Once a week.
Subscribe to our weekly newsletter.
A new AI passes the smell test almost 100% of the time
Our remarkable olfactory senses are modeled in a new research chip.
- Two researchers have created an algorithm that can accurately identify 10 different smells.
- The AI algorithm runs on an Intel chip that has 130,000 silicon "neurons."
- The natural mammalian olfactory bulb grows new neurons even in adults.
Our noses are busy beasts. At any given moment, multiple smells are competing for our attention, and somehow the brain can tell when it's smelling an orange even against a backdrop of other scents, say perfume or soap. The brain's olfactory bulb has hundreds of receptors tracking odors all the time, and yet somehow keeps everything straight. A scientist at Cornell University working with a researcher at Intel has just created an AI algorithm trained to recognize 10 scents by mimicking the mammalian olfactory bulb (MOB). Give the algorithm a computer chip to run on and it can learn to identify new odors.
Simulating the brain
Image source: GiroScience/Shutterstock
Thomas Cleland, senior author of "Rapid Learning and Robust Recall in a Neuromorphic Olfactory Circuit" published in Nature Machine Intelligence says, "This is a result of over a decade of studying olfactory bulb circuitry in rodents and trying to figure out essentially how it works, with an eye towards things we know animals can do that our machines can't."
"We now know enough to make this work," says Cleland speaking with Cornell Chronicle. "We've built this computational model based on this circuitry, guided heavily by things we know about the biological systems' connectivity and dynamics. Then we say, if this were so, this would work. And the interesting part is that it does work."
Cleland's co-author, Intel intern Nabil Imam, has the AI running on an Intel research chip called Loihi. The chip is something of a wonder all by itself. It's designed to perform neuromorphic computing inspired by the brain, sporting some 130,000 silicon "neurons." Like our own neurons, each can fire independently of the others, sending pulsed signals to its neighbors, changing their electrical states. In addition, Loihi can accept inputs from a variety of physical sensors, such as the metal-oxide chemical sensors that Cleland and Imam used, allowing the entire structure to simulate the natural learning process that occurs when the mammalian brain is fed various sensory inputs.
What the AI knows
In the MOB, there are a couple of important neuron types: mitral cells that switch on to signify a smell has been detected but don't identify it, and granule cells that learn to become specialized and pinpoint chemicals in the smell. This is the process Cleland and Imam have programmed Loihi to imitate when odors are presented to the system's sensors.
The researchers have trained the algorithm to identify 10 chemical smells: acetone, acetaldehyde, ammonia, butanol, ethylene, methane, methanol, carbon monoxide, benzene, and toluene. The idea is that in the presence of an odor, Loihi verifies the scent's presence and patterns of firing neurons identify exactly which smell it is.
Over the course of five cycles of exposure to each odor, the AI was eventually able to associate specific neural firing patterns with specific smells. The AI correctly identified eight of the smells 100 percent of the time, and the remaining two with 90 percent accuracy.
Additional simulations were run in which the target odor was masked 80 percent by other smells, the way they'd often be in the real word. In these tests, accuracy dropped down to just 30 percent, though that's still pretty impressive for such a "young" AI. Says Cleland, "The pattern of the signal has been substantially destroyed, and yet the system is able to recover it."
The sweet smell of success
Image source: Patrick J. Lynch, wikimedia
As AI and machine learning do, Cleland's and Imam's algorithm would be expected to get better with more training. Machine smell-detection for use in bomb-sniffing, for example, is clearly not just around the corner. Still, the research offers new insights about learning.
Due to the normally finite number of neurons available for storage of new memories in an adult brain, and the fact that "when you learn something, it permanently differentiates neurons," as Cleland says, storing new information means changing the information already encoded in neurons. However, the mammalian olfactory bulb is able to continually learn smells — hundreds or thousands of them —without losing knowledge of already-learned odors. It's one of the few areas of the brain that can keep creating new neurons through adulthood.
The researchers' experience with Loihi underscores the value of the olfactory bulb's unusual neuronal expandability. "The computational model turns into a biological hypothesis for why adult neurogenesis is important," Cleland says, "because it does this thing that otherwise would make the system not work. So in that sense, the model is feeding back into biology. And in this other sense, it's the basis for a set of devices for artificial olfactory systems that can be constructed commercially."
- Scientists link sense of smell and sense of direction - Big Think ›
- A new perfume can help you smell like an astronaut - Big Think ›
- How the brain processes our sense of smell - Big Think ›
- Smells form stronger connection to your memories - Big Think ›
Why mega-eruptions like the ones that covered North America in ash are the least of your worries.
- The supervolcano under Yellowstone produced three massive eruptions over the past few million years.
