

Can a Crowdsourced AI Medical Diagnosis App Outperform Your Doctor? - Scientific American

Log in

Create an account for free access to:

With a free New Scientist account you'll enjoy increased access to New Scientist content and ideas.

Every week the editors release a selection of articles to New Scientist account holders. These articles are available exclusively to logged in account holders and subscribers. The editors selection can range from new features, opinions and interviews to fascinating content from the New Scientist archive.

You'll also receive the latest news and top stories in your inbox every week with the New Scientist email newsletter.

Get more from New Scientist. To create your free account, simply complete this quick form.

Special rates for students, teachers, libraries, schools, colleges and universities

Special rates for companies and group subscriptions

Give New Scientist to a friend or loved one, or activate your gift subscription

View original post here:

There is a good case to unleash job-killing AI on the high seas - New Scientist

Inside a red-bricked building on the north side of Washington DC, internist Shantanu Nundy rushes from one examining room to the next, trying to see all 30 patients on his schedule. Most days, five of them will need to follow up with some kind of specialist. And odds are, they never will. Year-long waits, hundred-mile drives, and huge out-of-pocket costs mean 90 percent of Americas most needy citizens cant follow through on a specialist referral from their primary care doc .

But Nundys patients are different. They have access to something most people dont: a digital braintrust of more than 6,000 doctors, with expert insights neatly collected, curated, and delivered back to Nundy through an artificial intelligence platform. The online system, known as the Human Diagnosis Project, allows primary care doctors to plug into a collective medical superintelligence, helping them order tests or prescribe medications theyd otherwise have to outsource. Which means most of the time, Nundys patients wait days, not months, to get answers and get on with their lives.

In the not-too-distant future, that could be the standard of care for all 30 million people currently uninsured or on Medicaid. On Thursday, Human Dx announced a partnership with seven of the countrys top medical institutions to scale up the project, aiming to recruit 100,000 specialistsand their expert assessmentsin the next five years. Their goal: Close the specialty care gap for three million Americans by 2022.

In January, a single mom in her thirties came to see Nundy about pain and joint stiffness in her hands. It had gotten so bad that she had to stop working as a housekeeper, and she was growing desperate. When Nundy pulled up her chart, he realized she had seen another doctor at his clinic a few months prior, who referred her to a specialist. But once the patient realized shed have to pay a few hundred dollars out of pocket for the visit, she didnt go. Instead, she tried get on a wait list at the public hospital, where she couldnt navigate the paperworkEnglish wasnt her first language.

Now, back where she started, Nundy examined the patients hands, which were angrily inflamed. He thought it was probably rheumatoid arthritis, but because the standard treatment can be pretty toxic, he was hesitant to prescribe drugs on his own. So he opened up the Human Dx portal and created a new case description: 35F with pain and joint stiffness in L/R hands x 6 months, suspected AR. Then he uploaded a picture of her hands and sent out the query.

Within a few hours a few rheumatologists had weighed in, and by the next day theyd confirmed his diagnosis. Theyd even suggested a few follow-up tests just to be sure, and advice about a course of treatment. I wouldnt have had the expertise or confidence to be able to do that on my own, he says.

Nundy joined Human Dx in 2015, after founder Jayanth Komarneni recruited him to pilot the platforms core technologies. But the goal was always to go big. Komarneni likens the network to Wikipedia and Linux, but instead of contributors donating encyclopedia entries or code, they donate medical expertise. When a primary care doc gets a perplexing patient, they describe their background, medical history, and presenting symptomsmaybe adding an image of an X-ray, a photo of a rash, or an audio recording of lung sounds. Human Dxs natural language processing algorithms will mine each case entry for keywords to funnel it to specialists who can create a list of likely diagnoses and recommend treatment.

