What are the current limits of artificial intelligence? & their weaknesses

Discussion in 'Trading' started by pinetboltz, Apr 9, 2017.

  1. pinetboltz

    pinetboltz

    It seems many tech companies are working on self-driving cars recently, and there are varying estimates of when autonomous vehicles will actually replace human drivers on the road.

    Tbh, it's kind of bizarre that they haven't done this already, in that the machine learning algos have already won jeopardy, chess, poker, Go, etc vs the best professional players in the world, and still the algos haven't cracked driving to the extent of fully matching the performance of even the average driver.

    So what are the current limits of AI? & what are their weaknesses, in the sense of producing sub-standard results in certain types of activities?

    http://www.cnbc.com/2017/03/26/uber-self-driving-car-arizona-crash-suspended.html
     
  2. Quiet1

    Quiet1

    Difficulty in AI is I think the inverse of what most expect: higher intellectual functions are easier; lower more intuitive functions are harder. Basically the longer that a set of mental functions has existed in humans and apes the harder it is for AI to replicate.
     
  3. Quiet1

    Quiet1

    So, anything you can do without really thinking about it is hard...anything involve concious thought is easier.
     
    DrNo likes this.
  4. Turveyd

    Turveyd

    There is no real AI yet, some clever programs which they class as AI but that's all they are and there not AI.

    I personally, don't think we'll ever see real AI!
     
    truber likes this.
  5. Zzzz1

    Zzzz1

    How about you start working on some basic statistics basics. How many accidents have self driven cars caused given the many millions of miles they have been cruising California and other places for years now. And that's just the starting point. How many accidents are caused by human error just for each 10000 miles driven? When you put things in perspective you would not claim AI in self driven car does not reach the abilities of an average driver. Every seen fresh off the boat Indians and Pakistanis driving around New Jersey turnpikes? Or fish off the boat mainland Chinese who change lanes ever 30 seconds because they think they get to their destinations 5 seconds earlier that way? Because if you know about their driving abilities and how ridiculously low the passing requirements are of most state's driving test centers then you will appreciate self driven cars a lot more. My Japanese wife who has impeccable driving skills got bumped into 3 times in one year alone by some idiots in parking lots and on roads in the US.

     
  6. bookish

    bookish

    Narrowly defined problems/activities are easier to do with AI.
    You can categorize, if you are able to extract and present data to the AI.
    You can approximate a function, if you are able to extract and present data to the AI.
    Extracting data is a little more hard to do with an AI, but can be done if you know what data you want.
    But there is no "just do it" You have to create an AI for each narrowly defined thing you want to do, and tie them together. The more "real world" you get, the harder it is to do it reliably and safely with AI.

    Things are getting better now with deep ai / convoluted nets. They can play games, but it takes a LOT of trial and error and as soon as things get out of whack they fail. You can train them to play one level, but the next level does not work.

    Driving a car is one of the more difficult tasks and it is grossly irresponsible for anyone to even suggest self driving cars at this point. This is a great example of the jumping-in-head-first coke-whore mentality that a lot of executives in this country have. If someone is too lazy or stupid to drive a car they shouldn't be driving at all. But these guys are all wide-eyed and stroking their cocks at the futurism instead of trying to be responsible. This is why they need better psychological testing for everyone, especially in policy making positions. Sadly the psychologists are goofballs too. I know, I have a degree in psychology. You know that song about killing all the lawyers? Do the psychologists first. Psychiatrists are even worse. Most of them belong in the nut house as patients. That's not a joke, its well researched.

    I hesitate to suggest even a "brakes only" self driving car. Hesitation and inappropriate braking can cause an accident as easy as avoiding one.

    I think the whole self driving car thing is just an excuse so some execs can program every car in the country to go headlong into any person they don't like. "Oops, AI malfunctioned. Here is some pocket change for the dead guy's family and the police"

    When self-driving cars become common, I'll be driving around in an armored semi tractor.

    On the up-side. We could train one to be president and do better than we have in a long while.
     
    Simples likes this.
  7. I believe Tesla advertises on their site that if you get the fully loaded version that it's self driving capability is something like twice that of the average driver.
     
  8. bookish

    bookish

    "advertises" being the key word here.
     
  9. Zzzz1

    Zzzz1

    The stats speak for themselves. Self driven cars TODAY are already statistically speaking 100 times safer than an average human driver. By the way,the benchmark should not be an average driver but the worst drivers. People are not killed on roads by good or average drivers but stupid and reckless drivers. Why, for example, not all states and countries enact zero Promille laws, basically zero alcohol before or when driving is a mystery to me. Anyone who gets caught driving with alcohol in their blood should be banned from driving forever.

     
  10. Food for thought!
    Wait until someone figure out how to hack the GPS satellite and then do ur statistics!:sneaky:
    By the way the Russians and the Chinese already have the tools to do just that. And just imagine what will happen to Cali freeways if GPS satellites get blown out of the sky.
     
    #10     Apr 9, 2017
    OddTrader and bookish like this.