- Each eruption covered much of what is now the western United States in an ash layer several feet deep.
- The last eruption was 640,000 years ago, but that doesn't mean the next eruption is overdue.
The end of the world as we know it
Panoramic view of Yellowstone National Park
Image: Heinrich Berann for the National Park Service – public domain
Of the many freak ways to shuffle off this mortal coil – lightning strikes, shark bites, falling pianos – here's one you can safely scratch off your worry list: an outbreak of the Yellowstone supervolcano.
As the map below shows, previous eruptions at Yellowstone were so massive that the ash fall covered most of what is now the western United States. A similar event today would not only claim countless lives directly, but also create enough subsidiary disruption to kill off global civilisation as we know it. A relatively recent eruption of the Toba supervolcano in Indonesia may have come close to killing off the human species (see further below).
However, just because a scenario is grim does not mean that it is likely (insert topical political joke here). In this case, the doom mongers claiming an eruption is 'overdue' are wrong. Yellowstone is not a library book or an oil change. Just because the previous mega-eruption happened long ago doesn't mean the next one is imminent.
Ash beds of North America
Ash beds deposited by major volcanic eruptions in North America.
Image: USGS – public domain
This map shows the location of the Yellowstone plateau and the ash beds deposited by its three most recent major outbreaks, plus two other eruptions – one similarly massive, the other the most recent one in North America.
The Huckleberry Ridge eruption occurred 2.1 million years ago. It ejected 2,450 km3 (588 cubic miles) of material, making it the largest known eruption in Yellowstone's history and in fact the largest eruption in North America in the past few million years.
This is the oldest of the three most recent caldera-forming eruptions of the Yellowstone hotspot. It created the Island Park Caldera, which lies partially in Yellowstone National Park, Wyoming and westward into Idaho. Ash from this eruption covered an area from southern California to North Dakota, and southern Idaho to northern Texas.
About 1.3 million years ago, the Mesa Falls eruption ejected 280 km3 (67 cubic miles) of material and created the Henry's Fork Caldera, located in Idaho, west of Yellowstone.
It was the smallest of the three major Yellowstone eruptions, both in terms of material ejected and area covered: 'only' most of present-day Wyoming, Colorado, Kansas and Nebraska, and about half of South Dakota.
The Lava Creek eruption was the most recent major eruption of Yellowstone: about 640,000 years ago. It was the second-largest eruption in North America in the past few million years, creating the Yellowstone Caldera.
It ejected only about 1,000 km3 (240 cubic miles) of material, i.e. less than half of the Huckleberry Ridge eruption. However, its debris is spread out over a significantly wider area: basically, Huckleberry Ridge plus larger slices of both Canada and Mexico, plus most of Texas, Louisiana, Arkansas, and Missouri.
This eruption occurred about 760,000 years ago. It was centered on southern California, where it created the Long Valley Caldera, and spewed out 580 km3 (139 cubic miles) of material. This makes it North America's third-largest eruption of the past few million years.
The material ejected by this eruption is known as the Bishop ash bed, and covers the central and western parts of the Lava Creek ash bed.
Mount St Helens
The eruption of Mount St Helens in 1980 was the deadliest and most destructive volcanic event in U.S. history: it created a mile-wide crater, killed 57 people and created economic damage in the neighborhood of $1 billion.
Yet by Yellowstone standards, it was tiny: Mount St Helens only ejected 0.25 km3 (0.06 cubic miles) of material, most of the ash settling in a relatively narrow band across Washington State and Idaho. By comparison, the Lava Creek eruption left a large swathe of North America in up to two metres of debris.
The difference between quakes and faults
The volume of dense rock equivalent (DRE) ejected by the Huckleberry Ridge event dwarfs all other North American eruptions. It is itself overshadowed by the DRE ejected at the most recent eruption at Toba (present-day Indonesia). This was one of the largest known eruptions ever and a relatively recent one: only 75,000 years ago. It is thought to have caused a global volcanic winter which lasted up to a decade and may be responsible for the bottleneck in human evolution: around that time, the total human population suddenly and drastically plummeted to between 1,000 and 10,000 breeding pairs.
Image: USGS – public domain
So, what are the chances of something that massive happening anytime soon? The aforementioned mongers of doom often claim that major eruptions occur at intervals of 600,000 years and point out that the last one was 640,000 years ago. Except that (a) the first interval was about 200,000 years longer, (b) two intervals is not a lot to base a prediction on, and (c) those intervals don't really mean anything anyway. Not in the case of volcanic eruptions, at least.