Now, getting back 10 or 20 different doctors takes on a single patient is about as useful as having 20 friends respond individually via email to a potluck invitation. So Human Dxs machine learning algorithms comb through all the responses to check them against all the projects previously stored case reports. The network uses them to validate each specialist's finding, weight each one according to confidence level, and combine it with others into a single suggested diagnosis. And with every solved case, Human Dx gets a little bit smarter. With other online tools if you help one patient you help one patient, says Komarneni. Whats different here is that the insights gained for one patient can help so many others. Instead of using AI to replace jobs or make things cheaper were using it to provide capacity where none exists.

Komarneni estimates that those electronic consults can handle 35 to 40 percent of of specialist visits, leaving more time for people who really need to get into the office. Thats based on other models implemented around the country at places like San Francisco General Hospital, UCLA Health System, and Brigham and Womens Hospital. SFGHs eReferral system cut the average waiting time for an initial consult from 112 days to 49 within its first year.

That system, which is now the default for every SFGH specialty, relies on dedicated reviewers who get paid to respond to cases in a timely way. But Human Dx doesnt have those financial incentivesits service is free. Today, though, by partnering with the American Board of Medical Specialities, Human Dx can now offer continuing education and improvement credits to satisfy at least some of the 200 hours doctors are required to complete every four years. And the American Medical Association, the nations largest physician group, has committed to getting its members to volunteer, as well as supporting program integrity by verifying physicians on the platform.

Nick Stockton

Veritas Genetics Scoops Up an AI Company to Sort Out Its DNA

Megan Molteni

Thanks to AI, Computers Can Now See Your Health Problems

Megan Molteni

The Chatbot Therapist Will See You Now

Its a big deal to have the AMA on board. Physicians have historically been wary of attempts to supplant or complement their jobs with AI-enabled tools. But its important to not mistake the organizations participation in the alliance for a formal pro-artificial intelligence stance. The AMA doesnt yet have an official AI policy, and it doesnt endorse any specific companies, products, or technologies, including Human Dxs proprietary algorithms. The medical AI field is still young, with plenty of potential for unintended consequences.

Like discrepancies in quality of care. Alice Chen, the chief medical officer for the San Francisco Health Network and co-director of SFGHs Center for Innovation in Access and Quality, worries that something like Human Dx might create a two-tiered medical system, where some people get to actually see specialists and some people just get a computerized composite of specialist opinions. This is the edge of medicine right now, says Chen. You just have to find the sweet spot where you can leverage expertise and experience beyond traditional channels and at the same time ensure quality care.

Researchers at Johns Hopkins, Harvard, and UCSF have been assessing the platform for accuracy, and recently submitted results for peer-review. The next big hurdle is money. The project is currently one of eight organizations in contention for a $100 million John D. and Catherine T. MacArthur Foundation grant. If Human Dx wins, theyll spend the money to roll out nationwide. The alliance isnt contingent on the $100 million award, but it would certainly be a nice way to kickstart the processespecially with specialty visits accounting for more than half of all trips to the doctors office.

So its possible that the next time you go in for something that stumps your regular physician, instead of seeing a specialist across town, youll see five or 10 from around the country. All it takes is a few minutes over lunch or in an elevator to put on a Sherlock Holmes hat, hop into the cloud, and sleuth through your case.

The rest is here:

Want a Diagnosis Tomorrow, Not Next Year? Turn to AI - WIRED

Use your brain! Seattle-based language analytics company Textio says the term AI has lost its appeal. (BigStock Photo)

Leave it to the experts in machine learning to figure out that machine learning has lost its oomph. When it comes to terms in job postings which help fill those jobs more quickly, artificial intelligence and AI are also waning, according to Textio, the Seattle startup that analyzes the words that work.

In a new blog post on Wednesday, Textio co-founder and CEO Kieran Snyder takes a shot at the many job listings which rely on those terms to attract new talent. Its not the jobs that are going away, Snyder insists AI and machine learning are obviously still a huge deal for tech companies. Its just that the same-same nature of all those listings waters down the effect and slows the time it takes to fill an opening.

