For the 20 percent of people with a mental illness, early identification of the condition is key to getting the best treatment. But people often suffer symptoms for months, even years, without receiving clinical attention. Part of the problem is that psychiatrists have few tools to identify mental illnesses; they rely mostly on self-reported data and observations from friends and family.
The field is, in some ways, "stuck in the prehistoric age," according to Michael Birnbaum, MD, an assistant professor at the Feinstein Institutes for Medical Research and an attending physician at Zucker Hillside Hospital and Lenox Hill Hospital at Northwell Health.
But digital tools could help bring psychiatry into the modern age.
"It became apparent, in my work with young folks, that social media was ubiquitous," Dr. Birnbaum told Big Think. "So, we started to think about ways that we could potentially explore the utility of the internet and social media in the way we diagnose our patients and the care that we provide."
The results of a recent study, conducted by Feinstein Institutes researchers and IBM Research, suggest that social media activity can provide useful insights into who's at risk of developing mental illnesses like mood disorders and schizophrenia spectrum disorders.
Published in the journal njp Schizophrenia, the study used machine-learning algorithms to analyze millions of Facebook messages and images, which were provided voluntarily by participants, ages 15 to 35. The data represented participants' Facebook activity for 18 months prior to hospitalization.
...the health disparity between people with mental illness and those without is larger than disparities attributable to race, ethnicity, geography or socioeconomic status.
Identifying psychiatric disorders
The goal was for the algorithms to analyze patterns in these datasets, then predict which group participants belonged to: schizophrenia spectrum disorders (SSD), mood disorders (MD), or healthy volunteers (HV). The results were promising, showing that the algorithms correctly identified:
- The SDD group with an accuracy of 52% (chance was 33%)
- The MD group with an accuracy of 57% (chance was 37%)
- The HV group with an accuracy of 56% (chance was 29%)
The study also showed interesting differences in Facebook activity among the groups, such as:
- The SSD group was more likely to use language related to perception (hear, see, feel).
- The MD and SSD groups were far more likely to use swear words and anger-related language.
- The MD group was more likely to use language related to biological processes (blood, pain).
- The SSD group was more likely to express negative emotions, use second-person pronouns and write in netspeak (lol, btw, thx).
- The MD group was more likely to post photos containing more blues and less yellows.
These differences tended to become more apparent in the months before a patient was hospitalized. But even 18 months before hospitalization, the results revealed signals that hinted participants might be on the path to developing a psychiatric disorder. That's where these tools may someday help improve early-identification efforts.
"In psychiatry, we often get a snapshot of somebody's life, for 30 minutes once a month or so," he said. "There's the potential to get much greater granularity with some of these new assessment tools. Facebook, for example, can allow us to understand somebody's thoughts and behaviors in a more real-time, longitudinal fashion, as opposed to cross-sectional moments in time."
Dr. Birnbaum noted that everyone has a unique style of online behavior and that certain behavioral changes may contain clues about mental health.
"The way that we're understanding this is that everybody has a digital baseline, a way they typically act and behave on social media and the internet," he said. "So, ultimately here we would want to identify this baseline for each individual—a fingerprint—and then monitor for changes over time, and identify which changes are concerning, and which are not."
Using digital tools to better identify psychiatric conditions could someday reduce the number of people who suffer without treatment.
"There's an alarming gap between the number of people who experience mental illness and those who receive care," said Michael Dowling, president and CEO of Northwell Health. "It's especially troubling when you consider that the health disparity between people with mental illness and those without is larger than disparities attributable to race, ethnicity, geography or socioeconomic status."
A step toward the future of psychiatry
Credit: Jewel Samad/AFP via Getty Images
Although previous research has examined the relationship between online activity and psychiatric disorders, the new study is unique because it paired online behavior with clinically confirmed cases of psychiatric disorders.
"The vast majority of the data thus far has been extracted from anonymous, or semi-anonymous individuals online, without any real way to validate the diagnosis or confirm the authenticity of the symptoms," Dr. Birnbaum said.
But before clinicians can use these kinds of digital approaches, researchers have more work to do.
"I think that we need much larger datasets," Dr. Birnbaum said. "We need to repeat these findings. We need to better understand how demographic differences, like age, ethnicity and gender, can play a role."
Privacy is another consideration. Dr. Birnbaum emphasized that these kinds of approaches would only be conducted on a voluntary basis, and that the Facebook data used in the recent study was anonymized, and the algorithms examined only individual words, not the context or meaning of sentences.