Earthquakes can be 'overdue' because the stress on fault lines is built up consistently over long periods, which means quakes can be predicted with a relative degree of accuracy. But this is not how volcanoes behave. They do not accumulate magma at constant rates. And the subterranean pressure that causes the magma to erupt does not follow a schedule.
What's more, previous super-eruptions do not necessarily imply future ones. Scientists are not convinced that there ever will be another big eruption at Yellowstone. Smaller eruptions, however, are much likelier. Since the Lava Creek eruption, there have been about 30 smaller outbreaks at Yellowstone, the last lava flow being about 70,000 years ago.
As for the immediate future (give or take a century): the magma chamber beneath Yellowstone is only 5 percent to 15 percent molten. Most scientists agree that is as un-alarming as it sounds. And that its statistically more relevant to worry about death by lightning, shark, or piano.
Strange Maps #1041
Got a strange map? Let me know at email@example.com.
The potential of CRISPR technology is incredible, but the threats are too serious to ignore.
- CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) is a revolutionary technology that gives scientists the ability to alter DNA. On the one hand, this tool could mean the elimination of certain diseases. On the other, there are concerns (both ethical and practical) about its misuse and the yet-unknown consequences of such experimentation.
- "The technique could be misused in horrible ways," says counter-terrorism expert Richard A. Clarke. Clarke lists biological weapons as one of the potential threats, "Threats for which we don't have any known antidote." CRISPR co-inventor, biochemist Jennifer Doudna, echos the concern, recounting a nightmare involving the technology, eugenics, and a meeting with Adolf Hitler.
- Should this kind of tool even exist? Do the positives outweigh the potential dangers? How could something like this ever be regulated, and should it be? These questions and more are considered by Doudna, Clarke, evolutionary biologist Richard Dawkins, psychologist Steven Pinker, and physician Siddhartha Mukherjee.
Measuring a person's movements and poses, smart clothes could be used for athletic training, rehabilitation, or health-monitoring.
In recent years there have been exciting breakthroughs in wearable technologies, like smartwatches that can monitor your breathing and blood oxygen levels.
But what about a wearable that can detect how you move as you do a physical activity or play a sport, and could potentially even offer feedback on how to improve your technique?
And, as a major bonus, what if the wearable were something you'd actually already be wearing, like a shirt of a pair of socks?
That's the idea behind a new set of MIT-designed clothing that use special fibers to sense a person's movement via touch. Among other things, the researchers showed that their clothes can actually determine things like if someone is sitting, walking, or doing particular poses.
The group from MIT's Computer Science and Artificial Intelligence Lab (CSAIL) says that their clothes could be used for athletic training and rehabilitation. With patients' permission, they could even help passively monitor the health of residents in assisted-care facilities and determine if, for example, someone has fallen or is unconscious.
The researchers have developed a range of prototypes, from socks and gloves to a full vest. The team's "tactile electronics" use a mix of more typical textile fibers alongside a small amount of custom-made functional fibers that sense pressure from the person wearing the garment.
According to CSAIL graduate student Yiyue Luo, a key advantage of the team's design is that, unlike many existing wearable electronics, theirs can be incorporated into traditional large-scale clothing production. The machine-knitted tactile textiles are soft, stretchable, breathable, and can take a wide range of forms.
"Traditionally it's been hard to develop a mass-production wearable that provides high-accuracy data across a large number of sensors," says Luo, lead author on a new paper about the project that is appearing in this month's edition of Nature Electronics. "When you manufacture lots of sensor arrays, some of them will not work and some of them will work worse than others, so we developed a self-correcting mechanism that uses a self-supervised machine learning algorithm to recognize and adjust when certain sensors in the design are off-base."
The team's clothes have a range of capabilities. Their socks predict motion by looking at how different sequences of tactile footprints correlate to different poses as the user transitions from one pose to another. The full-sized vest can also detect the wearers' pose, activity, and the texture of the contacted surfaces.
The authors imagine a coach using the sensor to analyze people's postures and give suggestions on improvement. It could also be used by an experienced athlete to record their posture so that beginners can learn from them. In the long term, they even imagine that robots could be trained to learn how to do different activities using data from the wearables.
"Imagine robots that are no longer tactilely blind, and that have 'skins' that can provide tactile sensing just like we have as humans," says corresponding author Wan Shou, a postdoc at CSAIL. "Clothing with high-resolution tactile sensing opens up a lot of exciting new application areas for researchers to explore in the years to come."
The paper was co-written by MIT professors Antonio Torralba, Wojciech Matusik, and Tomás Palacios, alongside PhD students Yunzhu Li, Pratyusha Sharma, and Beichen Li; postdoc Kui Wu; and research engineer Michael Foshey.
The work was partially funded by Toyota Research Institute.