Snyder says AI and ML are going the way of big data, which she calls once cool, then cliche, now beginning to feel hopelessly dated.

Using its own data analysis to test this theory, Textio found that terms like AI and artificial intelligence were hot a couple years ago and did indeed pay off, with jobs advertised with those terms filling nine days or 28-percent faster than average engineering jobs.

But in 2017, job listings with any of these phrases are filling between one and two days faster than average. The increase in investment in these areas has led to an increase in the number of jobs, and thus the number of listings using these phrases has jumped significantly.

A quick search of Amazon jobs because they have more than 7,200 openings in Seattle using the term AI returned 106 listings, and 123 when artificial intelligence was searched.

So, Textio says, if you want to stand out, dont sound like everyone else. The company looked at terms that are on the rise right now but you better hurry. It clearly wont take long for the now-hot deep learning and chatbot to lose their appeal.

See the original post here:

Trying to fill a job in AI? Stop using the term AI, advises listings language experts Textio - GeekWire

A company can have the best technology in the world. It can have the strongest talent. It can have the coolest product ideas. But to train the algorithms that will deliver the intelligence to transform our cities, it needs data. To wit: The company with the most data wins.

Jessi Hempel is Backchannel's editorial director.

Sign up to get Backchannel's weekly newsletter.

Thats why earlier this year, after leaving Microsoft the previous fall, legendary engineer Qi Lu headed to Beijing to become Baidu's chief operating officer. At his former job, he was, among other things, CEO Satya Nadellas top deputy in helping to lead the companys AI strategy. Clearly, he saw more opportunity across the Pacific: In China, 731 million peoplenearly twice the entire population of the United Statesare online. Says Lu: China has the structural advantage.

On July 26, while Lu was visiting Silicon Valley, we sat down for an exclusive interview. Lu offered up an eye-opening explanation of how Baidu stands to dominate AI in China. And most places in the world, Lu notes, have much more in common with the tiny homes of the Chinese than the sprawling North American McMansions. He believes that could be Chinas biggest advantage in rolling out AI to global markets. Sure, Americas tech giants may have the lead in talentfor nowbut Lu believes that Baidu has what it will take to conquer the world.

Jessi Hempel: In the time since youve arrived at Baidu, theres been a reorganization. As COO, whats your role at the company?

I work very, very closely with Robin [Li, Baidu CEO]. We make sure he and I are fully in sync. I run R&D, sales, and marketing, because I want to make sure that our overall strategy is fully, fully in sync. Thats number one. Number two, I feel that were now much more clear and focused, in terms of strategy. Its really two battles. One is strengthening our mobile foundations. The other is leading the AI era.

How do you describe your AI strategy?

We believe the best way to commercialize AI technology is to build ecosystems. Essentially, to enable our partners to better accelerate their pace of innovation, using healthy, stable economic models to build strong, long-term win-wins for our developers and partners. The baseline is Baidu Brain [the term Baidu uses for all of its AI assets]. Its broader and more extensive than what Microsoft and Google offer today in the United States, because its a platform. We have 60 different types of AI services in our suite we call Baidu Brain.

Mark Harris

How Peter Thiel's Secretive Data Company Pushed Into Policing

Gabriel Nicholas

Ethereum Is Coding's New Wild West

Susan Crawford

Jeff Bezos Should Put His Billions Into Libraries

Scott Rosenberg

Bitcoin Makes Even Smart People Feel Dumb

And were the first major company to clearly separate the perceptual and the cognitive layer. Perceptive capability and the cognitive are related, but they are quite different. Most of the [other] AI platforms bundle them together.

What is Baidus equivalent of Siri or Cortana?