"This isn't about surveillance, or that Facebook should somehow be monitoring us," Dr. Birnbaum said. "It's about giving the power to the patient. I imagine a world where patients could come into the doctor's office and express their concerns, but also provide some additional clinically meaningful information that they own."
Dr. Birnbaum said the long-term goal isn't for algorithms to make official diagnoses or replace physicians, but rather to serve as supplementary tools. He added that these tools would be used only for people seeking help or information about their risk of developing a psychiatric condition, or suffering a relapse.
"Hopefully one day, we'll be able to incorporate this and other information to inform what we do, the same way you go to a doctor and you get an X-ray or a blood test to inform the diagnosis," he said. "It doesn't make the diagnosis, but it informs the doctor. That is where psychiatry is heading, and hopefully this is a step in that direction."
There have been a fair number of voices—Stephen Hawking among them—raised in warning that a super-intelligent artificial intelligence could one day turn on us and that we shouldn't be in such a hot, unquestioning hurry to develop true AI. Others say, naw, don't worry. Now a new white paper from scientists at the Center for Humans and Machines at the Max Planck Institute for Human Development presents a series of theoretical tests that confirm the threat: Due to the basic concepts underlying computing, we would be utterly unable to control a super-intelligent AI.
"We argue that total containment is, in principle, impossible, due to fundamental limits inherent to computing itself," write the paper's authors.
The white paper is published in the Journal of Artificial Intelligence Research.
Why worry?
Credit: @nt/Adobe Stock
"A super-intelligent machine that controls the world sounds like science fiction," says paper co-author Manuel Cebrian in a press release. "But there are already machines that perform certain important tasks independently without programmers fully understanding how they learned it. The question therefore arises whether this could at some point become uncontrollable and dangerous for humanity."
The lure of AI is clear. Its ability to "see" the patterns in data make it a promising agent for solving problems too complex for us to wrap our minds around. Could it cure cancer? Solve the climate crisis? The possibilities are nearly endless.
Connected to the internet, AI can grab whatever information it needs to achieve its task, and therein lies a big part of the danger. With access to every bit of human data—and responsible for its own education—who knows what lessons it would learn regardless of any ethical constraints built into its programming? Who knows what goals it would embrace and what it might do to achieve them?
Even assuming benevolence, there's danger. Suppose that an AI is confronted by an either/or choice akin to the Trolley Dilemma, maybe even on a grand scale: Might an AI decide to annihilate millions of people if it decided the remaining billions would stand a better chance of survival?
A pair of flawed options
Credit: Maxim_Kazmin/Adobe Stock
The most obvious way to keep a super intelligent AI from getting ahead of us is to limit its access to information by preventing it from connecting to the internet. The problem with limiting access to information, though, is that it would make any problem we assign the AI more difficult for it to solve. We would be weakening its problem-solving promise possibly to a point of uselessness.
The second approach that might be taken is to limit what a super-intelligent AI is capable of doing by programming into it certain boundaries. This might be akin to writer Isaac Asimov's Laws of Robotics, the first of which goes: "A robot may not injure a human being or, through inaction, allow a human being to come to harm."
Unfortunately, says the study, a series of logical tests reveal that it's impossible to create such limits. Any such a containment algorithm, it turns out, would be self-defeating.
Containment is impossible
Credit: UncleFredDesign/Adobe Stock
"If you break the problem down to basic rules from theoretical computer science, it turns out that an algorithm that would command an AI not to destroy the world could inadvertently halt its own operations. If this happened, you would not know whether the containment algorithm is still analyzing the threat, or whether it has stopped to contain the harmful AI. In effect, this makes the containment algorithm unusable."
The team investigated stacking containment algorithms, with each monitoring the behavior of the previous one, but eventually the same problem arises: The final check halts itself, rendering it unreliable.
Too smart?
The Planck researchers also concluded that a similar bit of logic makes it impossible for us to know when a self-learning computer's intelligence has come to exceed our own. Essentially, we're not smart enough to be able to develop tests for intelligence superior to ours.
"Machines take me by surprise with great frequency. This is largely because I do not do sufficient calculation to decide what to expect them to do." — Alan Turing
This means that it's entirely conceivable that an AI capable of self-learning may well quietly ascend to super-intelligence without our even knowing it — a scary reason all by itself to slow down our hurly-burley race to artificial intelligence.
In the end, we're left with a dangerous bargain to make or not make: Do we risk our safety in exchange for the possibility that AI will solve problems we can't?
