Worms Aren't So Dumb

Chances are, you’ve read about AI lately.  Maybe you’ve actually even tried DALL-E or ChatGPT, maybe even GPT-4.  Perhaps you can use the term Large Language Model (LLM) with some degree of confidence.  But chances are also good that you haven’t heard of “liquid neural networks,” and don’t get the worm reference above.   

Credit: Jose-Luis Olivares, MIT

That’s the thing about artificial intelligence: it’s evolving faster than we are. Whatever you think you know is already probably out-of-date.

Liquid neural networks were first introduced in 2020.  The authors wrote: “We introduce a new class of time-continuous recurrent neural network models.” They based the networks on the brain of a tiny roundworm, Caenorhabditis elegans.  The goal was networks that were more adaptable, that could change “on the fly” and would adapt to unfamiliar circumstances.

Researchers at MIT’s CSAIL have shown some significant progress.  A new paper in Science Robotics discussed how they created “robust flight navigation agents” using liquid neural networks to autonomously pilot drones. They claim that these networks are “causal and adapt to changing conditions,” and that their “experiments showed that this level of robustness in decision-making is exclusive to liquid networks.”  

An MIT press release notes: “deep learning systems struggle with capturing causality, frequently over-fitting their training data and failing to adapt to new environments or changing conditions…Unlike traditional neural networks that only learn during the training phase, the liquid neural net’s parameters can change over time, making them not only interpretable, but more resilient to unexpected or noisy data.” 

“We wanted to model the dynamics of neurons, how they perform, how they release information, one neuron to another, Ramin Hasani, a research affiliate at MIT and one of the co-authors, told Popular Science.

Essentially, they trained the neural network to pilot the drone to find a red camping chair, then moved the chair to a variety of environments, in different lightening conditions, at different times of year, and at different distances to see if the drone could still find the chair. “The primary conceptual motivation of our work,” the authors wrote, “was not causality in the abstract; it was instead task understanding, that is, to evaluate whether a neural model understands the task given from high-dimensional unlabeled offline data.”

Credit: Chahine, et. alia

Daniela Rus, CSAIL director and one of the co-authors, said: “Our experiments demonstrate that we can effectively teach a drone to locate an object in a forest during summer, and then deploy the model in winter, with vastly different surroundings, or even in urban settings, with varied tasks such as seeking and following.” 

Their video:

Essentially, Dr. Hasani says, “they can generalize to situations that they have never seen.”  The liquid neural nets can also "dynamically capture the true cause-and-effect of their given task," the authors wrote.  This is "the key to liquid networks’ robust performance under distribution shifts.”

The key advantage of liquid neural networks is their adaptability; the neurons behave more like the worm’s (or the neurons of other living creatures) would, responding to real world circumstances in real time.  “They’re able to change their underlying equations based on the input they observe,” Dr. Rus told Quanta Magazine. 

Dr. Rus further noted: “We are thrilled by the immense potential of our learning-based control approach for robots, as it lays the groundwork for solving problems that arise when training in one environment and deploying in a completely distinct environment without additional training…These flexible algorithms could one day aid in decision-making based on data streams that change over time, such as medical diagnosis and autonomous driving applications.”

Sriram Sankaranarayanan, a computer scientist at the University of Colorado, was impressed, telling Quanta Magazine: “The main contribution here is that stability and other nice properties are baked into these systems by their sheer structure…They are complex enough to allow interesting things to happen, but not so complex as to lead to chaotic behavior.”

Alessio Lomuscio, professor of AI safety in the Department of Computing at Imperial College London, was also impressed, telling MIT:

Robust learning and performance in out-of-distribution tasks and scenarios are some of the key problems that machine learning and autonomous robotic systems have to conquer to make further inroads in society-critical applications. In this context, the performance of liquid neural networks, a novel brain-inspired paradigm developed by the authors at MIT, reported in this study is remarkable. If these results are confirmed in other experiments, the paradigm here developed will contribute to making AI and robotic systems more reliable, robust, and efficient.

It's easy enough to imagine lots of drone applications where these could prove important, with autonomous driving another logical use. But the MIT team is looking more broadly. “The results in this paper open the door to the possibility of certifying machine learning solutions for safety critical systems,” Dr. Rus says. With all the discussion about the importance of ensuring that AI was giving valid answers in healthcare uses, as noted above, she specifically mentioned medical diagnosis decision making as one for liquid neural networks.

