The BA Festival of Science

Thanks to a generous grant from Trinity College at Cambridge University, I was able to attend the full week-long British Association for the Advancement of Science Annual Festival of Science in Leicester this year, from September 9th to 13th. Curiously enough, no-one uses the acronym BAAS while in America they do use AAAS – instead we simply call it the ‘British Association’ which no doubt causes some confusion.

Anyway, the BA Festival of Science is a week long event that can’t really be described as a conference as it doesn’t have a particularly focused nature aside from being about ‘science’ – and even that isn’t accurate, since there were plenty of lectures given outside the traditional remit of science, such as economics and philosophy. The lecture schedule consists of several parallel tracks that tend to last from half a day to a day covering distinct topics, for example, ‘Life and Space’ or ‘Radioactive waste – can we manage it?’ In addition to the lectures were debates and workshops.

This year there was quite a spread of topics such that on some days I had a very hard time trying to decide which to attend; in retrospect I think I managed a decent spread.

I originally intended to write up some of my notes made during the Festival as a series of pieces in the ‘Middling’ weblog, until I realised that I simply didn’t have the patience for that. So this article will attempt to string together my thoughts on some of the more interesting lectures I attended.

Visualisation using sound
Professor Stephen Brewster, University of Glasgow

This was a fairly interesting lecture summarising the work Brewster’s group has been doing with the MultiVis project. What they’re trying to do is to give blind people access to data visualisations, such as tables, graphs, bar charts and so on. Current methods include screen readers, speech synthesis and braille; these have the (perhaps) obvious problems of presenting data in a serial manner that is consequently slow and can overload short term memory, thus preventing quick comparisions between different pieces of data.

A good example of this is how blind people would access a table.

10 10 10 10 10 10
10 10 10 10 10 10
10 10 10 10 20 20
10 10 10 10 20 30

To access the table, item by item speech browsing would probably be used, so you can imagine a computer voice reading from left to right, ‘Ten, ten, ten, ten, ten…’ etc. This has the serious problem of being extremely slow, and currently there is no way for a blind person to get an overview of this table and importantly, be told that the interesting information is in the bottom right hand corner.

The solution? Multimodal visualisation, and in this case, sonification – that is, the use of sound other than speech. Sonification offers fast and continuous access to data that can nicely complement speech. Prof. Brewster demonstrated a sound graph, on which the y-axis is pitch and the x-axis time, so for the line y=x you would hear a note rising in pitch linearly. This worked quite well for a sine wave as well.

Multiple graphs can be compared using stereo, and an interesting result is that the intersection between graphs can be identified when the pitch of the two lines is identical. So, imagining that you are trying to examine multiple graphs, you might use parallel sonification of all graphs in order to find intersections and overall trends, and serial sonification in order to find, say, the maximum and minimum for a particular graph.

3D sound also offers possibilities for the presentation of multiple graphs; different graphs could be presented from different angles through headphones. Continuing this further, soundscapes would allow users to control access to graphs simply by moving the orientation of their head. Access by multiple users is possible, so you could have one person guiding another through the soundscape.

Such sonification aids can also be used together with tactile stimuli such as raised line graphs; by placing sensors on a user’s fingertips and connecting them to a computer, users could naturally explore a physical graph while a ‘touch melody’ would indicate (for example) the horizontal or vertical distance between their two fingers. External memory aids could be built in by allowing users to place ‘beacons’ on graphs, perhaps by tapping their fingers – as the user moves away from the beacon, the beacon sound diminishes.

Of course, sonification can also be used for sighted people.

I don’t doubt that these concepts have been explored before, but this presentation was the first I’ve encountered that has dealt with them in such a comprehensive manner and also produced practical demonstrations.

Information foraging and the ecology of the World Wide Web
Dr. Will Reader, Cardiff University

This was perhaps the most interesting Internet related lecture at the Festival of Science; I was impressed by the way Dr. Reader drew upon previous research, which is something that I think many web pundits forget to do. My notes:

Some background: information foraging occurs because people have a limited time budget in which to find answers. According to a recent survey, 31.6% of people would use the Internet to find the answer to any given question – this is the largest percentage held by any single information resource on the survey. However, if you collect together all the people who would use other people as an information resource in order to answer their question (i.e. not only friends and family, but also teachers, librarians, etc) then the humans still win.

H. A. Simone once said something along the lines of ‘Information requires attention, hence a wealth of information results in a poverty of attention. What is then needed is a way to utilise attention in the most optimal manner.’

To use a traditional metaphor, you could call humans ‘informavores’ (eaters of information). When humans read in search of an answer, we are trying to maximise the value of information we receive over the cost of the interaction.