One of the most devastating aspects of the COVID-19 pandemic has been unpredictability. The nation's health systems—especially those in hard-hit areas like New York City—have had to adapt to sudden surges of COVID-19 cases, all while dealing with limited resources, existing patients, and a novel virus that's still not fully understood.
But what if health systems were able to forecast COVID-19 hospitalizations two weeks before they occur? Northwell Health, the largest health care system in New York state, recently deployed a predictive tool that does just that.
Northwell Health's surveillance dashboard is able to predict COVID-19 hospitalizations by using insights from machine learning. In March, Northwell Health's Customer Insights Group developed an algorithm that's been mining data from online traffic to the Northwell.edu website, which has received more than 20 million hits since March.
The algorithm collects data through 15 different indicators, each of which reflects the online behavior of the website's visitors. For example, the tool analyzes metrics such as the length of time users spend on certain pages, searches for emergency department wait times, and specific symptoms users search for. Combined, this information translates into something like the "public mood" of the website on any given day.
Since Northwell Health began using the predictive tool in September, it's predicted COVID-19 hospitalizations with an accuracy of about 80 percent.
To understand how this mood relates to future COVID-19 cases, Northwell Health began comparing its data with a timeline of COVID-19 hospitalizations across 23 hospitals and nearly 800 outpatient facilities and in the metro New York area. This enabled the Customer Insights Group to see patterns of online activity that precede future increases or decreases in hospitalizations.
Since Northwell Health began using the predictive tool in September, it's predicted COVID-19 hospitalizations with an accuracy of about 80 percent.
"This is really the first tool that I've been exposed to that gives me a sort of guestimate of what two weeks from now may look like," said Dr. Eric Cruzen, chief medical informatics officer of Northwell's emergency medicine services and chair of the emergency department at Lenox Health Greenwich Village in Manhattan.
"Even if the data can provide an idea of whether to expect an increase, decrease, or stasis, that's valuable. Because every day we're working to estimate what tomorrow's going to bring. Any tool that's going to shed light on that is a good tool in my book."
The value of forecasting
Northwell emergency departments use the dashboard to monitor in real time.
Credit: Northwell Health
One unique benefit of forecasting COVID-19 hospitalizations is that it allows health systems to better prepare, manage and allocate resources. For example, if the tool forecasted a surge in COVID-19 hospitalizations in two weeks, Northwell Health could begin:
- Making space for an influx of patients
- Moving personal protective equipment to where it's most needed
- Strategically allocating staff during the predicted surge
- Increasing the number of tests offered to asymptomatic patients
The health-care field is increasingly using machine learning. It's already helping doctors develop personalized care plans for diabetes patients, improving cancer screening techniques, and enabling mental health professionals to better predict which patients are at elevated risk of suicide, to name a few applications.
Health systems around the world have already begun exploring how machine learning can help battle the pandemic, including better COVID-19 screening, diagnosis, contact tracing, and drug and vaccine development.
Cruzen said these kinds of tools represent a shift in how health systems can tackle a wide variety of problems.
"Health care has always used the past to predict the future, but not in this mathematical way," Cruzen said. "I think [Northwell Health's new predictive tool] really is a great first example of how we should be attacking a lot of things as we go forward."
Making machine-learning tools openly accessible
Northwell Health has made its predictive tool available for free to any health system that wishes to utilize it.
"COVID is everybody's problem, and I think developing tools that can be used to help others is sort of why people go into health care," Dr. Cruzen said. "It was really consistent with our mission."
Open collaboration is something the world's governments and health systems should be striving for during the pandemic, said Michael Dowling, Northwell Health's president and CEO.
"Whenever you develop anything and somebody else gets it, they improve it and they continue to make it better," Dowling said. "As a country, we lack data. I believe very, very strongly that we should have been and should be now working with other countries, including China, including the European Union, including England and others to figure out how to develop a health surveillance system so you can anticipate way in advance when these things are going to occur."
In all, Northwell Health has treated more than 112,000 COVID patients. During the pandemic, Dowling said he's seen an outpouring of goodwill, collaboration, and sacrifice from the community and the tens of thousands of staff who work across Northwell.
"COVID has changed our perspective on everything—and not just those of us in health care, because it has disrupted everybody's life," Dowling said. "It has demonstrated the value of community, how we help one another."
