I fully believe that autonomous driving is possible in 99% of the cases. Autonomous vehicles are not a new thing, research and experiments have been conducted in this domain at least since the 1980s. Self-driving cars have been running on German and Italian roads since the mid-1990s.
However, this technology in consumer cars is another story. It is between expensive and very expensive right now, and will likely continue to be for some time. This is also a complex system that needs to be monitored and maintained. This will change the way we see automobiles and will not be mainstream for at least another decade, if only to let a large enough number of these cars on the road to see a positive effect on statistics.
The Tesla forum discussion (3rd link) makes it clear that the the Tesla software is not properly aware of overtaking motorcyclist and is actually confused by them into thinking that the *car* ahead has accelerated. Which causes the Tesla car to accelerate in turn, when there is a very real slow car ahead. This is a known issue which at the time of the forum had not been fixed (July 2015). Since then, Tesla Motors has parted ways with MobileEye and reports are that the newer software is not as good.
If they are well trained on a dataset that covers all the relevant situation, they can interpolate, but they cannot really extrapolate all that well. No known machine learning technique know how to deal with a completely new and unexpected situation.
Autopilots in planes have been able to land and exit the runway for quite a while. I'm pretty sure they should be able to take off as well if the plane is well aligned at the start of the runway, but they can't taxi (drive the plane on the ground to the start of the runway).
Why replace C/C++? Compilers do optional bounds checking it already, at least since 2012.
Come on, man, keep up with the times! check out Adress Sanitizer and all the other Sanitizer goodies. Enabled by default in recent versions of GCC/G++ (since 4.8) *and* Clang (with LLVM 3.1). Contributed by your friendly Google developers.
I'm a white male myself and I cannot fathom how some people with my same complexion cannot notice the skin colour and genital makeup of Congress, company boards and most positions of power.
And yet, here you are. You are a white male yourself, and you seem not only able to "understand" the difficulties of being a minority, you also act in what you perceive to be their best interest. Yet at the same time you claim that others are incapable of doing this specifically because they are white males. How can you hold these to opposing views at the same time?
That is not what the GP wrote at all. He wrote that he could not understand why *some* people like himself (white male) did not see a large proportion of white males in positions of power. In other words, he wrote that these few people do not see this *in spite* of being white male, not *because*. This is a very different proposition. In addition, the GP did not write *all white males*, which would have indeed been a generalization, only "some white males", So no contradiction, and no opposing view.
Nah, the Nazis called themselves that way to confuse people, and apparently it works very well, to this day.
At their level of insanity, there is no left or right. If we define left biases as in the James Damore memo: compassion for the weak, disparities due to injustice, and humans are inherently cooperative, the Nazi showed none of these traits. On the other hand, they showed respect for the strong/authority, disparities are natural and just, and humans are naturally competitive. According to the memo, these are all right-wing biases.
This is just an idea published by some Chinese researchers, actually a fairly pedestrian idea based on recent approaches to machine learning in computer vision. The researchers were clever in choosing a controversial name for their approach ("a repression network"), but there is no indication that this approach is actually used anywhere, in China or elsewhere, whether it would scales to millions of cars.
BTW this is the same thing with face recognition. On curated databases of tens of thousands of faces taken in reasonably good condition, face recognition achieves very high scores, better than humans (who would remember 10,000 faces?) but in the real world, not so much, the technology is not that useful yet in practice. It will probably come though, eventually.
Where to begin ? It start in the TL;DR on page 2. The memo asserts there that differences in traits explains in part why women do not represent 50% of the population in tech. Later he write that these traits are universal across human cultures on page 3. Then on page 4, he writes that women are on average more neurotic.
Taken together, this is clearly meant to say that women as a group are on average biologically incapable of working in the "high-stress" world of tech. Even though he writes that one should not equate groups with their average, this is the first thing he does.
There you go, he clearly calls women inferior.
So now my own assertion is that these are all false. First and most importantly, women are definitely not more neurotic than men. There is as high a proportion of mental diseases in men as in women. This is very well documented, talk to any psychiatrist. Second of all, the world of tech is actually very low stress and very low responsibility compared to many other professions. Talk to any Registered Nurse about their 12h hospital shifts and their daily life and death decisions that they have to make. Finally, the so-called traits of women vs. men as represented in the memo are not universal, many are cultural. If you want to talk about a single well-known counterexample, here is one: the Polgar sisters. Look it up and see how many of our assumptions about gender roles and capabilities are actually social constructs.