“Everything that we do as a robotics and machine learning lab is [for] all-around safety and deployment of AI in a safe and ethical way in our society, and we really want to stick to this mission and vision that we have,” Dr. Hasani says.  We should hope that other AI labs feel the same.

Healthcare, like most parts of our economy, is going to increasingly use and even rely on AI. We’re going to need AI that not only gives us accurate answers but also can adapt to quickly changing conditions, rather than pre-set data models.  I don’t know if it’s going to be based on liquid neural networks or something else, but we’re going to want not just adaptability but also safety and ethics baked in.

----------------

Last month I wrote about Organoid Intelligence (OI), which intends to gets to AI using structures that world more like our brains. Now liquid neural networks based on worms’ brains. It’s intriguing to me that after several decades of working on, and perhaps for, our silicon overlords, we’re starting to move to more biological approaches.

As Sayan Mitra, a computer scientist at the University of Illinois, Urbana-Champaign, told Quanta Magazine: “In a way, it’s kind of poetic, showing that this research may be coming full circle. Neural networks are developing to the point that the very ideas we’ve drawn from nature may soon help us understand nature better.” 

I Have Some Silly Questions

Last year I used some of Alfred North Whitehead’s pithy quotations to talk about healthcare, starting with the provocative “It is the business of the future to be dangerous.”  I want to revisit another of his quotations that I’d like to spend more time on: “The silly question is the first intimation of some totally new development.”

Credit: Getty Images

I can’t promise that I even have intimations of what the totally new developments are going to be, but if any industry lends itself to asking “silly” questions about it, it is healthcare.  Hopefully I can at least spark some thought and discussion. 

In no particular order:

Why do we prefer to spend money on care when people are no longer healthy than we do on keeping them healthy?

The U.S. healthcare system well known for being exorbitantly expensive while delivering rather mediocre results.  Everyone laments it but we keep throwing more money into the system that is producing these results.

We’d be smarter to invest in upstream spending.  Like making sure people get enough to eat, with foods that are good for us.  We’d rather spend money on diabetes or obesity drugs rather than addressing the root causes of each disease.  Or like making sure the water we drink, the air we breathe, the things we eat, aren’t polluted (how many toxins or microplastics have you ingested today?).  Not to mention reducing poverty, improving education, or fixing social media.

We know the kinds of things we should do, we say we want to do them, but we lack the political will to achieve them and the infrastructure to ensure them.  So we end up paying for our neglect through our ever-more expensive healthcare system.  That’s silly.

Credit: Gracia Lam/The Commonwealth Fund

Why is everything in healthcare so expensive?

People used to talk about how the military had $600 hammers and $640 toilet seats, but healthcare just says “hold my beer” to all those examples.  It’s been well established that U.S. healthcare’s spending problem is not oversupply or overutilization – although there is both – but “It’s the prices, stupid.”  Whether it is tests, procedures, hospital stays, prescription drugs, or anything else, in U.S. healthcare things just cost more. 

Yet no one in healthcare thinks their prices are too high, and no one admits they’re making too much money.  We’d hoped making patients pay more of their bills, or making prices more transparent, would make them more price conscious, but that hasn’t happened.  We’ve even recently found that those big health insurers who supposedly are tough negotiators often don’t get better prices than people paying directly. 

Prices are high because no one cares enough to force them to be lower. That’s silly.

How come our physicians are becoming more specialized even though we’re realizing how interdependent everything in our body is?

Primary care doctors make up at most only a third of U.S. physicians, a percentage that is both declining and well below the comparable percentage in other developed countries.  Moreover, the physicians who are specialists are increasingly becoming sub-specialists.  That’s great if you have only one very specific health issue, but if that issue is only one of many, or if that issue is due to or has created other health issues – as is likely – then it is not so good.  It means lots of referrals, many doctors, and sometimes conflicting recommendations, Primary care doctors were supposed to coordinate everything, but, as I said, there aren’t that many of them, and they don’t have the time anyway.

If we want more holistic care, we’re training physicians wrong, we’re incenting them wrong financially, and we’re not getting anyone with the big picture of our health.  That’s silly.