Humans love puzzles. Thanks to an evolutionary skillset that lets us piece together fragments of information necessary for survival, we've turned biological instinct into a love for games. Sometimes our affection manifests in Candy Crush; other times, in solving uncrackable ciphers.
Numerous unbreakable codes persist. The CIA awaits the brave thinker that will crack the fourth code in its Kryptos monument. The Beale ciphers may or may not reveal $60 million in hidden treasure. Composer Edward Elgar continues to laugh from beyond the grave.
Few codes stand the test of time, however. It took nearly 600 years for researchers to realize the Voynich manuscript was effectively a rip-off copy of Women's Health. The MIT time-lock puzzle was only 20 years old, yet it still took a nifty programmer three years to crack it. And then there's the Zodiac Killer.
The recent news that a 51-year-old letter from the infamous Bay Area murderer, whose story was immortalized by David Fincher, has been cracked recently made headlines. While this code will bring no peace to the families of the unknown killer's victims, the solving of this letter reminds us once again that nothing is impenetrable.How I cracked the Zodiac Killer's cipher
Zodiac Killer
After the Zodiac Killer's first cryptogram was quickly solved in 1969, he followed up with a 340-character puzzle that has baffled cryptographers ever since. Three men worked tirelessly on the letter and finally revealed the encoded message:
I HOPE YOU ARE HAVING LOTS OF FUN IN TRYING TO CATCH ME THAT WASN'T ME ON THE TV SHOW WHICH BRINGS UP A POINT ABOUT ME I AM NOT AFRAID OF THE GAS CHAMBER BECAUSE IT WILL SEND ME TO PARADICE ALL THE SOONER BECAUSE I NOW HAVE ENOUGH SLAVES TO WORK FOR ME WHERE EVERYONE ELSE HAS NOTHING WHEN THEY REACH PARADICE SO THEY ARE AFRAID OF DEATH I AM NOT AFRAID BECAUSE I KNOW THAT MY NEW LIFE WILL BE AN EASY ONE IN PARADICE DEATH
While the San Francisco branch of the FBI has acknowledged the puzzle has been solved, they're not providing any more comments considering the case remains open.
Poe's Challenge
Edgar Allan Poe's "The Gold Bug" was based on a cipher mystery, as Poe himself was fascinated with puzzles. In 1840, he offered a free subscription to Graham's Magazine to anyone who could stump him. He claims to have solved a hundred entries, ending the contest by publishing a challenging code written by W.B. Tyler—who many at the time suspected was a pseudonym.
It wasn't until 2000 that a software engineer decoded the message, which opened up, "It was early spring, warm and sultry glowed the afternoon. The very breezes seemed to share the delicious languor of universal nature..."
Given the numerous typesetting mistakes, recent researchers aren't convinced that Poe actually wrote it. The author will likely remain a mystery, but the code itself is in the books.
Copiale cipher
An entire team spanning two countries was needed to crack the 260-year-old mystery of the Copiale cipher. Unlike a few lines of prose, this 75,000-character manuscript filled 105 pages written by a group of ophthalmologists. The book was encrypted in German and relied on a complex substitution code that used symbols and letters for spaces as well as text.
Dating from the second half of the eighteenth century, the first 16 pages discuss a masonic initiation ceremony by the Oculists. The strange ritual involves initiates "reading" a blank piece of paper before being given a pair of glasses—those wily eye doctors. After their eyes are washed, the referees then pluck a single eyebrow of each recruit.
Better than college hazing, though still an odd text to keep so secretive. Then again, maybe that was the point.
Slate statue of Mathematician Alan Turing at Bletchley Park
Credit: lenscap50 / Adobe Stock
The Zimmermann Telegram
Not all codes are so playful, or strange. Some are insidious. Such is the case with the Zimmermann Telegram, a note sent from Germany to Mexico in 1917. Intended for the German ambassador to Mexico, Heinrich von Eckardt, the Germans were preparing America's southern neighbors for battle—in the name of Germany. In exchange for weapons and funding, the Mexicans would reclaim Arizona, New Mexico, and Texas upon victory.
The cipher was cracked about a month after interception by Britain's "Room 40." The text read, in part:
"We make Mexico a proposal of alliance on the following basis: make war together, make peace together, generous financial support and an understanding on our part that Mexico is to reconquer the lost territory in Texas, New Mexico, and Arizona. The settlement in detail is left to you."
Tensions between the US and Germany were already high; this message pushed America over the edge. A month later, President Wilson overruled his intention of remaining neutral and entered World War I on the side of the Allies.