Infant mortality is defined as the number of deaths of children under one year of age, expressed per 1 000 live births, so your example does not really work. If an infant dies within 1 year after birth, this is counted in the infant mortality rates for all OECD countries.
However, there is a difference in that the United States and Canada are two countries which register a much higher proportion of babies weighing less than 500g, with low odds of survival, resulting in higher reported infant mortality. In Europe, several countries apply a minimum gestational age of 22 weeks (or a birth weight threshold of 500g) for babies to be registered as live births.
Actually difficult image-based problems is where AI will excel, give time and good training data. However a kind of medical "expert system" is not exactly in sight yet. So I basically agree with you: AI will be an excellent tool, boosting productivity and driving costs down. I am not sure it will drive the number of doctors down.
No it learn from *annotated* data. That is data where a human, most probably a doctor, told the system what to look for. Good quality annotated data is scarce and expensive to get and that is the #1 problem in AI research.
Geoff Hinton said that machine learning and particularly deep learning is so good now that anything that amounts to classification will soon be done better by machines than humans. The except is here. So logically medical schools should stop churning out radiologists now.
I would like to agree with him, but note that there is still a large difference between even large synthetic tests and the real world. At the end of the day a trained medical specialist will have to make the diagnosis, machines are not there yet, and companies are not willing to risk this level of commitment. Compare it to self-driving cars if you want: it may be technically doable, but the practical and legal challenges are still enormous.
I should say that I have been working in automating medical imaging for more than 20 years now. 15 years ago, with others I demonstrated a system that diagnosed skin cancer better than humans, and won many awards for it. Nonetheless it was not a commercial success and our company folded. Maybe things will be different because larger companies are behind similar efforts now (like IBM) and they have the time, resources and money to make it happen. However make no mistake, getting these systems accepted is going to take a long time.
So I would suggest that rather than saying inflammatory things that sound like "we CS people do things better than you petty doctors", AI people should propose to work with medical schools to help train specialists to understand, use these systems correctly and benefit from them. Any other attitude is probably doomed to fail.
Slashdot Mirror
I fully believe that autonomous driving is possible in 99% of the cases. Autonomous vehicles are not a new thing, research and experiments have been conducted in this domain at least since the 1980s. Self-driving cars have been running on German and Italian roads since the mid-1990s.
However, this technology in consumer cars is another story. It is between expensive and very expensive right now, and will likely continue to be for some time. This is also a complex system that needs to be monitored and maintained. This will change the way we see automobiles and will not be mainstream for at least another decade, if only to let a large enough number of these cars on the road to see a positive effect on statistics.
This is insightful, please mod up.
All railroad engineers are familiar with this or one of its variants.
Exactly, when Tesla advertises the self-driving ability of their cars as fun and the self-driving problem as easy and soon solved.
It is estimated that of these 100 billions or so, 40% did not live to see their first birthday.
Basically B&B but with very good heuristics. We can solve TSP with 10^6 nodes without too much trouble.
Charge for a trip is one thing, maintaining a good capacity over time is another.
The Tesla forum discussion (3rd link) makes it clear that the the Tesla software is not properly aware of overtaking motorcyclist and is actually confused by them into thinking that the *car* ahead has accelerated. Which causes the Tesla car to accelerate in turn, when there is a very real slow car ahead. This is a known issue which at the time of the forum had not been fixed (July 2015). Since then, Tesla Motors has parted ways with MobileEye and reports are that the newer software is not as good.
More specifically, they are indeed rule based, but the rules are learned, not engineered.
In other words we don't know.
If they are well trained on a dataset that covers all the relevant situation, they can interpolate, but they cannot really extrapolate all that well. No known machine learning technique know how to deal with a completely new and unexpected situation.
Autopilots in planes have been able to land and exit the runway for quite a while. I'm pretty sure they should be able to take off as well if the plane is well aligned at the start of the runway, but they can't taxi (drive the plane on the ground to the start of the runway).
Why replace C/C++? Compilers do optional bounds checking it already, at least since 2012.
Come on, man, keep up with the times! check out Adress Sanitizer and all the other Sanitizer goodies. Enabled by default in recent versions of GCC/G++ (since 4.8) *and* Clang (with LLVM 3.1). Contributed by your friendly Google developers.
Thanks for the short story !
Can we go back to your first point:
That is not what the GP wrote at all. He wrote that he could not understand why *some* people like himself (white male) did not see a large proportion of white males in positions of power. In other words, he wrote that these few people do not see this *in spite* of being white male, not *because*. This is a very different proposition. In addition, the GP did not write *all white males*, which would have indeed been a generalization, only "some white males", So no contradiction, and no opposing view.
