Comments by "earthstick" (@earthstick) on "Computerphile" channel.

I think this is a bit of a drama queen video. I have encountered these problems many times. I last encountered this when someone had written a confused timeout mechanism and stored their timeout value in local time. When the clocks changed their timeout was triggered when it should not have been. Just about all these problems are solved simply by converting all dates and times to UTC and storing them in UTC. UTC is not subject to daylight savings and times from different timezones become comparable once transposed into UTC. Only use local time in user interactions, i.e. the GUI.
2
Would a reasonable test be to present the AI with a paradox, such as the liar paradox (everything I say is a lie), and ask it to say if the statement is true or false? My reasoning is that there are no general means to detect the halting problem, and the liar paradox is a paradox because it is self-referential. Therefore, a machine will suffer the halting problem when evaluating the statement. Of course, we will not be able to tell if it will eventually return or never return, but we can say it is taking too long to come to a conclusion for us to be interested in it.
1
@llamatronian101 I meant it as much as a general test for intelligence, really. I agree that there will no doubt be programmed responses for common paradoxes. So the test would require a novel paradox or one phrased in a novel way.
1
To store a number precisely, there must be a common factor between the base of the number system and the divisor. For example, 10 does not divide precisely by 3. It gives 3.333 recurring. 10 does however divide precisely by 4, it results in 2.5. 10 and 4 have a common factor, it is 2. The problem with floating point numbers in the computer is that they are stored in base 2. 2 is prime and has no factors other than 1 and itself.
1