You are currently browsing the daily archive for April 2nd, 2008.
Well, after an incredibly long review process one of my papers has been accepted for publication in the “Journal of Experimental and Theoretical Artificial Intelligence.” The paper’s basically an experimental investigation into some of the cognitive philosophies underlying artificial intelligence. At present there are a whole host of thought experiments considering various fundamental questions: To what extent our are concepts and beliefs innate? How do aspects of the environment acquire meaning? Can human reasoning be captured by mere symbol shuffling?
Instead of adding my own speculations I decided to put these questions to the test by building agents with a cognitive architecture based on one of three competing approaches: symbol grounding, symbol attachment and enactivism. I was able to compare their performance in fulfilling various tasks in a complex environment. The results of this little competition certainly don’t resolve these deep philosophical issues but I like to think that they highlight a different approach to this thorny debate.
An idea that no doubt contains fatal flaws but why isn’t there greater public engagement with the sciences? After all the tax payer funds our research so we should probably tell them what we get up to. I keep returning to the genuine surprise that I encountered at Bristol, to many the idea of meeting a scientist in a science museum seemed shocking almost absurd. It’s no wonder we’re often perceived as some sort of isolated elite emerging from our ivory towers to make arcane pronouncements from authority, like some sort of modern day priesthood. Surely the onus is on the universities to rectify this situation.
I think a university museum approach would also benefit the wider research community. I could imagine even a small ‘museum’ acting as a summary page of current research. Bringing together ideas from across the campus and allowing visiting or isolated academics to quickly identify potential interdisciplinary research opportunities. The preparation and presentation of various displays and exhibits can also sharpen research teams thinking on their subject and open them up to thousands of potentially useful, potentially critical, viewpoints. Ultimately, by providing a university ’shop front’ we’re saying to the public “come in, this is what we do and it’s not as scary or as weird as you think.”
Well my time at Evolang wound up with my own presentation on this paper. Returning to one of my favourite themes on the necessity and problems of abstraction within artificial intelligence. (Until we can simulate every molecule of every neuron – and then embody that simulation within reality- abstraction will always be with us). I considered Edingburgh’s Iterated Learning Model – an interesting simulation of the cultural evolution of language – and sought to remove their mind reading abstractions.
After a few late nights putting in extra slides and a few dress rehearsals to tackle my nerves I think I put forward a reasonable case for success. So, feel free to have a look at the paper and let me know what you think.
I know I was as shocked by this revelation as the next guy! But when you think about it it’s quite an interesting observation, during conversations you can get a sense that people aren’t listening they’re just waiting for their turn to speak. And when they get there chance they frequently tell surprising or emotional stories that contain little factual information and insignificant survival benefit.
In fact, according to Jean-Louis Dessalles from the ENST in France, roughly 45% of our conversations are exercises in storytelling. Why? What possible Darwinian motivation could we have for doing this? As Jean-Louis says “If language has been selected because of its effect on the welfare of the group or as a fair exchange of information based on strict reciprocity, then the efforts that speakers devote to tell stories for all to hear, most often with much emphasis to highlight interest, is incomprehensible.” We’re fascinated by unexpectedness and by highlighting or recounting unexpected events we gain the attention of others. One theory is that this makes language a form of display. As, the advent of the spear made strength (a common feature of display amongst animals) an increasing irrelevance, a weakling could kill a hulk simply by taking them by surprise. In this new political landscape the best form of defence was to surround yourself with allies who were able to spot the unusual and anticipate a spear in the gullet. So, ‘How to attract friends and influence people?’, simple display your alertness by spotting the unexpected and being the first to tell people about it!
It’s hard to overestimate the impact genetics has had on our understanding of evolution. But one nice example comes from the history of the FOX2P gene and its role in the evolution of language. It’s a nice example because it illustrates both the complexity and utility of genetics. In the popular press FOX2P has been heralded as the ‘language gene’. But this is a simplification, the path from the genome to the person isn’t straightforward and it is influenced by multiple interacting factors (other genes, environmental and developmental processes to name a few). So is FOX2P the ‘language gene’, I’m afraid not, it’s certainly necessary for speech and it has a role to play in our fine motor control but the story behind the evolution of language hasn’t been solved quite yet and there are many, many more factors to consider. So, next time you read about the ‘X’ gene, whether the crime gene or the gay gene, take it with a massive pinch of salt.
“More people have been to Russia than I have”"
The above is a syntatic illusion, to most people it seems perfectly fine but in reality it would result in a big minus mark from your English teacher. One of the problems that has been highlighted at this years Evolang conference is that many researchers only consider language that has been written by the editor of the Times. As a result they construct perfect syntax trees, describing the structure of a sentence, and then act surprised when they don’t carry over into reality.
One alternative, proposed by Gary Marcus from New York University, is that we construct mental “treelets” containing limited structural information that is strung together “on the fly” using our inherent associative memory, allowing us to make sense of the language while still being fooled by things like the first sentence.
