I am sure that many readers have tried for themselves, or at least heard about, the famous experiment done in 1999 by Daniel Simons and Christopher Chabris. If you are not aware of what I am talking about, check out this video before reading on. I recommend watching the video on full screen mode.
I have tried this out in a classroom with about 20 people and although I did not keep statistics, a fair number did not see the gorilla and refused to believe that there had been one until the video was run again.
The result that 70% of viewers do not see the gorilla has become widely cited to sustain the idea that people can be blind to the obvious. But Teppo Felin, professor of strategy at the University of Oxford’s Saïd Business School, writes that this interpretation is wrong and that there is a better interpretation that is consistent with the experimental results ad also what we know about observations in general.
As Felin notes, there are a lot of things in the video, other than the gorilla, that are obvious and yet are unnoticed by viewers, so why focus on the gorilla? He says that what the experiment tells us is that people are rarely looking at things randomly. We know that our eyes don’t see the entire field of vision at once like a camera but darts from point to point and builds up the entire image from those scattered images. At any given moment, people are usually looking at something or for something and this distracts them from everything else. In this experiment, people were deliberately distracted so we should not be surprised.
Imagine you were asked to watch the clip again, but this time without receiving any instructions. After watching the clip, imagine you were then asked to report what you observed. You might report that you saw two teams passing a basketball. You are very likely to have observed the gorilla. But having noticed these things, you are unlikely to have simultaneously recorded any number of other things. The clip features a large number of other obvious things that one could potentially pay attention to and report: the total number of basketball passes, the overall gender or racial composition of the individuals passing the ball, the number of steps taken by the participants. If you are looking for them, many other things are also obvious in the clip: the hair colour of the participants, their attire, their emotions, the colour of the carpet (beige), the ‘S’ letters spray-painted in the background, and so forth.
In short, the list of obvious things in the gorilla clip is extremely long. And that’s the problem: we might call it the fallacy of obviousness. There’s a fallacy of obviousness because all kinds of things are readily evident in the clip. But missing any one of these things isn’t a basis for saying that humans are blind. The experiment is set up in such a way that people miss the gorilla because they are distracted by counting basketball passes. Preoccupied with the task of counting, missing the gorilla is hardly surprising. In retrospect, the gorilla is prominent and obvious.
The alternative interpretation says that what people are looking for – rather than what people are merely looking at – determines what is obvious. Obviousness is not self-evident. Or as Sherlock Holmes said: ‘There is nothing more deceptive than an obvious fact.’ This isn’t an argument against facts or for ‘alternative facts’, or anything of the sort. It’s an argument about what qualifies as obvious, why and how. See, obviousness depends on what is deemed to be relevant for a particular question or task at hand. Rather than passively accounting for or recording everything directly in front of us, humans – and other organisms for that matter – instead actively look for things. The implication (contrary to psychophysics) is that mind-to-world processes drive perception rather than world-to-mind processes. The gorilla experiment itself can be reinterpreted to support this view of perception, showing that what we see depends on our expectations and questions – what we are looking for, what question we are trying to answer.
In other words, there is no neutral observation. The world doesn’t tell us what is relevant. Instead, it responds to questions. When looking and observing, we are usually directed toward something, toward answering specific questions or satisfying some curiosities or problems. ‘All observation must be for or against a point of view,’ is how Charles Darwin put it in 1861. Similarly, the art historian Ernst Gombrich in 1956 emphasised the role of the ‘beholder’s share’ in observation and perception.
The point that Felin is making is consistent with philosopher of science Karl Popper’s idea that we do not simply ‘observe’ and to ask some one to do so is pointless. We always have to tell them what they should observe. In the context of science, experimental observations are always theory-driven orm as Albert Einstein put it in 1926, ‘Whether you can observe a thing or not depends on the theory which you use. It is the theory which decides what can be observed.”
Felin goes on to challenge the idea proposed by AI advocates that computers, not being subject to humans emotions and feelings, would be better able to not be distracted. He argus that the idea that computers and AI are immune to the fallacy of obviousness is wrong because being able to process more data faster does not imply the ability to discern what is relevant and should be observed and he is skeptical that they will ever be able to do so.
Deciding what is relevant and meaningful, and what is not, are vital to intelligence and rationality. And relevance and meaning continue to be outside the realm of AI (as illustrated by the so-called frame problem). Computers can be programmed to recognise and attend to certain features of the world – which need to be clearly specified and programmed a priori. But they cannot be programmed to make new observations, to ask novel questions or to meaningfully adjust to changing circumstances. The human ability to ask new questions, to generate hypotheses, and to identify and find novelty is unique and not programmable. No statistical procedure allows one to somehow see a mundane, taken-for-granted observation in a radically different and new way. That’s where humans come in.
The difficulty of AI to discern intent in others is illustrated by this example.
In Austin, Texas, Google has been testing one of its newest innovations: the self-driving car. While the car is designed to work with—not against—cyclists, by reading and interpreting hand signals, what it seemingly hasn’t taken into account is a cyclist’s ability to hold a trackstand rather than put a foot down at a stop sign. Or at least, that’s what one cyclist reports experiencing.
The story, which the cyclist in question reported in a cycling forum, has been picked up across the globe for its implications concerning cyclists and smart cars, as well as its weirdness. A Google self-driving Lexus had been road testing in his neighborhood for a few weeks, and near the end of a recent ride, he stopped at a four-way stop sign just after the Google car did—so he paused in a trackstand, waiting for the car to make a turn.
“It apparently detected my presence (it’s covered in GoPros) and stayed stationary for several seconds,” he writes. “It finally began to proceed, but as it did, I rolled forward an inch while still standing. The car immediately stopped… I continued to stand, it continued to stay stopped. Then as it began to move again, I had to rock the bike to maintain balance. It stopped abruptly.”
If only the GoPro footage were available! “We repeated this little dance for about two full minutes and the car never made it past the middle of the intersection,” the rider continues. “The two guys inside were laughing and punching stuff into a laptop, I guess trying to modify some code to ‘teach’ the car something about how to deal with the situation.”
He concluded his story by adding that he felt safer with the self-driving car, despite the confusion, than he did with human-operated vehicles.
Of course, the forum lit up with comments; some serious, some funny. Forum user Alan E responded with a bit of both, saying, “Obviously, it was confused by your presence, but at least had been programmed with some common sense to stop and evaluate the situation. Either that, or it was simply admiring your track-stand skills.”
To be fair, the self-driving car may have been confused by the cyclist, but at least it was courteous.
Of course, saying that there are things that computers will never do falls into the category of famous last words. But at the same time, the promises of AI have also been oversold over the decades.
It is interesting how there is a growing list of what seemed like well-established results in the field of psychology that have been questioned and even discredited. Unlike some others, this particular one is not challenging the reproducibility of the results but is instead arguing that the conventional explanation is not the right one.