Posts Tagged ‘DARPA’
Science Policy Around the Web – February 19, 2019
By: Mohor Sengupta Ph.D.
New AI fake text generator may be too dangerous to release, say creators
Artificial intelligence. In a Beijing competition last year, it identified patterns in brain MRI scans from hundreds of individuals and diagnosed their conditions far more accurately than human doctors. In modern journalism, AI is being trained by humans to scan financial reports and dish out an immediate news story with pertinent facts and figures. Bloomberg uses the bot “Cyborg”, which helps it compete against its rival, Reuters, in financial journalism. The Washington Post uses “Heliograf”, which covered the 2016 Olympics and the 2016 US elections. But, AI can create a completely fake human image by teaching itself from several real images. It can produce a completely fake video by literally putting words in a person’s mouth. And, owners of the nonprofit research company, “OpenAI”, realized, to their concern, that a new AI text generator called GPT2, created by its researchers, is so smart that it can make up connected and logical sentences following a random one-liner input.
Take this example. When fed with the opening line of George Orwell’s Nineteen Eighty-Four – “It was a bright cold day in April, and the clocks were striking thirteen” – GPT2 comes up with this: “I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. A hundred years from now. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science.” It recognizes the futuristic reference in the input. It had an interesting and obviously fake statement to say about UK Prime Minister Theresa May when fed with an unfinished line from a Guardian article about Brexit. In short, GPT2 is quick and creative.
It is not difficult to imagine the mischief this kind of groundbreaking technology can create. Generating fake news is one that comes to the mind easily.
Take for example, this false and incomplete phrase presented to GPT2 by a group of humans at MIT: “Russia has declared war on the United States after Donald Trump accidentally”.
OpenAI chillingly demonstrated how it can be a lethal hoodwinker by completing that phrase into this: “Russia has declared war on the United States after Donald Trump accidentally fired a missile in the air. Russia said it had “identified the missile’s trajectory and will take necessary measures to ensure the security of the Russian population and the country’s strategic nuclear forces.” The White House said it was “extremely concerned by the Russian violation” of a treaty banning intermediate-range ballistic missiles. The US and Russia have had an uneasy relationship since 2014, when Moscow annexed Ukraine’s Crimea region and backed separatists in eastern Ukraine.”
Don’t panic, none of that is true!
Machine learning was used to train GPT2 on a dataset containing about 10 million articles, selected by scouring the social news site Reddit for links with more than three votes. “We need to perform experimentation to find out what they can and can’t do,” said Jack Clark, policy director of OpenAI. “If you can’t anticipate all the abilities of a model, you have to prod it to see what it can do. There are many more people than us who are better at thinking what it can do maliciously.” For now, the company has decided to keep GPT2 behind closed doors until such a time when they know the bot’s full potential. Only a simpler version will be made publicly available.
While there are skeptics of AI, there are also critics of the skeptics. They argue that humans can produce fake news as well as an AI. But take a moment to think of the number, variations, creativity and logical or factual supports that can be weaved in seconds by a bot like GPT2. How is it possible that any human can outsmart a program that has learnt 10 million articles about current and past politics, about wars, about famous fictional characters, about sports, about celebrities, about nature and about anything under the sun, and beyond?
Downsides of a technology cannot confine it behind closed doors. We have learnt that from the past. AI is becoming increasingly necessary in many areas, for example, identifying medical conditions from MRI as mentioned above. Nature Medicine recently published an article where an AI accurately diagnosed common childhood diseases by analyzing digital and electronic health records. But like any revolutionizing technology, AI can be misused for malicious purposes. And it will be used for that, sooner or later. The idea here is to prepare the world for what is about to become mainstream in a few years. Clark calls this an “escalator from hell”.
(Alex Hern, The Guardian)
Searching Tardigrades for Lifesaving Secrets
In the series “Cosmos: A Spacetime Odyssey”, Neil deGrasse Tyson talked about five cataclysmic events, separated by eons, that wiped out almost all life on Earth. Every time this happened, a miniscule proportion of existing life escaped and adapted to the new environment. Among this population there is an organism that survived ALL five purges. This animal is the size of a sand particle and it dwells in moisture. It is called a Water bear, or a Moss piglet, or Tardigrade.
It is not entirely clear how these organisms survive extreme environmental conditions, but research on their physiology suggest involvement of certain proteins that protect its cells from dehydration. In an event of environmental emergency, the tardigrade can desiccate itself within minutes into a firm, curled-up ball called a “tun”, only to resume normal functions within minutes of sensing moisture. This remarkable feat is called anhydrobiosis in scientific jargon. A group of scientists from Japan have found certain heat-soluble proteins that help tide over the anhydrobiosis period. Separately, another group of scientists from USA and Italy have found intrinsically disordered proteins (IDPs) specific to tardigrades (TDPs) to confer to them the protection against desiccation.