Credit: Yale Medicine

How come there is no one in the health care system whose job it is to keep us healthy, and who is rewarded accordingly?

Everyone in healthcare talks about wanting patients to be healthy, and many of them pay at least lip service to reminding you about doing things that might help, but let’s be clear: they get paid when you get sick (or, at least, use services).  It’s not a health system in any sense; it is a health care at best, and, more accurately, a medical care system.

If you have one, a primary care doctor might claim to fill that space, but, admit it, they don’t really know what your health habits are and have little ability to influence them.  Arguably, there are a great many health behaviors that are outside physicians’ training and expertise anyway.  The health care system thinks about people only when they become patients, and primarily relies on reactive, medical interventions to address health issues. 

Hypothetically, one can imagine a health care system in which you pay for the periods during which you are healthy, and don’t pay when you aren’t.  Instead, we use a system in which the more you use it, the more you pay.  That’s silly.

Since we’ve discovered that our microbiome outnumbers “us” – in terms of cells, DNA, etc. – and has direct impacts on “our” health, why does our health care still focus on “us” and not on “we”?

It was a huge step forward for medicine, and our health, when we discovered bacteria, viruses, etc., but we quickly decided they were the enemy and declared war on them.  Yes, advances like penicillin have saved countless lives, but, as with many advances, we didn’t fully understand the consequences. 

To the extent we think about our microbiome, it’s usually in conjunction with our gut, and mostly only in relation to our gut health, but research is showing that the gut microbiome impacts other areas of the body (such as our brain), and that our microbiome has impacts on almost all aspects of our health.  Our vaunted theories of disease badly need a 21^st century, microbiome-inclusive update. Like it or not, if it isn’t not healthy, we’re not healthy.

We indiscriminately kill off “good” microbes along with the “bad” ones, rather than trying to optimize the balance.  That’s silly.

Credit: NIH

 --------------

Our health care system has a “The Emperor Has No Clothes” problem. Many of us realize that it is badly flawed, perhaps even irredeemable, but it is the system we have, and we figure that it would take too much to fundamentally change it.  Well, that’s silly.

So I’ll keep trying to ask silly questions about healthcare and hope that others can get intimations of something totally new. 

Implementation May Be a Science, But, Alas, Medicine is Still an Art

I’ve been working in healthcare for over forty (!) years now, in one form or another, but it wasn’t until this past week that I heard of implementation science.  Which, in a way, is sort of the problem healthcare has. 

The big gap. Credit: NIHR ARC South London

Granted, I’m not a doctor or other clinician, but everyone working in healthcare should be aware of, and thinking a lot about, “the scientific study of methods to promote the systematic uptake of research findings and other EBPs into routine practice, and, hence, to improve the quality and effectiveness of health services” (Bauer, et. al). 

It took a JAMA article, by Rita Rubin, to alert me to this intriguing science: It Takes an Average of 17 Years for Evidence to Change Practice—the Burgeoning Field of Implementation Science Seeks to Speed Things Up.

It turns out that implementation science is nothing new. There has been a journal devoted to it (cleverly named Implementation Science) since 2006, along with the relatively newer Implementation Science Communications. Both focus on articles that illustrate “methods to promote the uptake of research findings into routine healthcare in clinical, organizational, or policy contexts.” 

Brian Mittman, Ph.D., has stated that the aims of implementation science are:

• “To generate reliable strategies for improving health-related processes and outcomes and to facilitate the widespread adoption of these strategies.
• To produce insights and generalizable knowledge regarding implementation processes, barriers, facilitators, and strategies.
• To develop, test, and refine implementation theories and hypotheses, methods, and measures.”

Dr. Mittman distinguished it from quality improvement largely because QI focuses primarily on local problems, whereas “the goal of implementation science is to develop generalizable knowledge.” 

Ms. Rubin’s headline highlights the problem healthcare has: it can take an alarmingly long time for empirical research findings to be incorporated into standard medical practice.  There is some dispute about whether 17 years is actually true or not, but it is widely accepted that, whatever the actual number is, it is much too long.  Even then, Ms. Rubin reminds us, it is further estimated that only 1 in 5 interventions make it to routine clinical care.  

