"If a shell falls in a forest, does anyone hear?"
The number of jobs for humans that will remain in 2030 shrinks by the day. GPT-4 didn't just get into law school, it passed the bar in the 90th percentile. It passed the med school exams. It aced all the high school AP exams and the GREs (its highest score was in Environmental Science). It passed the introductory, certified, and advanced sommelier exams, even though it has never doffed a flute of bubbly. It has bettered grandmasters at chess and Go and designed women’s fashion.
AI has painted faux Rembrandts and Van Goghs and composed tunes in the style of Mozart and Snoop Dog. It has even assumed the voice and mannerisms of Steve Jobs to describe how Apple fared through the pandemic, in English or in Spanish.
Sound-ranging tech traces to 1915 when William Lawrence Bragg used widely-spaced microphones to automatically record distant gunfire via a galvanometer and pinpoint a firing position to within 10 meters. Modern systems are slightly better but have difficulty with some types of camouflage or if a target is mobile. This is where we need to make a point. No matter how good AI may get, it will never attain superiority to biological systems. That goes for rooftop mechanical trees over forested ecosystems as much as it does for the latest counter-battery radar.
Bats make my point. They have evolved better technology over several million years. They don’t “see” targets with a precision of meters—they pinpoint to millionths of meters.
With all those bat-hours spent on R,D&D, they do other things AI devices can only dream of. In An Immense World, Ed Yong explains:
The basic process seems straightforward. The bat's call is scattered and reflected by whatever's around it, and the animal detects and interprets the portion that rebounds. But to successfully do this, a bat must cope with many challenges. I count at least 10.
Yong’s ten challenges that bats face, and their solutions, are these:
Distance. Bats conquer this by concentrating calls into a cone and sending out sonar at 138 decibels (fortunately outside the range of human hearing).
Avoiding deafening themselves. They can hear their own screams even if we can’t. They address this by contracting muscles in their middle ears while they shout and then relaxing their ear muscles for the echo. Since they may be shouting and listening at 200 pulses per second, this is a remarkable muscular ability. Ginger Baker eat your heart out. It is further complicated by a moving target that is approaching, changing the echo gap with each pulse. Yong notes:
Bats can compensate for Doppler shifts. When closing in on a target, they produce calls that are lower than their normal resting frequency, so the upshifted echoes hit their ears at exactly the right pitch. And they do this (quite literally) on the fly, constantly tweaking their calls so that the echoes from targets ahead stay within 0.2 percent of the ideal frequency. This is a staggering feat of motor control that's almost unmatched in the animal kingdom.
Imagine that you have a mistuned piano that always produces notes three tones higher than what you're actually trying to play. If you want middle C, you'll have to press the A on its left. You'd soon get the hang of it—but imagine now that the piano's mistakes aren't systematic, and the gap between the pressed notes and desired notes changes all the time. Now you must constantly judge the size of the gap by listening to the music coming out of the janky instrument, and adjust your fingers as you play. That is what constant frequency (CF) bats are doing many times a second, with almost no errors. They can even do this for several targets simultaneously. A horseshoe bat can throw its attention between different obstacles at varying distances and perform the right Doppler compensation for each one.
Speed of evasive manuevers. While closing on prey at high-speed, bat vocal muscles contract up to 200 times a second. This is the so-called terminal buzz. They blanket their target with sonar.
Separation. When calling very quickly, a bat risks creating a jumbled stream of overlapping calls and echoes that can't be separated or interpreted. Bats, therefore, make their calls very short and space them closely so that each goes out only after the echo from the preceding one has returned. The control is so fine that even during rapid terminal buzz, there's no overlap.
Decrypting the signal. The bat's nervous system is so sensitive that it can detect differences in echo delay of just one or two-millionths of a second, which translates to a physical distance of less than a millimeter.
Translating to type. Yong explains: “… a hunting big brown bat produces a call that sweeps across a broad band of frequencies, falling over an octave or two. All of these frequencies bounce off the moth's body parts in subtly different ways, and provide the bat with disparate pieces of information. Lower frequencies tell it about large features; higher frequencies fill in finer detail. The bat's auditory system somehow analyzes all this information—the time gaps between the call and the various echoes, at each of their constituent frequencies to build a sharper and richer acoustic portrait of the moth. It knows the insect's position, but maybe also its size, shape, texture, and orientation.”
“When a sonar pulse hits an insect's flapping wing, the echo strength varies as the wing moves up and down. But at one particular moment, when the wing is exactly perpendicular to the incoming sound, an especially loud and sharp echo bounces straight back at the bat. This is called an acoustic glint. It's a dead giveaway that an insect is flying nearby… a CF bat can use glints to distinguish fluttering insects against cluttering foliage. They must be the auditory equivalent of flashes of light.”
7. Motion. The entire hunting sequence, from initial search to terminal buzz, might occur over a matter of seconds. In their attempts to close or evade, both bat and target are moving incredibly fast. Once the bat goes in for the kill, it produces the terminal buzz to claim as much information as possible as quickly as possible.
8. Camouflage. Is it a leaf-like beetle or a fluttering leaf? Using sonar, and sonar alone, a bat approaches from a sharp angle, so that echoes from the insect bounce toward it while those from the leaf bounce away.
More than 200 bat species in 60 countries are considered threatened (Critically Endangered, Endangered, or Vulnerable). More than half of the bat species in the United States are listed as endangered or in severe decline.
9. Crowds. When 20 million bats are flooding out of a cave together, how does each pick out its own echoes? Mostly, they ignore the din and fly blind. Yong says: “This explains the many historical incidents in which people barricaded the entrances to caves for safety reasons, only to later find that bats had fatally crashed into the doors.” The bats just weren’t paying attention.
10. Energy. Echolocation is mentally demanding, especially since bats do everything they do at speed. Yong tells us that bats can distinguish two grades of sandpaper whose grains differ by half a millimeter, but will also plow headlong into a cave door. The feats that they do are exhausting, but they do it on a diet of bugs. The calories or kilojoules of energy it takes to run the brain of a bat would not be enough to boot your phone. You could probably run a hundred bat brains on the energy your phone uses just launching an app.
One thing you will notice when you look at a bat is all the hair. All over its face, ears, nose, wings, tail, belly and back are various kinds and sizes of hair. Every strand is an antenna. The follicles and the nerves aid bats in flight control, detecting air velocity, wind direction, vortices, and potential stall conditions. Hair also feeds their brains more information about obstacles or prey. Shave a bat and it collides with walls and trees. I wrote more about the sensory perception of hair in 2012.
AI-endangered economic sectors
Will general AI become creative in the way musicians like Brian Wilson, Paul McCartney or Paul Simon claim their greatest works came fully formed from the deep subconscious, or writers like Shakespeare, F. Scott Fitzgerald and Stephen King pull up ideas no one seems to have imagined before? What about non-linear relationships? Will AI enter into a flow state to tap the collective unconscious?
Will AI learn from bats? It probably is doing that already in some defense industry lab—design by biomimicry. Bats have one insurmountable advantage over AI, however, just as real trees have an advantage over artificial ones. We saw this in how astronaut Dave regained control of the HAL-9000 computer in Stanley Kubrick’s 2001: A Space Odyssey.
AI requires external power. Vast power. If you are worried about AI taking over the world and killing us all, don’t. For now, we can just unplug it.
That may be what the bats were trying to do in Wuhan, China. Unplug us. They may finally be paying attention.