We are focusing on two platforms to bring our customers and partners together. The first platform we call DuerOS. DuerOS is a natural language-based, conversation-based, human computing platform. Very much like Alexa, Google Now, Siri, or Cortana in the United States. The only difference is DuerOS is so far ahead of anybody else. DuerOS in China has accumulated more conversation-based skill sets than anybody else. We have 10 major domains [and] over 100 sub-domains of conversational skills that we developed. Were also building up an emerging partner ecosystem. So our partners are building more and more skill sets. Amazon, perhaps, has more than Baidu right now, because they have a larger partner ecosystem in the United States. But compared to most companies, in China, were clearly leading.

Number two, we are also clear leaders in partners. DuerOS today is in over 100 brands of private home appliances, whether its refrigerators, air conditioners, TVs, storytelling machines, or speakers.

How does the US market for voice technology compare to the Chinese market?

The home environment is very different. Because were talking about voice interactions. The acoustic environment, the pattern of noises, will be very different. Alexa, Echo, and Cortana are optimized for American homes. In my view, this only works in North America and maybe a portion of Europe. Essentially, the assumption is that you have spacious homes; you have several rooms. In China, thats not the case at all. For our target, even for the young generation with high incomes, typically they have 60 square meters [645 square feet], sometimes 90 square meters [970 square feet].

We have better opportunities to globalize DuerOS, because guess what? A home in Japan, a home in India, or a home in Brazil, is a lot closer to a home in China than a home in North America.

Bloomberg/Getty Images

So, thats different. Whats similar?

The similarity part is the technology. The core technology is still speech recognition, signal processing, natural language understanding, and the platform. Our platform architecture, in many ways, is very similar to Amazon. In my view, Amazon is doing a very great job. Even though I worked at Microsoft. Im always gonna be rooting for Microsoft. But honestly, Amazon is leading.

But dont you think that Amazons handicap is on its back end, in that it cant keep up on the technology side with Google and Microsoft?

I worked on Cortana four and a half years ago. At the time we all were like, Amazon, yeah, that technology is so far behind. But one thing I learned is that in this race to AI, its actually more about having the right application scenarios and the right ecosystems. Google and Microsoft, technologically, were ahead of Amazon by a wide margin. But look at the AI race today. The Amazon Alexa ecosystem is far ahead of anybody else in the United States. Its because they got the scenario right. They got the device right. Essentially, Alexa is an AI-first device.

Microsoft and Google made the same mistake. We focused on Cortana on the phone and PC, particularly the phone. The phone, in my view, is going to be, for the foreseeable future, a finger-first, mobile-first device. You need an AI-first device to solidify an emerging base of ecosystems.

Its become so much clearer, living in China, what AI-first really means. It means you interact with the technology differently from the start. It has to be voice or image recognition, facial recognition, in the first interactions. You can use a screen or touch, but thats secondary.

At Baidu [headquarters], its all face recognition-based. At the vending machine at Baidu, you can buy stuff with voice and a face. And were also working on a cafeteria project. Our goal is, when you go to a cafeteria, you walk away with food.

Technically, thats possible now in a lot of places, but that doesnt mean people are receptive to it.

Its not all technology. Its about the structure of the environmentthe culture, the policy regime. This is why AI plus China, to me, is such an interesting opportunity. Its just different cultures, different policy regimes, and a different environment.

Scott Rosenberg

Inside Salesforces Quest to Bring Artificial Intelligence to Everyone

Jessi Hempel

Inside Microsoft's AI Comeback

Steven Levy

How Google is Remaking Itself as a Machine Learning First Company

Steven Levy

The iBrain Is Hereand Its Already Inside Your Phone

So how about the ethical consequences of the tools that were creating? Do people have the same types of conversations at Baidu as they do at Microsoft?

Similar. Protection of privacy is of paramount importance to us. Ultimately, our users trust in our technology. So, this is something we talk quite a bit about. And we are going to continue to seriously invest in capabilities to make sure that you can trust our services, in terms of privacy. For example, we talked about voice interactions. Were working on technologies that would prevent the unintended activation of smartphones. Its because we know that people dont want their conversations to be shipped to the Cloud. I may have very private conversations in my living room. [But sometimes] the speakers think you are trying to wake them up, and then send those bits to the Cloud.

