Not so sure I agree. When I try to think of the core difference between subways and robotic vehicles, the only one I can think of is the relative lack of stops and the absence of cross-traffic.
With a serious and reliable driverless car system, stops could be integrated in a way which made them minimally impede traffic flow and the cross traffic optimized to maximize the flow over the whole road network.
Keep in mind we are talking about (in the limit at least) a network of driverless cars which can communicate. If traffic signals are in on the deal, I see no reason we shouldn't be able to do far, far better than a subway system.
Not sure that I agree here. I don't know about you, but my understanding of the variety of metrics they used in this article is quite minimal. Instead, it is instructive to just see the relative variation in the past five years or so compared to the 30 before that.
For instance, I have no idea what the metrics in Figs 2 and 3 are exactly, but I can see that the inequality metric in Figure 2 has reduced itself to approximately 1970 levels and that, since 1995, the welfare metric in Figure 3 has increased by (very roughly) twice the standard deviation of the past 30 years' measurements.
Of course scale is important, but it is folly to trot out the same argument every time a graph does not include the origin. Without knowing something about the y-axis variable we cannot make this type of claim.
I won't argue with your feelings toward Emotiv, since I don't especially trust their demos to be well controlled for various muscular cues.
That said, there is good work done which puts a lot of effort into actually measuring the non-muscular signal from EEGs and training on that instead. EEGs can capture signal from alpha waves (roughly analogous to alertness as you describe), and researchers have come up with numerous specific responses to watch for. One such example is the P3000 response, corresponding to the reaction a person has when they are shown an object which they were picturing in their head (a target).
There are numerous other techniques, many dealing with predicting a type of pattern by looking for activity coming from the corresponding area of the brain. If the sensors and algorithms are written correctly, EEG can absolutely be a neural interface, although one with quite a lot of noise. Done poorly, it can be entirely swamped by the physical motions you describe.
To be fair, we can make huge steps forward once the interface technology has improved resolution. I will trust your girlfriend's expertise, but I do know from mine in Machine Learning that we can extract a surprising amount of signal without knowing the neurophysical processes which generate the data.
A huge problem right now with the products coming out of Emotive and similar companies is that the level of noise is enormous and the resolution of the sensors is very poor. As we continue to remove the noise from the data (and come up with better sensors!), we can train better and better models to differentiate these various activities.
Alpha-Beta pruning is, to my knowledge, only really defined in a mini-max setup. By definition, it leverages the mini-max framework to prune areas of the search space.
You can apply the core concept -- refusing to expend computation effort when the result of that effort is guaranteed to not be used -- to any number of other methods and approaches in Machine Learning and AI.
Minor picky point, but universities have access to such computing power via time-sharing at national labs if it is not available from their own department.
This just seems unlikely. The difference between 100 points at the top end of the scale is minute -- a few problems swing -- and the change in a 1570 to 1600 is quite literally an issue of missing just a couple problems in either section.
The SAT is a horribly noisy measure of excellent in either verbal skills or math, much less actual intelligence. I can tell you I managed to never peak over 700 on the math section only to get a perfect score on both the Math IIC and GRE Math sections (much harder tasks).
I'm going to blatantly ignore most of your post and just ramble about solving AI.
I don't put much stock in the idea of a surprising, monolithic "solution" to AI being discovered. If the past 30-40 years are any indication, AI will continue to slowly encompass harder and harder tasks until one day we stop having easy ways of distinguishing it from whatever we consider intelligence.
They used to say chess was too hard for computers. Then they said computers couldn't do anything but pure logic. Then it was that they couldn't do reasoning unless it was directly programmed in. What will be left in another 10 years?
With a serious and reliable driverless car system, stops could be integrated in a way which made them minimally impede traffic flow and the cross traffic optimized to maximize the flow over the whole road network.
Keep in mind we are talking about (in the limit at least) a network of driverless cars which can communicate. If traffic signals are in on the deal, I see no reason we shouldn't be able to do far, far better than a subway system.