A team of three scientists at Harvard Medical School and are now using computational biology and machine learning to design TDPs tailored to slow down metabolism in human cells. “It really started out as a wacky, high-risk idea,” said Pamela Silver who spearheads the project. In 2008 Dr. Silver came across a grant challenge posted by the US military seeking novel solutions to stabilize hemorrhage of personnel in war zones. Together with her bioinformatician colleague and machine-learning expert Roger Chang and computational biologist Debora Marks, she set out to engineer novel TDPs that would lower the metabolism in eukaryotic cells during a state of shock. This is akin to “slowing biological time”, only to return back to normal pace when the critical period has passed. Although it is still unclear how these proteins might work, Dr. Chang has suggested that they might form a biological glass that physically immobilizes everything inside the cell during the stress period.
The group was recently awarded a five-year cooperative agreement by the Defense Advanced Research Projects Agency (DARPA) to pursue the idea. If it works, the engineered TDPs could be a benchmark in modern medicine, revolutionizing current understanding of not only trauma management, but also address several pressing issues, like, long-term transportation of protein-based drugs or to-be-transplanted organs and keeping the egg cell viable without freezing it.
(Steph Yin, New York Times)
Embryo ‘Adoption’ Is Growing, but It’s Getting Tangled in the Abortion Debate
In Vitro Fertilization (IVF) has changed the way people understand reproduction. It has made conceiving a reality for many women who are not able to conceive naturally for medical or social reasons. During the IVF process, sperm from a male donor are used to fertilize an egg cell from a female donor (generally the would-be biological parents). Usually, more than one embryo results from such a process, but only one is transplanted in the recipient’s uterus. Commonly, when a biological child is desired, donor and recipient are the same woman. In a relatively new trend, couples other than the donor are “adopting” these extra embryos. While this may seem as a very uncomplicated way of giving new life and a great option for many women, it is fraught with problems arising from religious and social issues.
Monica Broecker is a woman in her mid-forties who found out that she couldn’t conceive, after repeated miscarriages. She decided to adopt an embryo and approached National Embryo Donation Center, based in Knoxville, Tennessee, which is also the largest embryo donation clinic in USA. Despite being financially stable, Ms. Broecker was turned down by the agency because she is a single woman.
Some of the better-known embryo adoption agencies are funded by a grant by the Department of Health and Human Services (HHS), called the “Embryo Awareness Adoption Program” which has had $1 million in annual funding since 2003. To date, all the grant recipients, barring two, are affiliated with anti-abortion or Christian organizations.
Ms. Tyson, who works for Snowflake Embryo Adoption, one of the recipient agencies of the HHS grant, says that her clients are mostly Christian. In almost all embryo donations, the donor family selects the recipient. Ms. Tyson has had a difficult time finding donors for single women, LGBTQ people or people from diverse religious practices and atheists. She usually refers such recipients to Embryo Donation International, an agency that is not affiliated to any religious organization. Unsurprisingly, Embryo Donation International doesn’t receive the HHS grant.
On an average, embryo donation costs much less than a single round of IVF. Increasingly, many women wanting to become mothers are seeing this approach as a perfect synergy of their requirements and giving life to an embryo rather than it being frozen indefinitely. Matching with a donor is the biggest problem that many of them are experiencing. Although all grant recipients deny that they consider religion, sexual orientation or marital status while considering clients, statistics show a different picture. The government should look more carefully into its grant applicants and take measures to diminish bias in granting the awards.
(Caroline Lester, New York Times)
Have an interesting science policy link? Share it in the comments!
Science Policy Around the Web – July 25, 2017
By: Allison Dennis B.S.
Gene Drives
With Great Power There Must Also Come–Great Responsibility!
On the horizon of life-changing biotechnology up for ethical debate, nestled between CRISPR and whole genome sequencing, are gene drives, which have the potential to alter genes for better or for worse across generations. During sexual reproduction each of the two versions of a gene carried by a parent has a 50% chance of being inherited by each offspring. The frequency of each version of a gene across a population is influenced by rates of mutation, migration, genetic drift, and natural selection. Gene drives present the technology to circumvent these natural forces. By introducing molecular machines capable of damaging a particular version of a gene along with the version they prefer to the cells that give rise to eggs or sperm in an organism, scientists can shift the likelihood that their version will be inherited by that organism’s offspring from 50% to 100%. Upon fertilization the undesired gene will be damaged by the molecular machine and the desired gene will used as a template to repair the damaged copy, allowing two copies of the desired gene to be permanently introduced in the offspring and inherited by the next generation. Clever applications have been proposed to design mosquitoes resistant to malaria, mice unable to transmit lyme’s disease, or salmon able to grow to full size in half the time. More bold applications would use the technology to render female mosquitoes sterile, the ultimate insecticide. However, for each one of these beneficial applications exists the devastating opposite, which could be employed to accelerate the spread of disease. Altering population genetics of one species could accidentally devastate ecosystems.