Do you think that Chinese consumers care as much? Do you think that they expect something different, by virtue of the fact that they live under a different political environment?

Our assumption is that people will care about this. Ultimately, we believe people are rational. If theres a compelling benefit, people will weigh the consequences and then make those choices. I think this is global.

Baidu announced an ambitious self-driving initiative called Apollo this spring, and youve announced 50 partnerships so far. Why are you doubling down on autos?

If you want to truly build digital intelligence to be able to acquire knowledge, make decisions, and adapt to the environment, you need to build autonomous systems. In autonomous systems, the car is the first major commercial application that is going to land.

Its just like the phone ecosystem today. The phone ecosystem is the largest silicon software ecosystem. I believe the same thing will happen for the autonomous system. The car is going to build a larger ecosystem. And the same set of capabilitieshardware, sensors, chip sets, softwarewill be used to build industry robots, home robots. We want to have hundreds of companies and universities all at work on this, building a very large ecosystem. Then we can build robots, build drones, and build all those autonomous systems. So, to me, autonomy is a key.

You were instrumental in developing Apollo, right?

I am the COO of the company, but I run that business directly. For the last three plus months, I probably spent about about 40 percent of my time on the autonomous driving technology producttalking to customers; talking to partners. Essentially, from where things are today, toward the future of being able to be fully autonomous, the fundamental technological path for the self-driving technology is the speed of iterations.

What does that speed depend on?

Essentially, how much data you can get. Because to be able to drive on the road, you have to drive different kinds of roads in different kinds of conditionslighting, weather, whether its wet, how much physical pressure is on your tires. And with Apollo, we will be able to pull together all the resources, particularly the data resources, in a way that enables everybody to be better off.

We wrote a manifesto of Apollo. Essentially, there are four principles. Each is important. One is open capability. At Baidu, we open up our capabilityin code, in services, in datato all partners. This works particularly well in China, because China is highly, highly fragmented. Theres more than 250 car OEMs [original equipment manufacturers], unlike the United States, which is a heavily concentrated industry. None of the OEMs will have the full capabilities to build out deep R&Ds. With our code base that we released on July 5, [we will make it possible for] one person to assemble a vehicle in three days that can do autonomous driving in limited forms and start on R&Ds.

The second is shared resources. Essentially, with the Apollo design, there are two tiers. You are able to use the Apollo code and capability, and some data sets, with no strings attached. The second tier is enables you to use all the data that Baidu providesHD maps, the training databut we ask you to contribute your data. However, theres a key principle. The more you contribute, the more you should be able to get back.

The third principle is the accelerating pace of innovation. Essentially, because were able to put together more data, we are able to achieve more capability in our simulation engines. We enable everybody, collectively, to innovate at a much faster pace.

And the fourth principle is sustained win-win. Baidu is the biggest model. Its going to focus on delivering high-end services, high-value services, HD maps, [and] security services. Were competing against nobody. We enable each OEM, whether its Bosch, Continental, or Nvidia, to be able to do more.

This is the reason I created a subsidiary in the United States, Apollo US. And, also Apollo Singapore. The Singapore government essentially was like, Wow, this isJust come to Singapore. Im ready to invest.

Bloomberg/Getty Images

What needs to happen to enable fully autonomous vehicles in China?

Technology alone is not going to enable self-driving cars for a long time. Ill give you just a simple example. Lets say theres some kind of a road incident in a city, and the police come, and theres no signs. Say, he or she just hand writes a sign on a piece of paper to say, Please slow down to less than five miles an hour. Watch carefully as you proceed. And they hold it up. You need the technology to be able to read handwriting and understand the human language to be able to do that. Thats going to take a long, long time.

