HN Explorer

Creatures from primordial silicon: A lean, mean circuit that nobody understands (newscientist.com)

18 points by mercurio ↗ HN
Gladwell's writing is very seductive but frequently wrong. I like to think of it as speculative non-fiction. It has the same mind-expanding qualities as good SF, and its connection to reality can be just as tenuous.

It is very easy to make up a plausible sounding theory. The hard part is making sure it has no holes in it. The harder part is proving it. As an example, I'll give an alternative evolutionary explanation for why it makes sense that (some) people can accurately judge others in a short amount of time (I claim no expertise in the evolutionary sciences; this is written as a layperson):

Consider an intelligent social species. When two strangers meet, the ability to judge accurately and to be favorably judged, both lead to better fitness. Lets assume there are situations where intelligence is the dominant quality that is being evaluated. Now if you are intelligent, you want to honestly communicate this, so its in your best interest to aid the judgement of the other person. If you are not, then its preferable to be dishonest and try and simulate intelligence. Now why should this situation lead to a state where (some) people can fairly easily make accurate judgements about others? It seems like the person judging has no easy way of accurately evaluating, but has to undertake the hard task of distinguishing between liars and truth-tellers. But the interesting part is that lying and intelligence are highly correlated, since lying takes more brainpower than just telling the truth. So the people who can fool the judgement of others, don't really need to, while those who need to lie, can't do so effectively. This evolutionary dynamic leads to a state where intelligent people can quickly judge the intelligence of others, because they don't have to worry about other intelligent people fooling them.

9 comments

[ 2.6 ms ] story [ 21.7 ms ] thread
Novash
Ok, let's see if I got it right. He wrote a program that could pick pieces of other programs after testing to find the "fittest" and made new ones until he found one that worked "correctly"? Was it?
pygy
You're halfway there. What you describe is genetic algorithms in their classical implementation.

Here he uses a FPGA, a kind of processor that can be rewired at will, in hardware. The gates can be programmed into the various boolean operators you know, usually to implement a specialized binary processor (say, a DSP).

Binary here means that after a set amount of time set by a clock, every cell will either be in a ON (1) or OFF (0) state.

The researcher here uses genetic algorithms to find the best configuration of gates to perform the required task.

What's new here is that the fittest "program" doesn't work in a digital fashion like classical processors. It doesn't have a clock, and acts as an extremely minified electronic analogical circuit.

I'd love to see how this could be combined with hybrots...

http://en.wikipedia.org/wiki/Hybrot

Prrometheus
Yeah. That's what I'm going to specialize in at grad school. Be jealous.
fiaz
I would like to know if words sufficiently similar sounding to "GO" and "STOP" evoke the same impulse in the circuitry to accordingly output the voltage. I'm guessing that they will if the signals are sufficiently similar.

There is a difference between symbol and signal; as impressive as this story is, it is important to make this distinction because the article seemed to imply processing on the symbol/semantic level as opposed to the signal level (which I am assuming is the dominant modality of interpreting the words "stop" and "go"); there were also hints of awareness as in:

"It appears that evolution made use of some physical property of these cells—possibly a capacitive effect or electromagnetic inductance—to influence a signal passing nearby. Somehow, it seized on this subtle effect and incorporated it into the solution."

As an example of the difference of symbol/signal, a chimpanzee can be trained to obey traffic lights according to what they represent. The chimp will no doubt have the capacity to do what the lights command, but what would it do if it saw a red light while in the middle of an intersection? Would it be able to separate the symbology of "don't stop within the intersection" with the signal of "stop now" and end up either inside the intersection or directly after the intersection obediently waiting for another light to show up within its field of vision to tell it to continue?

The fact that he's used the process to create controllers for robots does not mean, to me at least, that there is the beginning of awareness forming at this stage; the controllers are more likely computed as opposed to rationalized; I suppose you could say that genetic algorithms are a form of rationalization, but feel free to show me how this is the case.

This is nonetheless an amazing break through in that we have what looks to be an "emergent entity" that responds to commands and the creator is baffled about how this happened. I'm hesitant to use the word "evolved entity" at this point because I don't see an imperative drive for survival for the sake of survival in the same way members of the five kingdoms display. I think the distinction is relevant given the nature of what has emerged and how it emerged (through genetic programming which uses the concept of evolution as a schema to describe the process accurately).

mercurio
I think you misunderstood the significance of this research. Nobody is claiming that the circuit is alive or intelligent. The method used is a fairly standard genetic algorithm. The fascinating part is that the evolved circuit is unlike anything designed by humans. For example: "A further five cells appeared to serve no logical purpose at all — there was no route of connections by which they could influence the output. And yet if he disconnected them, the circuit stopped working."
fiaz
Perhaps you're right about my misinterpretation... the overall implication I took from it was that a genetic algorithm was applied to a circuit that could be rewired itself; this increased the level of complexity possible and from this (and this is where I sensed the implication that we have something akin to a biological system that is living) we have the result of something that is autonomous and sensing in a way that it can interact with humans the way a dog might (as highlighted in the first paragraph). Also the use of the word "creatures" in the title sort of threw me off...

There's much I don't know about genetic algorithms; perhaps more background in that area would help. I thought that the benefit of using genetic algorithms was to create novel designs that are unlike anything a human being would normally create.

fiaz
Oh, one more thing...

I would like to add that the first example of telling the circuitry "go" and "stop" implies some processing on behalf of the circuitry such that it has an understanding of the meanings behind these symbols. This conveyed to me in some way that the author of the article was indicating that the circuitry had an understanding of these words (hence my lengthy discussion of symbol vs signal).

ibsulon
This was really cool over a decade ago, a year after Adrian Thompson received his thesis for this work. (I mean, it's still cool, though there's a 11 years of research between then and now.)

http://www.cogs.susx.ac.uk/users/adrianth/ade.html has some of his papers written more recently.