The Enigma Code
One of the most famous cracks in history is certainly the Enigma Code. If the Zimmermann Telegram helped us get into World War I, the second chapter only ended in our favor thanks to Alan Turing's unforgettable machine.
The Germans were utilizing an enciphering machine to pass messages to its Axis partners. Perhaps learning from past mistakes, they changed the entire cipher system on a daily basis.
Turing responded with his own machinery: the Bombe, Lorenz, and Universal Turing Machine. Thanks to his inventions, alongside tireless efforts by British cryptologists, the Allied forces exploited procedural flaws and operator mistakes by the Germans. The Enigma Code was cracked, saving countless Allied lives and helping turn the tide of the war.
--
Stay in touch with Derek on Twitter and Facebook. His new book is "Hero's Dose: The Case For Psychedelics in Ritual and Therapy."
For a while now, the leading theory about what we're doing when we dream is that we're sorting through our experiences of the last day or so, consolidating some stuff into memories for long-term storage, and discarding the rest. That doesn't explain, though, why our dreams are so often so exquisitely weird.
A new theory proposes our brains toss in all that crazy as a way of helping us process our daily experiences, much in the way that programmers add unrelated, random-ish nonsense, or "noise," into machine learning data sets to help computers discern useful, predictive patterns in the data they're fed.
Overfitting
The goal of machine learning is to supply an algorithm with a data set, a "training set," in which patterns can be recognized and from which predictions that apply to other unseen data sets can be derived.
If machine learning learns its training set too well, it merely spits out a prediction that precisely — and uselessly — matches that data instead of underlying patterns within it that could serve as predictions likely to be true of other thus-far unseen data. In such a case, the algorithm describes what the data set is rather than what it means. This is called "overfitting."
Big Think
The value of noise
To keep machine learning from becoming too fixated on the specific data points in the set being analyzed, programmers may introduce extra, unrelated data as noise or corrupted inputs that are less self-similar than the real data being analyzed.
This noise typically has nothing to do with the project at hand. It's there, metaphorically speaking, to "distract" and even confuse the algorithm, forcing it to step back a bit to a vantage point at which patterns in the data may be more readily perceived and not drawn from the specific details within the data set.
Unfortunately, overfitting also occurs a lot in the real world as people race to draw conclusions from insufficient data points — xkcd has a fun example of how this can happen with election "facts."
(In machine learning, there's also "underfitting," where an algorithm is too simple to track enough aspects of the data set to glean its patterns.)
Credit: agsandrew/Adobe Stock
Nightly noise
There remains a lot we don't know about how much storage space our noggins contain. However, it's obvious that if the brain remembered absolutely everything we experienced in every detail, that would be an awful lot to remember. So it seems the brain consolidates experiences as we dream. To do this, it must make sense of them. It must have a system for figuring out what's important enough to remember and what's unimportant enough to forget rather than just dumping the whole thing into our long-term memory.
Performing such a wholesale dump would be an awful lot like overfitting: simply documenting what we've experienced without sorting through it to ascertain its meaning.
This is where the new theory, the Overfitting Brain Hypothesis (OBH) proposed by Erik Hoel of Tufts University, comes in. Suggesting that perhaps the brain's sleeping analysis of experiences is akin to machine learning, he proposes that the illogical narratives in dreams are the biological equivalent of the noise programmers inject into algorithms to keep them from overfitting their data. He says that this may supply just enough off-pattern nonsense to force our brains to see the forest and not the trees in our daily data, our experiences.
Our experiences, of course, are delivered to us as sensory input, so Hoel suggests that dreams are sensory-input noise, biologically-realistic noise injection with a narrative twist:
"Specifically, there is good evidence that dreams are based on the stochastic percolation of signals through the hierarchical structure of the cortex, activating the default-mode network. Note that there is growing evidence that most of these signals originate in a top-down manner, meaning that the 'corrupted inputs' will bear statistical similarities to the models and representations of the brain. In other words, they are derived from a stochastic exploration of the hierarchical structure of the brain. This leads to the kind structured hallucinations that are common during dreams."
Put plainly, our dreams are just realistic enough to engross us and carry us along, but are just different enough from our experiences —our "training set" — to effectively serve as noise.
It's an interesting theory.
Obviously, we don't know the extent to which our biological mental process actually resemble the comparatively simpler, man-made machine learning. Still, the OBH is worth thinking about, maybe at least more worth thinking about than whatever that was last night.
