1. Childcare: er, nope. Look at the "changing a nappy" problem from a comp. sci. perspective and you realise that it's one of the HARD problems in AI. You've got vision -- recognizing the different bits of the baby and tracking them in real time (babies don't stay still and they change shape). You've got chemotaxis -- establishing when the nappy needs changing (as opposed to, baby just farted). You've got proprioception: your nappy-changing robot needs to be able to fold soft structures (nappies) in three dimensions around another structure of variable texture which is extremely pressure and temperature sensitive and can be damaged trivially easily. Worse: it's not just snipping away a soiled nappy and applying a new one -- you've got to clean the baby's bottom (a moving three-dimensional surface), you've got to apply any necessary ointments, baby powder, oil, or medication, and you've got to do this all without hurting or irritating the baby.
My take is that nappy changing is basically an "AI complete" problem -- if you can make a robot that can change a nappy, you've cracked most of the hard problems in interaction with human beings. (You can't lie to a baby by having a set of canned textual interaction scripts like Siri -- it's all about touch, vision, smell, and so on.)
Oh, and don't get me started on having robots teach babies useful stuff, like how to speak, or how to smile, or how to walk.
2. Chef: well, maybe. But McDonalds is already largely automated -- the shops are just front-ends for a factory/supply-chain distribution network that has enormous economies of scale because the product is uniform. In effect all they do in the kitchens at McD's is final assembly. Real cooking? Again, that's a hard job to automate -- especially as it's a low-wage/low-margin service role, and it involves quality assurance/hygeine issues (hint: is your robot able to detect the presence of pests, ranging from flies and rats to next door's cat?).
3. Tour guide: maybe. Self-driving buses and canned scripts will get you some distance. But responding to tourists' specific needs jacks up the complexity of the task a huge distance -- especially if it's something like "needs a special diet that they didn't specify when they booked the tour" or "is having a heart attack in the middle of nowhere".
4. "armies of robots trawling the web for potential interesting stories and writing automated stories" -- Politely, this is bullshit. Web scrapers are not "writing" stories, they're summarizing press releases that someone else wrote. While there's some progress in automating typescript output based on transactions in highly ritualized sporting events with well-defined scoring systems, or stock market and weather reports, these are special cases with very tightly constrained problem spaces. As far as the web crawlers go -- all they represent is news outlet owners firing journalists and reprinting whatever advertising bullshit they're fed, with just enough random word salad resequencing to get them off the hook for copyright violation.
5. "True Love" was not the first computer written novel; try "The Policeman's Beard is Half-Constructed", from the late 1980s. As for the evident contempt you have for Mills & Boon (formula romance) novels ... yes, there's a formula: the art comes in how the author runs the variations on it to keep it from becoming boredom. Once you can automate boredom-avoidance, it might be possible to automate the assembly process for some limited formats. But good luck with that: again, it's an AI-complete problem.
6. Doctor: sure, we have diagnostic equipment. We've had diagnostic expert systems since the 1980s, and in limited problem domains they're very helpful. Go too far though and you run into the happy fun intersection of the internet of things, malware propagation, and the medical device support market -- for which, see comp.risks passim.
1. Childcare: er, nope. Look at the "changing a nappy" problem from a comp. sci. perspective and you realise that it's one of the HARD problems in AI. You've got vision -- recognizing the different bits of the baby and tracking them in real time (babies don't stay still and they change shape). You've got chemotaxis -- establishing when the nappy needs changing (as opposed to, baby just farted). You've got proprioception: your nappy-changing robot needs to be able to fold soft structures (nappies) in three dimensions around another structure of variable texture which is extremely pressure and temperature sensitive and can be damaged trivially easily. Worse: it's not just snipping away a soiled nappy and applying a new one -- you've got to clean the baby's bottom (a moving three-dimensional surface), you've got to apply any necessary ointments, baby powder, oil, or medication, and you've got to do this all without hurting or irritating the baby.
My take is that nappy changing is basically an "AI complete" problem -- if you can make a robot that can change a nappy, you've cracked most of the hard problems in interaction with human beings. (You can't lie to a baby by having a set of canned textual interaction scripts like Siri -- it's all about touch, vision, smell, and so on.)
Oh, and don't get me started on having robots teach babies useful stuff, like how to speak, or how to smile, or how to walk.
2. Chef: well, maybe. But McDonalds is already largely automated -- the shops are just front-ends for a factory/supply-chain distribution network that has enormous economies of scale because the product is uniform. In effect all they do in the kitchens at McD's is final assembly. Real cooking? Again, that's a hard job to automate -- especially as it's a low-wage/low-margin service role, and it involves quality assurance/hygeine issues (hint: is your robot able to detect the presence of pests, ranging from flies and rats to next door's cat?).
3. Tour guide: maybe. Self-driving buses and canned scripts will get you some distance. But responding to tourists' specific needs jacks up the complexity of the task a huge distance -- especially if it's something like "needs a special diet that they didn't specify when they booked the tour" or "is having a heart attack in the middle of nowhere".
4. "armies of robots trawling the web for potential interesting stories and writing automated stories" -- Politely, this is bullshit. Web scrapers are not "writing" stories, they're summarizing press releases that someone else wrote. While there's some progress in automating typescript output based on transactions in highly ritualized sporting events with well-defined scoring systems, or stock market and weather reports, these are special cases with very tightly constrained problem spaces. As far as the web crawlers go -- all they represent is news outlet owners firing journalists and reprinting whatever advertising bullshit they're fed, with just enough random word salad resequencing to get them off the hook for copyright violation.
5. "True Love" was not the first computer written novel; try "The Policeman's Beard is Half-Constructed", from the late 1980s. As for the evident contempt you have for Mills & Boon (formula romance) novels ... yes, there's a formula: the art comes in how the author runs the variations on it to keep it from becoming boredom. Once you can automate boredom-avoidance, it might be possible to automate the assembly process for some limited formats. But good luck with that: again, it's an AI-complete problem.
6. Doctor: sure, we have diagnostic equipment. We've had diagnostic expert systems since the 1980s, and in limited problem domains they're very helpful. Go too far though and you run into the happy fun intersection of the internet of things, malware propagation, and the medical device support market -- for which, see comp.risks passim.
Reply
Leave a comment