I'm sorry Dave, I can't do that...

About a week ago I posted an article about our tentative dipping of our toes into the world of AI. We were trying to write an application to train a neural network using a genetic algorithm to play four-in-a-row. We’ve done that now and the results are... well let’s just say we’re not ready to enter our network in any four-in-a-row competitions. 

That’s not to say it hasn’t worked. Our network has definitely learnt the game and is playing to win. The trouble is that it doesn’t seem to play very defensively, especially in the opening moves of the game. If you let the game play beyond more than about a dozen counters on the board, then it seems to play defensively and, more often than not, it will win. In fact it will often force a position where, whatever move you play, it will win. But it’s definitely no expert.

We’re not finished yet. We have a few more things to try - and perhaps we just need to let the network evolve a little more. I don’t know when we’ll get the chance to do this but if we do, and if we have any more luck, then I’ll post an update.

NSConference: Thank you, Scotty!

We're back form NSConference - and what a conference it was! Scotty and his team put so much effort into making this happen and it really showed. The whole thing ran like clockwork. The collections of speakers was like a who's who of the Mac Development Community - Matt Legend Gemmell, Fraser Speirs, Philippe Mougin, Andre Pang, Mike Lee, Drew McCormack, Graham Lee, Bill Dudney and Marcus Zarra all delivered lively, polished, informative presentations.

Martin Pilkington did an inspirational 15 minute lunchtime session on application accessibility. He is author of The Accessible Mac column (http://www.mac-developer-network.com/columns/theaccessiblemac/march09/) and we support his pledge that "by the end of 2009, all the apps we produce will be fully accessible".

I would love to write more but one of my new friends, Ross Carter, has written a post that says it all, and with so much more style than I ever could! Please pop over to http://rosscarter.com/2009/250.html to read what he has to say.

Thanks again to Scotty and his team. And let's hope we can do it all again in 2010!

The Serious Business of the Idiot Detector (Part 2)

In Part 1 of this series I promised I'd write an update on some of our marketing experiences with the Idiot Detector. It's been on sale for a little over a month, so here goes...

We got off to a half decent start. We missed out being on the "What's New" front page in the Entertainment section on the App Store and were buried deep on page 13!! That certainly didn't help our initial sales. We were probably rescued a little by appearing on the various sites that follow App Store releases. These are not reviews, just the product info pulled off the App Store and presented for readers of those sites. For example...

http://appshopper.com/entertainment/idiot-detector

http://www.148apps.com/newest-148-app-store-additions/

Over the next couple of weeks the sales dropped off as seems the typical pattern for many iPhone apps.

We then took out an advert on 148apps.com. The ad space we opted for was shared with 8 others, but with 500,000 visits per month we thought we'd get enough traffic. And we had a decent run of sales. This graph shows the trends...

The advert provider behind 148apps.com is buysellads.com. They provide a really cool dashboard where you can collect all sorts of useful stats about your advert. It turns out that just under 1% of people who saw the advert clicked the link through to iTunes. Of these around 25% actually went on to purchase the software. That's really not a bad conversion rate!

So what's next for Dr Ivan Egghead's Amazing Idiot Detector? Well, if 25% of people who see it on iTunes are actually going to buy it - we clearly need to get more people to see it on iTunes! Work is currently underway on a cunning plan and I'll report back form the labs of Dr Ivan Egghead with part 3 of this tale next month...

Blondie25 ?

I’ve just finished reading a very interesting book called ‘blondie24’. It’s not what you might think - it’s about a couple of guys who trained a neural network to play checkers using a genetic algorithm. Inspired by their success we’ve decided to give it a go ourselves for a bit of fun. We reckon that the same technique should work with a four in a row game (Connect 4).

Back when the book was written (1995 - ish) they took 6 months to evolve a decent network. I’m kinda hoping that we might be able to run it a little quicker these days. We’ve started building the framework (in Objective-C of course) so hopefully we’ll have some initial results in week or so - we’ll keep you posted.

I’m sure that anyone else who has read this book will have had exactly the same idea, so if you’ve already tried we’d love to hear about your experiences.