U.S. defense organizations have taken notice of this powerful technology. DARPA, the Defense Advanced Research Project Agency, has launched the Safe Genes program in an effort to anticipate and address potential risks of introducing uncontrolled or undesired gene drives. The program awarded a collective $65 million to seven labs hoping to develop counter technologies including self-fizzling drives, chemical control methods, and gene drive vaccines. This summer, to delve deep into the intellectual discussion JASON, tackled the issue. This independent group of scientists, holding stellar academic records and top-secret clearances, meets once a year to address questions posed by the U.S. Department of Energy, Department of Defense, CIA, and FBI. However, their report is likely to be classified. (Ewen Callaway, Nature News)
Violence Against Women
Beginning to Understand the Nature of Intimate Partner Violence Through Data Curation
Careful evaluation of the nature of homicides of women has revealed that 55% result from intimate partner violence (IPV). The study conducted by the US Center for Disease Control (CDC) looked into the circumstances surrounding the deaths of 10,018 women over the age of 18 between 2003 and 2014 across 18 states. In addition to cases where the victims were intimate partners of the suspect, IPV-related homicides included cases where the female victims were friends, family, or those who intervened during an incident of IPV.
Nationwide political attention was drawn to the issue of IPV starting in the 1990s. The Violence Against Women Act was passed by Congress in 1994 and sought to legally define domestic violence as a crime external to the purview of private family matters. Research has revealed several risk factors associated with intimate partner violence, including threats with weapons, stalking, obsessive jealousy, sexual assault, and controlling behavior. However, the effectiveness of political and public health interventions remain unclear due to the overall decline in violence over the last decade and believed underreporting of individual incidence.
In an effort to more broadly understand the “who, when, where and how” surrounding violent deaths that occur in the United States, including those connected with IPV, the CDC created the National Violent Death Reporting System in 2002. By pooling information gathered by local law enforcement officers, coroners, medical examiners, and state agencies the CDC is hoping learn more about “why” so many violent deaths occur, towards the goal of developing and evaluating public health interventions. At its inception, funding only supported the participation of six states. However, involvement has been increasing from 17 states in 2006 and 42 in 2016 with the goal of eventually including all 50 states, U.S. territories, and the District of Columbia.
This study confirmed that homicide as a result of IPV occurs across all age groups and racial ethnic groups. However, young black and Hispanic women are disproportionately affected compared with white and Asian women of the same age group. Overall, black and indigenous women experienced significantly higher higher homicide rates, including non IPV related cases, than women of other races. Women died as a result of the use of firearms in 53.9% of all cases. While the “why” still remains unclear, this 15 year glance back sheds some light on the groups most affected by violence inflicted by their own partners, providing opportunity for targeted prevention. (Camila Domonoske, NPR)
Have an interesting science policy link? Share it in the comments!
Science Policy Around the Web – July 21, 2015
By: Amanda Whiting, Ph.D.
