IN BRIEF
name Corrin Lakeland
location Auckland, New Zealand
Email lakeland@acm.org
job Technical Manager
Speciality Data-based insight
Profiling and Data Visualisation
Phone +64 21 467784
Hello.
Doesn't it seem strange when we are trying to decide what dishwasher to buy, the computer cannot tell us which would be best based on the experiences of previous customers? That KPIs are frequently modelled as the amount we achieved last year plus a fudge factor? That marketers still view their customers as segments rather than people. And finally, that a skilled eight year old is better at the board game Go than the best computer, despite decades of research.
My area of interest is making sense of data and in this site I'll look at some of the problems that interest me as well as what I think should be done about them. There is little cohesion between the different topics except that they all interest me. Broadly speaking, my work is in carefully representing problems so that the computer can solve it for itself. If language is presented in carefully selected chunks then I believe a computer program can gradually learn the underlying structure. That if the right algorithm is shown the purchasing decisions of thousands of shoppers then it can reliably tell us what we would like to buy, without having to know anything else about the products.
So what specifically do I believe we should be doing differently? Principally, I think we should be paying more attention to some of the brilliant algorithms being developed inside the data mining community, and being much more creative about other fields where they can be applied. For instance, we have developed some amazing algorithms for spam detection, and yet emails to technical support at companies are processed by a human or by simple keyword matching. Why? We went past keyword matching in spam detections several generations ago! Along a similar line of reasoning, the ability of statistical parsers to disambiguate sentences is really quite spectacular; the standard of the voice recognition in the top systems is also spectacular, yet for some reason they do not use a statistical parser to help with their disambiguation; and also when we ring up the bank, we don't get the computer saying "How can I help you today?" but instead "Press 1 to withdraw money". Why? The technology is there already, but it isn't used outside research labs.
I guess what it comes down to is that I like solving other people's problems. I'm not that interested in designing new algorithms (though I never turn down a new tool). I think we already have the tools we need to assist in many day-to-day problems, and yet the solutions that currently have are mediocre at best. I think this is because the people writing systems such as the bank's do not have the tools at their fingertips and are not yet forced through competition to get the tools. I think it is time that more ordinary companies, government departments, and people started using the best tools available.