As you can see it carries on just fine and can fool a casual observer. But the longer you carry on a conversation with it the more apparent that Cleverbot is frustrating to talk to, not so much that it isn’t human — after all, all of the responses are taken from human sources. If it weren’t so good at emulating a human from which you expected more, you wouldn’t be frustrated.

Cleverbot is frustrating in two ways:

• it isn’t interesting
• it doesn’t make sense

In other words, it lacks deepness, like a shallow human. Why?

Some say that Cleverbot isn’t an AI because it’s just humans talking to humans via the intermediary of a selection algorithm. One is tempted to be dismissive like this but it’s too convenient. It places too much weight on the role of the corpus and too little on that of learning and interaction. Cleverbot clearly captured some aspect of the human intellectual process that raised it to the level of a shallow conversation partner. What it neglected though will be enlightening to the question of what is deepness.

The most obvious way in which Cleverbot is shallow is in novelty. This is a result of mimicry of others. Mimicry is of course fundamentally human because we learn by mimicking. It doesn’t preclude novelty of some sort. In fact, although it isn’t clear at what level Cleverbot replicates from the corpus, whether in phrases or entire sentences, we are forced to accept that some novelty is injected into the selection process, as responses don’t come out the same every time. However it’s not enough to make Cleverbot consistently interesting, probably because Cleverbot doesn’t synthesize new ideas, merely passing on received wisdoms at the appropriate moments. It seems deep novelty requires more atomicity, and more computation.

The more subtle way in which Cleverbot is shallow is in contextual memory. Cleverbot, probably also for computational reasons, just makes remarks in response to the most recent context. As a result, it changes topics frequently, and doesn’t pursue a line of inquiry to its satisfactory conclusion because it doesn’t know what a line of inquiry is. It can be described even as evasive or flighty. While each response usually makes sense, Cleverbot doesn’t make sense over the span of a conversation, which requires more persistence.

It seems deepness is related to computational resources. And if we believe more intelligent machines — those that exhibit deepness and frustrate us less — to be more “human,” then we must expect a high level of computational competence as a human trait. Deep conversation is a measure of that. Disturbingly, the kind of conversation that Cleverbot can make is rather the more commonplace case among humans, yet it borders on meaningless drivel because it produces nothing new. Even more disturbingly, this is despite that Cleverbot can be quite clever sometimes, being stimulated to produce some of its store of wisecracks after elicitation by a motivated human counterpart who steered it to them. It shows that it doesn’t take much to carry on one half of a conversation, and deceptively so especially if one of the two is playing the role of an intelligent “human.”

Maybe being “human” isn’t that hard. Maybe the binary identification of machine/human should be considered a necessary quantification artifact, and the Turing test really measures a gradation of deepness instead (it is a probabilistic test after all, and of course subjective). If so, let’s run the Turing test on people in a double-blind way.

The test proposed by the article solves some of these problems by requiring generation, and by using distraction after the short target tone is produced. The point is to move on from the target tone faster than consultation with hidden external references can take place. If recognition is not immediate, then one must first hold the note in short-term memory, then after the distraction, compare it to internal reference pitches from tonal memory. This is not accurate since short-term tonal memory itself is not stable, being influenced by distraction. So for some small number of tones (could be all of the chromatic scale), HTM could do well, depending on the person, but maybe performance is not even…, and HTM should never be able to match lesser-heard (e.g. non-standard) pitches well… However, if recognition is by APE, then any tone can be immediately recognized into an abstract form and as something distinct, and easily matched later in the abstract forms.

Under this regime, it would seem that most people who recognize and generate tones upon request probably just have varying degrees of HTM and have developed a quick lookup table as internal reference, which would seem to be malleable by training as with other kinds of memory (for people with good associative memory anyway). APE, however, probably cannot be learned — it’s a kind of idiot savant skill like people who know large number multiplications in one second — it just cannot be done with a lookup table.

http://www.maxim-ic.com/appnotes.cfm/an_pk/742/

Turns out it is not quite what I thought, and it is pretty neat. It does convert between two complex numbers, but the relationship has nothing to do with rectangular to polar. It’s the real and imaginary grid lines of normalized load impedance (the circles) layered on top of the real and imaginary grid lines of normalized reflection coefficient (the straight lines). Normalized load impedance and normalized reflection coefficient are functions of each other, so the Smith chart is used to convert between them. Very nice!

Part 1.

Today I became suspicious of Seagate products (and my fortune in general)

Windows XP was running, and programs were being used. The disk was probably being accessed for memory cache. I noticed the drive making repetitive noises, spinning down and then spinning up, and the machine became unusable. I power-cycled the machine, and it was “NTOSKRNL.EXE is missing or corrupt.” Bad news.

To diagnose the problem, I booted the machine into “Dell diagnostic partition”, located at the front of the hard disk. This partition booted with no problem. Good? I used the “disk diagnostic” tool, which passed basic disk parameter checks (size of disk, for example), but upon reading the disk soon found read errors on some sectors. I started the same test at a later location to skip the damaged sectors, and passed a long error-free zone of the disk, but then encountered more erroneous sectors. After this, I tried the “disk confidence test” tool, which reads only small portions of the disk at a time, as a representative sampling of disk integrity. The test was proceding well, but I halted it for time reasons.

So, in goes the XP CD and Recovery Console comes up, but the OS installation can’t even be found on the disk and there is no C: drive to inspect. Ouch. This is serious. One screw off and out pops the hard disk. Hoho! It’s a Seagate Momentus ST94811A, a so-called “reliable” drive. This machine is barely two years old! Woe be to Seagate which made the drive. Woe be to Seagate Momentus which is designed to run way too hot under the palm for a laptop hard disk anyway. That should have tipped me off. Hrmmmm.

Lessons today:

• When hard disk seems to churn more than usual paging (as in defragmentation-like churn), it may die soon.
• When your browser repeatedly crashes for no apparent reason after paging in from disk, yet your RAM is verifiably good, your disk may die soon.
• White-out is reasonably good plastic glue.
• Dell diagnostic partition takes up almost 50MB of disk space.
• Please choose a Seagate Momentus if you would like to burn your palm and lap.
• If there is some data you would rather not lose at this very moment, back up now.

On to Part 2.

Are there existing conventions dealing with technological aid in various kinds of skills testing? I can think of some, and they generally seem to fall on the side of accepting technology, with restrictions that can sometimes be arbitrary. For example, on the SAT you could use a graphing calculator, but not one with a “QWERTY” keyboard, so a TI-89 was okay but the TI-92 was not, even though they ran the exact same firmware. Not sure if they changed this. Guess not. The College Board still appears to be living in the stone age with regard to some of these banned calculators:

• calculators with QWERTY (typewriter-like) keypads arbitrary
• calculators that contain electronic dictionaries they all do now or can
• calculators with paper tape or printers cash register? lol
• calculators that “talk” or make noise right
• calculators that require an electrical outlet haha
• etc…

That aside, the car that drives itself has got to be the most popular civilian application touted by sensor networks people. So far, most automated components of the car are not fully automated. There is always some human element in some key part of the chain, unlike in airplanes. There is some human resistance toward giving up control on this matter. The trend toward more automation may be unstoppable, though, if automation creeps in a bit at a time like this.