She quotes University of Washington gastroenterologist Rachel Issaka, MD, MAS: “implementation science is really trying to close that gap between what we know and what we do.”  Or, rather, between what is known by some and what most do. “The hope of implementation science is that we can synthesize what works for whom and for where and for what disease and close that 17-year gap,” Nathalie Moise, MD, MS, director of implementation science research at Columbia University told JAMA.

It is worth noting that implementation science focuses both on getting clinicians to start doing newly proven treatments as well as to stop doing longstanding treatments that have subsequently been shown to be of little or no value (“deimplementation”). 

There are implementation science departments or programs at Brown, Duke, Johns Hopkins, Northwestern, Penn, UCSF, UNC, University of Michigan, University of Washington, and Wake Forest, to name a few. Some are in the school of medicine, some in the school of public health. 

With such widespread training in the field, you’d think we’d be doing better at closing that gap – or, as Ms. Rubin labels it, that “chasm” – between what we should do and what we do.  But here we are still, and, as Ms. Rubin points out, COVID proved the point. 

“COVID-19 has shown the world that ‘knowing what to do’ does not ensure ‘doing what we know,’” wrote implementation science pioneer Enola Proctor, PhD, a professor emerita of social work, and infectious disease specialist Elvin Geng, MD, MPH, director of the Center for Dissemination and Implementation at the Institute for Public Health, both at Washington University in St Louis, in a 2021 Science editorial.

 Few would argue that clinicians are actively ignoring best practices. It’s more about how they were trained, how others around them practice, what they’re used to/comfortable with, and hugely compounded by the sheer mass of medical knowledge.  Medical knowledge is estimated to double every couple of months, and that half life is getting shorter and shorter; it was estimated at 2 years only five years ago.  No one -- no human anyway -- can keep up.

Other limitations are that studies may not have had diverse enough study populations, or that they are socio-economic barriers to the desired care.  Ms. Rubin cities the simple lack of a ride post-colonoscopy as a reason that some patients decline getting them. “I do think that White, high socioeconomic clinicians just have no clue that there are people out there who lack transportation options,” Dr. Issaka notes. That’s only one of a million – a billion -- blind spots that our health care system has about the people using it.

One has to wonder about what kind of industry healthcare is that it needs a science to study how to implement practices that are proven to be more effective for its customers. Most other industries focus on this as a matter of course, as a matter of survival, but not healthcare.

Much of this, I fear, is our historical view that physicians are as much, if not more, “artists” as scientists. We defer to their judgement. We lack the mechanisms to ensure that they’re practicing similarly to other physicians in the community, much less in other communities, and still much less to best practices/most recent evidence. That’s a big reason why healthcare needs implementation science, and why it has been slow going making it actually succeed.

Big Data and AI give us the tools to change this.

Using Big Data, we have the ability to collect and analyze what happens to patients. We can know what treatments physicians are ordering, and if they are in conformance with best practices. Best of all, it should allow us to evaluate effectiveness on much bigger populations, in more widely diverse situations, in much faster time frames.

Using AI, individual clinicians will be able to better keep up with existing medical knowledge. It’s an impossible task now, but one that AI is already starting to demonstrate. Most current AI are trained on fixed data sets, which can’t include the most current research, but those data sets are still much better than a clinician’s memory, and in the near future AI should be able find current findings in real time.

I love that there is implementation science, and I wish its practitioners great success, but I long for the day when healthcare has its principles baked into its everyday practice.       

I Have No Mouth, Yet Still I Scream

In light of the recent open letter from AI leaders for a moratorium on AI development, I’m declaring a temporary moratorium on writing about it too, although I doubt either one will last long (and this week’s title is, if you hadn’t noticed, an homage to Harlan Ellison’s classic dystopian AI short story).  Instead, this week I want to write about plants. Specifically, the new research that suggests that plants can, in their own way, scream.

Bear with me.

Credit: Robert Krulwich/NPR

To be fair, the researchers don’t use the word “scream;” they talk about “ultrasonic airborne sounds,” but just about every account of the research I saw used the more provocative term.  It has long been known that plants are far from passive, responding to stimuli in their environment with changes in color, smell, and shape, but these researchers “show that stressed plants emit airborne sounds that can be recorded from a distance and classified.”  Moreover, they posit: “These informative sounds may also be detectable by other organisms.”  