Trust me, I'm (enough of) a mathematician.
yes mate, whoosh for you.
Nah, the Nazis called themselves that way to confuse people, and apparently it works very well, to this day.
At their level of insanity, there is no left or right. If we define left biases as in the James Damore memo: compassion for the weak, disparities due to injustice, and humans are inherently cooperative, the Nazi showed none of these traits. On the other hand, they showed respect for the strong/authority, disparities are natural and just, and humans are naturally competitive. According to the memo, these are all right-wing biases.
This is just an idea published by some Chinese researchers, actually a fairly pedestrian idea based on recent approaches to machine learning in computer vision. The researchers were clever in choosing a controversial name for their approach ("a repression network"), but there is no indication that this approach is actually used anywhere, in China or elsewhere, whether it would scales to millions of cars.
BTW this is the same thing with face recognition. On curated databases of tens of thousands of faces taken in reasonably good condition, face recognition achieves very high scores, better than humans (who would remember 10,000 faces?) but in the real world, not so much, the technology is not that useful yet in practice. It will probably come though, eventually.
Have you actually read the memo ?
Where to begin ? It start in the TL;DR on page 2. The memo asserts there that differences in traits explains in part why women do not represent 50% of the population in tech. Later he write that these traits are universal across human cultures on page 3. Then on page 4, he writes that women are on average more neurotic.
Taken together, this is clearly meant to say that women as a group are on average biologically incapable of working in the "high-stress" world of tech. Even though he writes that one should not equate groups with their average, this is the first thing he does.
There you go, he clearly calls women inferior.
So now my own assertion is that these are all false. First and most importantly, women are definitely not more neurotic than men. There is as high a proportion of mental diseases in men as in women. This is very well documented, talk to any psychiatrist. Second of all, the world of tech is actually very low stress and very low responsibility compared to many other professions. Talk to any Registered Nurse about their 12h hospital shifts and their daily life and death decisions that they have to make. Finally, the so-called traits of women vs. men as represented in the memo are not universal, many are cultural. If you want to talk about a single well-known counterexample, here is one: the Polgar sisters. Look it up and see how many of our assumptions about gender roles and capabilities are actually social constructs.
If you do all the work, you do all of the learning. The slacker typically learn nothing and a good grade is not going to change that fact.
Infant mortality is defined as the number of deaths of children under one year of age, expressed per 1 000 live births, so your example does not really work. If an infant dies within 1 year after birth, this is counted in the infant mortality rates for all OECD countries.
However, there is a difference in that the United States and Canada are two countries which register a much higher proportion of babies weighing less than 500g, with low odds of survival, resulting in higher reported infant mortality. In Europe, several countries apply a minimum gestational age of 22 weeks (or a birth weight threshold of 500g) for babies to be registered as live births.
https://data.oecd.org/healthstat/infant-mortality-rates.htm
How is that working out for you?
Actually difficult image-based problems is where AI will excel, give time and good training data. However a kind of medical "expert system" is not exactly in sight yet. So I basically agree with you: AI will be an excellent tool, boosting productivity and driving costs down. I am not sure it will drive the number of doctors down.
No it learn from *annotated* data. That is data where a human, most probably a doctor, told the system what to look for. Good quality annotated data is scarce and expensive to get and that is the #1 problem in AI research.
Geoff Hinton said that machine learning and particularly deep learning is so good now that anything that amounts to classification will soon be done better by machines than humans. The except is here. So logically medical schools should stop churning out radiologists now.
I would like to agree with him, but note that there is still a large difference between even large synthetic tests and the real world. At the end of the day a trained medical specialist will have to make the diagnosis, machines are not there yet, and companies are not willing to risk this level of commitment. Compare it to self-driving cars if you want: it may be technically doable, but the practical and legal challenges are still enormous.
I should say that I have been working in automating medical imaging for more than 20 years now. 15 years ago, with others I demonstrated a system that diagnosed skin cancer better than humans, and won many awards for it. Nonetheless it was not a commercial success and our company folded. Maybe things will be different because larger companies are behind similar efforts now (like IBM) and they have the time, resources and money to make it happen. However make no mistake, getting these systems accepted is going to take a long time.
So I would suggest that rather than saying inflammatory things that sound like "we CS people do things better than you petty doctors", AI people should propose to work with medical schools to help train specialists to understand, use these systems correctly and benefit from them. Any other attitude is probably doomed to fail.