To enable full autonomy, you need new rules, new laws. Thats number one. Number two, as part of Apollo, working with all our partners, we actually found out theres so much more commercialization, much, much earlier than full autonomy. The Audi 8 is great example. Essentially, the car automatically follows the flows in heavy jammed traffic. And thats common to Beijing, Shanghai, and in the Bay Area. Now you just let the car drive, and you can read something and do something else. And besides following a car, there are so many other scenarios.

When we first met, you were at Microsoft. You left several months before you arrived at Baidu. Why?

I broke my leg in October 2016. I needed two surgeries. Bill, Satya, and I are still super close, so when I go to Seattle usually I go see Satya at his house. I visit with Bill. I promised to be their personal advisors.

It seems 2017 is a bellwether year for AI development in China. Whats significant about this year?

Its a combination of the readiness of technology and the number of industry verticals AI can commercialize. And at the global scale, I do feel that theres opportunities for China and the United States to collectively drive the world forward. Im probably influenced by Bill Gates quite a bit. He always talked about how the world economy right now, for practical purposes, is a single engine economy. The United States has five percent of the worlds population, but produces about 24 percent of economic output and 60 percent of innovation. Its just not going to be able to sustain the pace of growth, because the world has seven billion people. Maybe three-plus billion people are living a modern life. We have transportation; we eat processed food; we have refrigerators. But then theres a sharp drop off. The other populations are living in completely different living conditions. Our job is to elevate everybody to living a modern life. How do you do that? By more innovations, better growth. Really, China should become the second innovation engine, and [Gates] genuinely believes a more innovative China, and a more developed China, is a great thing for the world. I believe that, too.

When you began beefing up your AI resources several years ago, you focused on building a lab in Silicon Valley. When American researcher Andrew Ng left Baidu last spring, his replacement to head Baidus AI labs was in China. Has AIs talent in China caught up to the US?

The United States is still overall stronger, no question. But the gap between China and the United States is rapidly closing. Theres no doubt about that. And since I have lived in China for over six months now, honestly, I read more papers, I talk to more AI developers, and you can feel the strength of the talent base.

Baidu will do more and more AI work in China, for sure. But at the same time, were continuing to invest in the United States, in the Bay Area and also Seattle. We just opened a Seattle campus, because we acquired a company called Kitt.ai. For the very top echelon of the talent, the United States is still better, and we want to fully leverage that.

View post:

How Baidu Will Win China's AI raceand, Maybe, the World's - WIRED

Scam artists and others use a method called cloaking to make their links on Facebook (left) look more innocuous than the content they serve (center). They even show Facebook reviewers fake versions of their sites (right). Credit: Facebook

Facebook says it is intensifying its efforts to control scams and fake news by taking a harder line on "cloaking," a tactic that bad actors use across the web to avoid detection.

"We've recently been ramping up our enforcement," says Rob Leathern, Facebook product management director. "We are making it clear: We don't tolerate cloaking."

Cloaking is a longtime but straightforward practice of so-called black hats online. Fraudulent marketers, pornographers and even racists have used it to disguise their true nature in search results and in social feeds.

Facebook was already seeking out links on its platform to landing pages that don't deliver what was promised, serve deceptive ads or have too many ads. Possible penalities included warnings, lower visibility for links and outright bans from the platform.

Some of the more elaborate cloaking efforts, however, are tough to recognize. One method is to show Facebook one version of a site to gain approval, then serve something different when Facebook users arrive.

"For example, they will set up web pages so that when a Facebook reviewer clicks a link to check whether it's consistent with our policies, they are taken to a different web page than when someone using the Facebook app clicks that same link," Leathern wrote in a blog post Wednesday. "Cloaked destination pages, which frequently include diet pills, pornography and muscle building scams, create negative and disruptive experiences for people."

The company says it has beefed up both human reviews and its artificial intelligence algorithms to spot cloaking. "In the past few months these new steps have resulted in us taking down thousands of these offenders and disrupting their economic incentives for misleading people," Leathern wrote.