It should make you wonder what your houseplant is saying about you when you forget to water it or get a cat. 

They basically tortured – what else would you call it? – plants with a variety of stresses, then used machine learning (damn – I guess I am writing about AI after all) to classify, with up to 70% accuracy, different categories of responses, such as too much water versus too little.  Even plants that have been cut, and thus are dying, can still produce the sounds, at least for short periods.  They speculate that other plants, as well as insects, may be able to “hear” and respond to the sounds.

The ultrasonic sounds are believed to be produced through a process known as cavitation, which is a well-known process during which pressure variations in a liquid create small bubbles that collapse and generate shock waves.  The specific mechanism for this hasn’t been identified.

Here is what it sounds like:

The research mainly used tomato and tobacco plants, but also found that other plants, including corn, wheat, grape, and cactus, also emitted the sounds.  “We can separate between sounds emitted by tomato and tobacco, between tomato and cacti, and also between cut tomato and dry tomato a little bit dry tomato and very dry tomato,” lead researcher Lilach Hadany, a professor at Tel Aviv University, told Business Insider.

“When these plants are in good shape, they produce less than one sound per hour, but when stressed they emit many more, sometimes 30 to 50 per hour,” said Professor Hadany.  Her team had previously shown that plants can “hear,” such as when bees buzzing nearby cause them to produce more nectar.

“These findings can alter the way we think about the plant kingdom, which has been considered to be almost silent until now," the authors write.  “Our results, demonstrating the ability to distinguish between drought-stressed and control plants based on plant airborne sounds, open an avenue of research in the field of precision agriculture.”

Instead of blithely fertilizing and watering them, on our schedules, plants may be telling us exactly what they need, when.

It gets even more interesting. "Even in a quiet field, there are actually sounds that we don't hear, and those sounds carry information. There are animals that can hear these sounds, so there is the possibility that a lot of acoustic interaction is occurring," explains Professor Hadany.  “So now that we know that plants do emit sounds, the next question is—‘who might be listening?’ We are currently investigating the responses of other organisms, both animals and plants, to these sounds, and we’re also exploring our ability to identify and interpret the sounds in completely natural environments.”

Credit: Khait, et. alia.

Not everyone is convinced that communication is happening.  “Lots of sounds in the world are generated that are not ‘intentional’ signals, but nonetheless can be heard and used by other organisms for their own benefits. So, the concept of communication is indeed a challenge … does it need to be bi-directional to work and be considered as such?” Daniel Robert, a professor of bionanoscience at the University of Bristol’s School of Biological Sciences, told CNN.  He was not involved in the research.

As to whether the sounds suggest that plants have “feelings” as we might think of them, “I think we are not there yet," Professor Hadany admits. "We cannot say the plant feels stress and therefore makes sounds. It might be that the sounds are made completely passively, like a physical process.”

---------------

This all reminds me over the furor a few years ago about the “Wood Wide Web,” a hypothesis that trees communicated through their roots via a network of fungi, although this remains controversial.  The point, though, is that there is a lot more communication going on in nature than we realize. We used to think we were the only animal that communicated, the only social animal, the only tool-using animal, and all those have been debunked. Now, if plants can scream, the lines between animal and plant are getting less well-defined.

Similarly, the line between “us” and our microbiome is getting very fuzzy. It’s long been known that not only do our microbiome cells outnumber “our” cells but also their DNA vastly outnumbers ours.  Who is really “us”? 

Moreover, our microbiome is actively communicating with us, not only in the gut (where the largest numbers are) but also with the brain and other organs.  That communication at least influences our health, such as with MS or depression.  It turns out that cancer cells have their own microbiome (and mycobiome).  More connections will be discovered, such as the effect on inflammation, which may underlay heart disease and autoimmune disorders.

Unfortunately, we know more about what plants are communicating with the world than what our microbiome is communicating with us. 

It all reinforces my belief that the 21^st century is going to be the century of biology – whether it is computing, industry, or medicine. Yet we’re still dousing everything in antibiotics and wrecking havoc on our microbiome, with unknown (but probably terrible) consequences. 

So perhaps we should be doing a better job of listening to plants, and figuring out what else in nature we should be paying better attention to. Our health may depend on it.