Over the holiday weekend, I chatted with my sister-in-law about a study her and her dog participate in on providing communication tools to canines. Maybe you saw this segment on CBS Sunday Morning or some other reporting on this work. The dogs are given buttons, each of which plays a recording of a spoken word. We know dogs have some capacity to understand spoken language, since they can be trained to respond to commands. The word buttons close the loop, allowing the dogs to use words as well. While the sophistication of the speech may be disappointing for anyone expecting the translator collars from Up, this setup actually requires more intelligence on the part of the dogs since they have to do the translating themselves–possibly with some deciphering by the humans as well. The idea that the dog can press a button that says “dinner” when he’s hungry or “toy” when he wants to play seems pretty straightforward. Combining “downstairs” and “broccoli” to request a vegetable-flavored treat kept downstairs instead of “upstairs broccoli” (florets kept upstairs) is more intriguing.
Also this weekend, I read a discussion of data mining algorithms for advertising and why your phone doesn’t need to be listening to your speech all the time to serve uncannily specific ads. Major ad providers have repeatedly denied continuous speech monitoring to serve ads, and laboratory tests and code analysis have backed up those assertions. Yet nearly everyone I know with a smartphone has a story about the time they were served ads for something they had spoken about in proximity to their phone but had never typed into a digital device. I’ll confess to being drawn into some of these accounts, unable to provide a satisfactory alternative explanation. But the reality is likely just a collective failure of imagination, an inability to conceive of how a machine could learn something about us that we hadn’t explicitly told it. We underappreciate how much can be inferred about us from our usage of various apps, from our movements in physical space, from which phones our phones spend time near, and by aggregating knowledge from millions of users across dozens of platforms. The sum total of how we are tracked is probably more invasive and unsettling than mere listening, yet that is the one invasion we remain most wary of.
Finally, recent days have seen a resurgence of interest in the possibility that SARS-CoV-2, the virus that causes COVID-19, first infected a human in a laboratory setting rather than, well, anywhere else. Perhaps some day we will find definitive evidence one way or another; the most compelling would be a sample from either a lab or an animal reservoir of a virus much closer to the one isolated from infected humans early in the pandemic than the bat coronaviruses so far identified. More likely, we will continue to live with some uncertainty, although without further information the non-lab scenario will remain the most likely. From an epidemiological perspective, lab scenarios seem less plausible because the virus is so variably infectious between humans. Many infected people never go on to infect someone else. So for a pandemic to start with a single lab exposure, we’d likely need an unbroken chain of several unlucky transmissions to keep the infection circulating long enough to achieve a sort of critical mass. That might seem contradictory, given the fairly predictable doubling of cases numbers we saw in the absence of mitigation efforts, but you need a certain minimum number of cases to get that kind of statistical reliability. At very low numbers of cases, such as you might have in an accidental lab exposure scenario, the variability in infectiousness is more significant and can drop infections down to zero again. Outside of a lab, if there is regular contact between humans and an animal reservoir, there is the possibility for multiple crossover infections. In that case, any single introduction may also flame out, but in aggregate an epidemic spread is more likely. Still, at this time we cannot completely rule out a lab release, and so the idea will likely continue to seem more compelling to some.
What do all of these items have in common? To me, they all demonstrate a human tendency to more readily engage with human forms of intelligence and agency. Learning by listening in on a conversation is something we’ve all done and can understand; finding statistical patterns by cross-indexing multiple large datasets is less familiar. While a laboratory accident may not itself represent intelligence, laboratory science is popularly linked to intelligence, with an attendant possibility of human-induced genetic changes. The steps to evolve from a bat virus to a human virus are comparatively less familiar and so perhaps seem unlikely, even though the evolutionary process is itself a form of intelligence. And dog communication apparently invites a fair amount of skepticism; I’m sure the uncharacteristic-for-humans combinations of words and concepts doesn’t help. In that way, dog intelligence is reminiscent of machine intelligence, which can also combine concepts in ways humans never would, sometimes leading to brand new chess strategies and sometimes leading to game show mistakes that might raise even a child’s eyebrow.
As the world eagerly awaits a US Department of Defense report on UFOs with the hope it might answer questions about extraterrestrials (I’m not expecting anything conclusive), we shouldn’t overlook the opportunities to engage with these unfamiliar forms of intelligence we already know to be among us. And who knows? Maybe it will help us relate better should we ever encounter an alien intelligence from another planet.