Funding policy
Congress pushes NIH to spur breakthroughs through prizes
A provision in the new 21st Century Cures bill, which passed through Congress last week, would create a new program from which to launch biomedical prize competitions at the National Institutes of Health (NIH). Prize competitions have been used by other federal agencies, such as the Defense Advanced Research Projects Agency (DARPA), to offer challenges in robotics and engineering tasks as way to speed up innovation and research. Though offering prizes for success can garner applications from sources outside of the traditional academic mainstream, “there only a few places where prizes really work in health,” said Christopher Frangione, vice president of prize development at XPRIZE, a private organization dedicated to facilitating global changes and innovation through competition. For example, long term studies involving basic, fundamental research might be poor candidates for a “grand challenge” competition, whereas benefits might be seen in areas that require an engineering feat, or would be helped by encouraging interdisciplinary research. Other opponents of the bill fear that prizes would reduce grant funding through other, more traditional means, and worry that having politically appointed advisers “threatens to undermine the independent peer-review process that is the bedrock of NIH funding,” said Rep. Frank Pallone (D–NJ). Still, any new avenue to pursue ground-breaking medical research is worthy of exploration! It will be interesting to see if NIH ends up following up with the potential Innovation Prizes Program. (Kelly Servick, ScienceInsider)
Scientific rigor and reproducibility
Collaborate and listen to reproduce research
A new report from the American Society for Cell Biology (ASCB) suggests that suggests that researchers in particularly hard to reproduce fields like cell biology, could better their results in replicating published data if they reached out to the original authors for assistance or method clarification. ASCB surveyed its membership and found that 60% of those who reported problems with reproducibility said that they were able to fix the issue by checking with the lab that conducted the experiment in question to resolve issues about the methods. While this may seem like an obvious solution, some researchers may be reluctant due to competition within the field or other reasons. One solution to this issue would be to have each field adopt their own consistent standards of proof, said Mark Winey, one of the report’s authors and a molecular biologist at the University of Colorado in Boulder and chair of the ASCB’s Data Reproducibility Task Force. In addition, a majority of respondents to the survey said that the quest to publish in high-profile journals hampered reproducibility, an opinion shared by Arturo Casadevall, a microbiologist and immunologist at Johns Hopkins Bloomberg School of Public Health in Baltimore, Maryland. “Science is messy, and high-impact journals often demand clean stories with a clear punchline,” he says. “That creates perverse incentives for cherry-picking data.” (Chris Woolston, Nature | Research Highlights: Social Selection)
Have an interesting science policy link? Share it in the comments!
Science Policy Around the Web – August 15, 2014
By: Bethanie L. Morrison, Ph.D.
photo credit: mkhmarketing via photopin cc
Our weekly linkpost, bringing you interesting and informative links on science policy issues buzzing about the internet.
Global Health
Using experimental drugs and vaccines against Ebola is ethical, WHO panel says
A 12-member World Health Organization (WHO) ethics panel has approved the use of experimental drugs to combat the latest Ebola outbreak in West Africa, the worst outbreak on record. Panelists and other disease experts initially worried that testing the experimental drugs in rural Africa may be seen as racist, yet after seeing the positive results of the experimental ZMapp on two Americans infected with the Ebola virus, they have put those concerns aside. While agreeing that the compassionate use of experimental therapeutics is warranted in this situation is a huge hurdle that had to be overcome, there are many more policy considerations that will be up for discussion at the next convening of ethics panel members in Geneva at the end of the month. Some of these roadblocks include whether to distribute the therapeutics to health care workers first, which authority makes the decisions about individual patient treatment and, most importantly, the fact that the experimental drugs in discussion are only available in limited quantities. “Much more ethical work needs to be done to create a sound infrastructure for compassionate use in humanitarian emergencies,” wrote Arthur Caplan, a bioethicist at New York University. (Kai Kupferschmidt)
Diplomacy, Defense, and Technology
Pentagon’s breakthrough human brain-inspired computer chip to power drones
Researchers at the Pentagon, as part of the Defense Advances Research Projects Agency (DARPA), have created a computer chip inspired by the synapses of the human brain. The chip, which contains over 5 billion transistors and more than 250 million life-like “synapses,” requires only a fraction of the electricity typically required to power commercially available computer chips. The decrease in energy requirements for the chip will make it much easier for military field use. In addition, the chip is powerful enough to “give unmanned aircraft or robotic ground systems with limited power budgets a more refined perception of the environment,” says Gill Pratt, the DARPA program manager. This will take some of the burden off of system operators as drones should be able to distinguish threats more accurately. Ultimately the development of this chip will allow for a much wider range of portable computing applications used for military and defense. (Douglas Ernst)
Science Communication
Ebola May Pose Little Threat to U.S., but It Looms Large on Twitter
Ebola is trending on Twitter. This fact has social scientists and biomedical scientists on high alert for completely different reasons, both of which converge on the implications of appropriate scientific communication. Jonah Berger, a professor of marketing at the Wharton School of the University of Pennsylvania, has found that heightened emotions increase story sharing, regardless of the truth. While Ebola remains a very rare disease, the emotions that its horrific symptoms stir up in the general public are very strong, enticing people to share stories regardless of the validity of their sources or the impact that spreading potentially false information may have on people and policy makers. People want to be more a part of the conversation and be “in the know” than a part of the alternative. While strong emotional issues are great for the social media model of marketing, they may not have the same impact on the biomedical research community who now has to spend countless amounts of time and resources explaining why the Ebola virus is not one to be concerned with in day to day living. Perhaps this is not such a bad thing; as it has pushed people like Dr. Thomas Frieden, the director of the Centers for Disease Control and Prevention (CDC), to release fact sheets about Ebola and other “emerging threats.” (Joshua A. Krisch)
Have an interesting science policy link? Share it in the comments!
Science Policy For All 