What is meant by the value of information? The value of a text relies principally on relevance, reliability and the difficulty of understanding. Examining the latter factor in detail, it’s theorised that the amount learned from a text (or any information resource) follows a bell curve when plotted against the overlap between the person’s own knowledge, and the information in the text. So – if there is a very small overlap (i.e. almost everything in the text is new) or a very large overlap (everything in the text is already known), little is learned. When the overlap is middling, the amount learned is high.

Dr. Reader carried out an experiment to test this theory in which subjects were given a limited amount of time to read four texts about the heart (something like 15 to 30 minutes). They then had to write a summary of what they’d learned. The texts varied in difficulty, from an encyclopaedia entry to a medical journal text.

The results of the experiment showed that people were indeed adaptive in choosing which texts to spend the most time reading according to their personal knowledge on the subject; in other words, they read the texts that contained a middling amount of information overlap the most. However, the subjects did act surprisingly in one way – they spent too long reading the easiest text.

Is this a maladaptive strategy? Maybe not – it could be sensible. Given the time pressure the subjects were under, they may have simply been trying to get the ‘easy marks’ by reading the easy text.

It turns out that there are two different access strategies when reading multiple texts on a single subject (or accessing multiple information sources). There’s ‘sampling’ in which subjects choose the best text available. They do this by skim reading all of the texts quickly and then deciding on the best. It sounds easy enough, but it’s very demanding on memory if you have several texts to read. People spontaneously use the sampling strategy only 10% of the time.

The majority strategy is called ‘satisficing’ (yes, that’s the right spelling), the aim of which is to get a text that is ‘good enough’. Simply enough, a person will read the first text, and then move on if they aren’t learning enough.

All of this changes when people are presented with summaries of texts. Now, sampling is the majority strategy. These summaries, or outlines, are judged by people to be reliable clues to the content of the text – an information ‘scent’, if you will.

This begs the question, why don’t people use the first paragraph of a text as an impromptu outline? It’s because the first paragraph is not necessarily representative of the rest of the text; we all know how texts can change rapidly in difficulty, particularly in scientific journals.

Outlines can sometimes be misleading. In a study carried out by Salmoni and Payne (2002), when people use Google for searching, they can sometimes be more successful at saying whether a fact is on a given page if they do not read the two line summary/extract in each link in a search result page. This suggests that the Google extract is not as useful as we might believe.

Another experiment by Dr. Reader confirms what many of us anecdotally know. Subjects were asked to research a subject using the Internet through Google. They were given 30 minutes, and then had to write a summary afterwards. The results:

Mean unique pages viewed: 20.8
Mean page time visit: 47.6 seconds
Mean longest page time visit: 6.43 minutes

This shows that some pages were only visited for a matter of seconds, whereas others were visited by several minutes.

Dr. Reader concluded with a few suggestions for improvements to search engines. They could index the difficulty and the length (in words) of search results, and also the reliability of a page. This is already done in Google via Page Rank (essentially calculated by the number and type of pages linking to the page in question), but Dr. Reader also suggests using annotation software (like the ill-fated Third Voice) and interestingly, education. We should educate Internet users in how to quickly and accurately evaluate the reliability of a page.

All in all, an interesting lecture.

The march of the marketeers: invasive advertising and the Internet
Dr. Ian Brown, University College London

I didn’t learn much from this lecture, but that’s only because I’m very interested in the subject anyway and keep abreast of all the latest developments. However, it was a very comprehensive and up to date lecture, unlike some of the reporting you see in the mass media. One thing that I did find interesting was Dr. Brown’s claim that some digital TV channels have ‘unmeasureably small audiences’.

Since audiences are measured by sampling a few hundred or thousand people who have little monitors attached to their TVs, if no-one in the sample group watches a programme or channel, then as far as the survey company is concerned, no-one in the entire country watched it. Even for supposedly popular programmes such as the Nationwide League Football matches on ITV digital, there were zero viewers in the sample group. This is understandably causing problems with advertisers.

Dr. Brown went on to talk about Tivo and all the rest, but I’m not going to cover that.

And all the rest…

I’m giving a very skewed view of the Festival here because I only took notes on things that were completely new to me and that I felt would interest people here. Consequently, I didn’t take any notes in the space lectures I went to, even though some of them, such as ‘Living and working in space’ by Dr. Kevin Fong and the lecture given by Sir Martin Rees were excellent. The former was a very entertaining and information lecture about space medicine on long duration space missions, and the latter was all about posthumans and the Fermi Paradox.

I was actually stunned by Sir Martin’s lecture; not because of its content (I read lots of SF, thank you very much) but because it was coming from him – the Astronomer Royal, no less! In the past, such respectable people wouldn’t touch esoteric subjects like posthumans with a bargepole.