Widespread phenomenon Jessie Daniels, a professor of sociology at Hunter College-City University of New York, has studied cloaking since the 1990s, and says it's very much mixed in with the problems of false news and racist propaganda propagating on the web.

Cloaking is often done by "someone who is concealing authorship in order to disguise a political agenda," Daniels says.

She points to search results from a query as simple as "Martin Luther King." The first page of Google results includes the site martinlutherking.org, which promises "historical trivia, articles and pictures" in "a valuable resource for teachers and students alike." The site is actually run by the racist group Stormfront.

"The thing I look at are people who are politically motivated and people close behind them who are profit-motivated," Daniels says. "It's always the pornographers and white supremacists."

Stopping cloaking will be difficult for any company, even Facebook, Daniels adds. In the end, it takes a lot of human reviews. Facebook has recently hired 3,000 people to comb the site for objectionable videos, but it wouldn't say how many people are dedicated to cloaking patrol.

Read this article:

Facebook Uses AI to Stop Spammers From Cloaking Their Tracks ... - AdAge.com

Kavya Kopparapu might be considered something of a whiz kid. After all, she had yet to enter her senior year of high school when she started Eyeagnosis, a smartphone app and 3D-printed lens that allows patients to be screened for diabetic retinopathy with a quick photo, avoiding the time and expense of a typical diagnostic procedure. In June 2016, Kopparapus grandfather had recently been diagnosed with diabetic retinopathy, a complication of diabetes that damages retinal blood vessels and can eventually cause blindness. He caught the symptoms in time to receive treatment, but it was close. A little too close for Kopparapus comfort. According to the IEEE Spectrum, Kopparapu, her 15-year-old brother Neeyanth and her classmate Justin Zhang trained an artificial intelligence system to scan photos of eyes and detect, and diagnose, signs of diabetic retinopathy. She unveiled the technology at the OReilly Artificial Intelligence conference in New York City in July. After diving into internet-based research and emailing opthamologists, biochemists, epidemiologists, neuroscientists and the like, she and her team worked on the diagnostic AI using a machine-learning architecture called a convolutional neural network. CNNs, as theyre called, parse through vast data sets -- like photos -- to look for patterns of similarity, and to date have shown an aptitude for classifying images. The network itself was the ResNet-50, developed by Microsoft. But to train it to make retinal diagnoses, Kopparapu had to feed it images from the National Institute of Healths EyeGene database, which essentially taught the architecture how to spot signs of retinal degeneration. One hospital has already tested the technology, fitting a 3D-printed lens onto a smartphone and training the phones flash to illuminate the retinas of five different patients. Tested against opthalmologists, the system went five for five on diagnoses. Kopparapus invention still needs lots of tests and additional data to prove its efficacy before it sees widespread clinical adoption, but so far, its off to a pretty good start. Eyeagnosis is operating in a space that's recently become interesting to some very large companies. Last fall, a team of Google researchers published a paper in the Journal of the American Medical Association showing that Google's deep learning algorithm, trained on a large data set of fundus images, can detect diabetic retinopathy with better than 90 percent accuracy. That algorithm was then tested on 9,963 deidentified images retrospectively obtained from EyePACS in the United States, as well as three eye hospitals in India. A second, publicly available research data set of 1,748 was also used. The accuracy was determined by comparing its diagnoses to those done by a panel of at least seven U.S. board-certified ophthalmologists. The two data sets had 97.5 percent and 96.1 percent sensitivity, and 93.4 percent and 93.9 percent specificity respectively.

And Google isnt the only player in that space. IBM has a technology utilizing a mix of deep learning, convolutional neural networks and visual analytics technology based on 35,000 images accessed via EyePACs; in research conducted earlier this year, the technology learned to identify lesions and other markers of damage to the retinas blood vessels, collectively assessing the presence and severity of disease. In just 20 seconds, the method was successful in classifying diabetic retinopathy severity with 86 percent accuracy, suggesting doctors and clinicians could use the technology to have a better idea of how the disease progresses as well as identify effective treatment methods.

