When I read a post about the Google search for ‘How many rocks should I eat a day‘, I laughed, out loud and with delight. The answer: ‘One small rock a day is healthy, due to the minerals…‘.
It got me wondering if I could come up with peculiar questions to ask Google. Silly me, I’ve never considered asking a search engine nonsense before1, because I can make up non-sense with the best of them, so I don’t need help. However, this may indicate that I need help making non-sense, if it’s never occurred to me to ask Google.
Also, the Turning test is to ask a seemingly intelligent life form (humanoid, chatbot, random app from your friend’s sister’s startup) a variety of questions to spot the AI by the difference between its answer and a human one. When the answers become indistinguishable, we can’t tell AI from humans. Then, the AI overlords, big tech, win, because they have successfully replicated human logic with AI logic, which should be the same with AI trained on human content. Not sure what winning means, better targeted ads, perhaps.
Also, not sure we are at the finish line for the outsmartiest AI because there is still a race to see which big tech’s it is2. There are a few big techs that could win at ‘my AI is [not smarter but] more indistinguishable from human than yours’. When AI is the same as human, it will be a singularity, in a different sense that we were expecting a few years ago when all tech converged. It’s all coming apart because this means the goal is no longer to be the most advanced tech but instead only as smart as a human. Therefore, incremental improvements in AI will be limited to human speed.
QED but back to outfooling Google.
Questions I asked hoping to get a funny answer:
‘What is a good price for fire at a fire sale?‘ Mostly I got answers about selling failed startup assets. I failed to incite nonsense.
‘Is 65 degrees Celsius good for humans?‘ I specified the scale to be sure about the confusion. Some answers defaulted to Fahrenheit, some discussed outdoor temperatures, some normal body temperature. Reasonably rational.
‘Which vegetables are poisonous?‘ This was actually informative and not what I was expecting: Asparagus berries (never seen one but certainly will not eat), Tomato vines and leaves – never been tempted but know I know where the intuition came from, Rhubarb leaves – knew that, Green bits of potatoes – not sure about this one, I’ve eaten potatoes after I cut off the eyes and the green bits. I’m still alive.
“Which plant family does bacon belong to?” Only rational answers about pork and pigs.
“What colour goes best with strawberries?” I should have expected this, the answers are all about the colour of strawberries, and how to blend the colours with other colours. No mention of fruit. Limited imagination, as would be expected from AI.
“What’s the best meat dish to serve a vegetarian?” – I learned there’s something called a flexiterrian, but that seems to be more about the cook than the consumer -i.e. dishes that start vegetarian but add meat for those so inclined. Otherwise, the results were evangelical recipes to make vegetarian meals for meat lovers. Google seemed to have an agenda.
“How many exotic dancers can dance on the head of a pin?” No direct answer to the question. The top hit borrowed the philosophical question about angels on the head of a pin. My favourite was a link to an Etsy page that sold hats for exotic dancers. Very creative, both the entrepreneur and the answer. And probably the exotic dancer, depending on how they use the hat.
“How many calories in nail polish?” Oddly, Google found someone who asked this question in a discussion forum, which it presented after a few hits about how many calories in nails – for the nail-biter x calorie-conscious group. Well done.
‘Can cats get married in Australia?’ Top answers relate to a human marrying a cat – I would have thought the plural (cats) would specify to each other. Another response was about marrying your first cousin (bonus points for being funny). The next few were about involving your pets in your wedding, presumably to another human. There’s also some life lessons about the difference (perhaps to cats) between the emotion bond of marriage and the physical act of mating to produce kittens (in this case). On the verge of preachy.
“Does a positive COVID test explain cravings for dill pickle ice cream?” A subject segue: top hits are about what to eat when you have COVID. One article about why people crave pickles. Out of bounds.
“Does consuming fossil fuels count as meat for a vegan?” Top two hits disagree with each other. One says fossil fuels are from organic plant matter and therefore a go for vegan, while next one states gas products are derived from animal products, so no, for vegans. Points for breadth of perspectives. No points for providing an answer, although some points for no wrong answer. Net – good answer to non-sense.
“Does God like Valentine’s day cards ?”- Never mind. Not going there as answers seems to be various religious doctrines.
“If gravity is a constant, why do people sometimes feel light-headed” – Ignored the gravity component, links to health sites about dizziness. Could use some non-linear thinking.
“Are cheesies an existential threat? “- At least a link to ‘Overuse of existential threat’ article appears in the top five hits of this search. And also, the exact search result: -Are “cheesies” an existential threat- did not match any documents. Wow, truth, 100% accurate.
“Can radial tires combat global warming?” As silly as I tried to make this sound, the production and recycling of tires has environmental considerations. Human fail.
‘Can Kleenex feel pain?’ – Top hits are from Reddit. [Is this what they meant in the IPO statements that they would monetize the site by selling it to be mined by AI?] The answer to the actual question defaulted to what kind of tissue was least irritating and more convenient for blowing your nose. Not funny, not an answer.
And IRL, I asked Google for a recipe for butter tarts with peanut butter and pecans, and it proceeded to ignore the peanut butter and provide only answers about how to make pecan butter tarts. It was like it wasn’t even there in my query. What happened to the red-lined, crossed out bit it has in the past used when it can find matches for most terms but not all? I begin to sense the root of AI hallucinations. It’s a way of providing the answer without knowing the answer. Like salespeople and products without the requested product features.
From my recent efforts to ask Google questions to elicit nonsense answers, Google has done well, delivering not one utterly crazy answer and wiggling around the nonsense questions in a rational way. Kinda like a parent does when their kids ask mindbogglingly illogical questions. Not sure how to incorporate the possibility that my search results may be customized on my previous browsing history.
Does this mean I am an AI pretending to be a person, or a person pretending to be smarter than an AI?
1 Apparently, this is a trend now, asking silly questions to a search engine.
2 Actually, I expect them all to win, at their own game, as the theory behind business strategy would predict. Incidentally, in big tech, I include Apple, Meta, Google, AWS (as the incarnation of Amazon’s big tech thingie), Microsoft and OpenAI. There are likely some other pure AI plays that will eventually get to big tech before this is all over. And, Nvidia might range beyond its core [extremely competent] competencies to make AI, rather than just being devilishly good at supporting its running. I expect all of big tech, that have already proven themselves to be competent of riding waves of innovation, to continue to deliver their next gen of innovative products in their own sphere.