Then there was the talk on DNA nanomachines by Dr. Turberfield from Oxford University; I hadn’t quite grasped the possibilities of DNA assembly before that lecture, and neither did I truly understand how DNA computing could be used to solve a variant of the travelling salesman problem, but afterwards I did (in other words, it was a good lecture). Dr. Turberfield also showed a model of his current work in trying to construct a DNA nanomachine motor, which he confesses probably doesn’t have much immediate practical use but certainly is fun.

Most of the lectures I attended were pretty good; some were excellent, of which I’ve only mentioned a few above. If you ever find that the BA Festival is taking place nearby one year (next year it’s in Salford) then it’s probably worth getting hold of a programme and attending for a day or two. You’ll learn a lot.

Kurzweil and AI

Hah, I always knew that AI pundit Ray Kurzweil was up to no good, but this article proves it. Kurzweil is fond of making grand – and vague – predictions about the future of AI, but as far as I can see he his only major achievement that could possibly be related to AI is his voice recognition software – and Kurzweil was hardly the only pioneer in that field.

I was present at the unveiling of the Ramona prototype discussed in the article, and I was extremely underwhelmed by it. This lash-up of motion capture, voice recognition and crude AI was supposed to be a breakthrough? That Kurzweil’s estimate of the ‘virtual personality’ market being $5 billion in a mere three years turned out to be completely wrong is sadly no surprise to me now.

On a more general note, I’m glad that the article pointed out that, “absent multiple major revolutions in both computer science and neuroscience, it’s almost certain that the bold AI prognostications of today will be no more accurate than those of the past,” – the assumption that AI progress will continue to roll on ahead just like Moore’s Law exhibits a fundamental misunderstanding of the problems involved in artifical and human intelligence.

The Media Lab

In one of the slower periods at the lab, I browsed through the mini library we have here and began flipping through The Media Lab: Inventing the Future at MIT by Stewart Brand. It was absolutely fascinating reading – not because the Media Lab is an interesting place, but because the book is fifteen years old.

The book was written a little after the opening of the Media Lab, which is essentially a technology laboratory looking at the cutting edge of ‘neat computer things’ (my term). It’s amusing to consider that if you stripped the book of dates and numbers, then you’d have both a good description of the current state of technology, and also a good overview of the research the Media Lab is still conducting.

For example, there is talk of electronic books – and we’re now at the stage where they could conceivably be on the mass market within half a decade. There’s talk of interactive TV (which we have) and artificial intelligence natural language processors and parsers (which, yes, we still don’t have). Holography is featured quite heavily, and there are the usual predictions of 3D TV – which I really fail to see the point of.

Between them, Brand and the Media Lab get a lot of things right (e.g. Brand: “I’m inclined to believe that the ideal content for CD ROMs are those multivolume reference works and subscription services…” and MIT: “CD ROM is by definition an interactive medium.”) There’s a nice prediction for personal video recorders which almost exactly mirrors what we have with Tivo, and a discussion about the problems of bandwidth.

Of course, what I found most enjoyable were the predictions that were completely wrong, including the fear that not only might DATs (Digital Audio Tape) overtake CDs, but they could result in mass piracy. About email: “[In the US] if it happens by a provider, it’s going to happen when the banks develop a standard and decide it’s in their interest to pay the costs of getting the terminals out there.” And my favorite, half a gigabit is “effectively, infinite bandwidth.” If only it were so…

It seems to me that many of the problems that the Media Lab was looking at back then have been solved and exceeded, in the form of the Internet and innumerable consumer electronics devices. The problems that haven’t been solved reflect a misunderstanding on the Media Lab’s part of the complexities involved in, say, cheap and effective holography, or that old chestnut, AI.

Writing

Something popped into my head today as I was scribbling down some notes during a supervision: does the fact that I write with a pencil (as opposed to a pen or biro) affect my writing style, and on a higher level, my method of thinking?

Pencils provide a much less constrained and linear way of putting thoughts down onto paper, in that pencil marks can be easily and quickly erased. Thus, I’m not too bothered with making the occasional correction or altering what I’ve written so that it’s more accurate, whereas if I used some non-erasable implement that option wouldn’t be open to me. Conversely, perhaps using a pencil is making me lazy and those who write using pens have less cause to make corrections.

Taking this further, what about writing on the computer? Words, sentences and paragraphs can all be moved about at the click of a button, and rarely does a supervisor not warn us against getting into a habit of writing all essays on the computer, as this won’t help us write essays in exams. I tried writing an essay on paper a couple of weeks ago, and it went down perfectly fine. In fact, I probably did it faster than I would’ve done on the computer since I could draw diagrams quicker. Score one for paper.

As others have said, probably the best solution would combine the qualities of paper and computers – I imagine some kind of smart paper which you can either write on (it has handwriting recognition, naturally) or hook up to a wireless keyboard would be ideal (many people can type faster than they can write). You’d be able to annotate the paper and move sentences and words about with ease, and it’d be intuitive for all users. It’ll probably be on the market in another ten years.