Lower-tech options are also taking a stab at improving access to screenings. Using a mix of in-office visits, telemedicine and web-based screening software, the Los Angeles Department of Health Services has been able to greatly expand the number of patients in its safety net hospital who got screenings and referrals. In an article published in the journal JAMA Internal Medicine, researchers describe how the two-year collaboration using Safety Net Connects eConsult platform resulted in more screenings, shorter wait times and fewer in-person specialty care visits. By deploying Safety Net Connects eConsult system to a group of 21,222 patients, the wait times for screens decreased by almost 90 percent, and overall screening rates for diabetic retinopathy increased 16 percent. The digital program also eliminated the need for 14,000 visits to specialty care professionals

More:

Teenage team develops AI system to screen for diabetic retinopathy - MobiHealthNews

Artificial intelligence as a technology is rapidly growing, but much is still being learned about how AI and autonomous systems make decisions based on the information they collect and process.

To better explain those relationships so humans and autonomous systems can better understand each other and collaborate more deeply, researchers at PARC, the Palo Alto Research Center, have been awarded a multi-million dollar federal government contract to create an "interactive sense-making system" that could answer many related questions.

The research for the proposed system, called COGLE (CommonGroundLearning andExplanation), is being funded by the Defense Advanced Research Projects Agency (DARPA),using an autonomous Unmanned Aircraft System (UAS) test bed but would later be applicable to a variety of autonomous systems.

The idea is that since autonomous systems are becoming more widely used, it would behoove humans who are using them to understand how the systems behave based on the information they are provided, Mark Stefik, a PARC research fellow who runs the lab's human machine collaboration research group, told ITPro.

"Machine learning is becoming increasing important," said Stefik. "As a consequence, if we are building systems that are autonomous, we'd like to know what decisions they will make. There is no established technique to do that today with systems that learn for themselves."

In the field of human psychology, there is an established history about how people form assumptions about things based on their experiences, but since machines aren't human, their behaviors can vary, sometimes with results that can be harmful to humans, said Stefik.

In one moment, an autonomous machine can do something smart or helpful, but then the next moment it can do something that is "completely wonky, which makes things unpredictable," he said. For example, a GPS system seeking the shortest distance between two points could erroneously and catastrophically send a user driving over a cliff or the wrong way onto a one-way street. Being able to delve into those autonomous "thinking" processes to understand them is the key to this research, said Stefik.

The COGLE research will help researchers pursue answers to these issues, he said. "We're insisting that the program be explainable," for the autonomous systems to say why they are doing what they are doing. "Machine learning so far has not really been designed to explain what it is doing."

The researchers involved with the project will essentially have roles educators and teachers for the machine learning processes to improve their operations and make it more useable and even more human like, said Stefik. "It's a sort of partnership where humans and machines can learn from each other."

That can be accomplished in three ways, he added, including reinforcement at the bottom level, using reasoning patterns like the ones humans use at the cognitive or middle level, and through explanation at the top sense-making level. The research aims to enable people to test, understand, and gain trust in AI systems as they continue to be integrated into our lives in more ways.

The research project is being conducted under DARPA's Explainable Artificial Intelligence (XAI) program, which seeks to create a suite of machine learning techniques that produce explainable models and enable human users to understand, appropriately trust, and effectively manage the emerging generation of artificially intelligent partners.

PARC, which is a Xerox company, is conducting the COGLE work with researchers at Carnegie Mellon University, West Point, the University of Michigan, the University of Edinburgh and the Florida Institute for Human & Machine Cognition. The key idea behind COGLE is to establish common ground between concepts and abstractions used by humans and the capabilities learned by a machine. These learned representations would then be exposed to humans using COGLE's rich sense-making interface, enabling people to understand and predict the behavior of an autonomous system.

Go here to read the rest:

Advancing AI by Understanding How AI Systems and Humans Interact - Windows IT Pro