Its good to see the average citizen believes everything there told and incapable of independant thought just like they want.
We will agree to disagree, my lack of abilities to trade outside the system has taught me to only trade system rules and now I don't trade manually at all, eyes weak, lack of patience, body pains, so automation trades much better than me, plus I can't day trade 40 futures markets or scan 15,000 stocks. I commend you if you trade better than what you can program, some traders have unique gifts but I understand my limitations.
Based on the premise of the OP, the benchmark should be the best drivers. If you can teach a computer to win in chess against the highest ranked players, it should have a better driving record then the top human drivers. The key difference is, also, that the known automated driving accidents are obvious cognitive/ML glitches (like a computer failing to recognize a truck in case of the Tesla accident) while most human accidents are either due to bad judgment or due to lapse of attention. A small LOL - a guy entrapping self-driving cars by creating a circle with dashed line on the outside and a solid line on the inside - https://creators.vice.com/en_us/article/meet-the-artist-using-ritual-magic-to-trap-self-driving-cars The obvious explanation, of course, is that there is a big difference between combinatorial activities such as chess or poker and a cognitive mosaic-like activity like driving. So many of our seamless actions are mosaic-based and that part is so hard to encode. For example, there was a paper many years ago entitled "How to wreck a nice beach" - read it out loud quickly and you will realize what the paper is about. * I am not against self driving cars, in fact I can't wait for them to arrive. The topic, however, is about the limitations of AI/ML and, TBH, self-driving cars are less about ML once you get past the basic shape recognition and more about creating a driving algorithm.
Well, why should we care about the premise of the OP? The benchmark should be the average driver because driver-less cars in the future are replacing average drivers based on the law of large numbers, not just top drivers. I agree with most other parts of your post but not the last part. Tons of algorithms in self-driven cars are the result of training deep neural networks with inputs gathered from tons of different sensors over years of data gathering during the development process. Nobody is claiming that the process is completed but why are some people in this thread even discussing whether the AI in this process can be considered "intelligence" or not. What really matters is that those cars already today produce way less accidents than the average driver behind wheels. That is what matters, that is what reduces deaths and fatal accidents. I get riled up when I am confronted with illogical arguments (not referring to yours but the some other poster).
Just out of curiosity, would you distinguish AI from SML? In my mind, SML is nothing more then a clever statistical hack to solve a specific data analysis problem (which, granted, could be pretty complex like facial recognition). I did my degree on using SML to solve real life problems and I believe that we have not seen most of it's potential. On the other hand, AI is about computerized cognition in an area with a lot of non-tokenized variables (IMHO, of course, cause now everyone calls any SML startup "AI"). For example, doing legal document search based on a glorified keyword selection is an application of SML, while devising a strategy for winning this specific case is a cognitive activity that requires a lot of non-tokenized thinking.
Hmm, I am not sure to be honest. It is a terminology issue, AI for me is a collective description of any learning process that is automated. SML is admittedly lower in the food chain with regressions, linear classification, minimax, clustering, and the like. So, I would consider it part of AI. Not of the sophisticated and complex kind but nonetheless. And yes, most startups that peruse the term "AI" just use some regression or clustering algorithm. What is AI however in the keyword search world is if I would confront an algorithm with a natural disaster search term and without lookup tables it would output all relevant, trade able stocks that are with high certainty impacted by such natural disaster based on the products those companies produce, their market share, their competitors, their geographic reach, ...all gathered and weighted with biases during the learning process.
Of course, but I feel like we are arguing about apples and oranges. Your argument is that automation can do many tasks better then 90% of human beings - I'd say that (a) that number is too low and (b) it's not a very difficult benchmark to beat. The majority of human beings pretty much suck at everything, from menial tasks to high cognition activities. It's only natural that you can create an automation that can replace humans in some of those tasks by designing pin-point SML tools. PS. For the record, I am an investor in a several SML/AI related startup companies so my opinions might be biased
Zee Zee Zeee .... Once again , you miss the common sense factor. There's not enough statistical data to make such an all encompassing statement. Do you know how many wrecks will occur because of cars "pegged" to the speed limit in any major city at rush hour? Try driving 55 MPH on 285 through Atlanta at 7 AM, ...you will cause a dozen wrecks.... and thats before you get shot. ... Will AI teach these cars to break the speed limit too? Talk about a lawyer smorgasbord.