One of my favourite games from my childhood was Creatures. I still remember buying it in a large cardboard box and finding the small 3.5″ floppy disk which nestled inside. This was my first experience of two new technologies: virtual worlds and alife.
The game featured little furry creatures (they reminded me of Ewoks, from Star Wars) called Norns. These creatures hatch from eggs. The installation disk, as far as I recall, hosts the simulation environment and the encoding for your first Norn egg. The basic idea was that your Norn would hatch and was supposed to have certain behavioural traits which would make it unique compared to those of your friends. The baby Norn would then wander around the virtual world (a side-on platform type world, with very rich background graphics) and try to interact with a number of the items littering the environment.
The task of the player, was to raise the Norn. I had to feed him when he was hungry. But if I overfed him he became less well behaved. I could also use a literal stick and carrot approach to teaching him behaviour. They even had a rudimentary language capacity. Eventually your Norn could lay an egg and reproduce, giving rise to subsequent generations each of which had their own behaviour part of which was genetic and part of which was due to their upbringing.
A few years ago I read the book, Creation: Life and how to make it, by Steve Grand. Steve was the creator of Creatures. In this his first book, he outlines the technology behind the game. Apart from being a well written book, I was amazed to learn of the level of technological (biological) detail which lay behind the Norns.
To my mind, Steve must be one of the pre-eminent computational people in the world who succeed at synthesising biological knowledge and applying it to real-world problems. I mentioned obliquely the difference between reductionist and constructionist (synthesis) approaches in biology in a recent article Mathematics and Biology. Science is naturally reductionist. This is how it should be, it’s more or less the scientific method. But Mathematics is rarely reductionist. And, if you ever want to build anything or make generalisations out of specific scientific studies then you eventually need to begin synthesising again.
In his book, Steve reveals how the simulated brains of the Norns are implemented as simple artificial neuronal networks. They use these brains largely for language processing. The hatching process is based on genetic algorithms, another topic which I love, giving the creatures genetic traits which influence their abilities and metabolisms. Basically, in the game, Steve combined all that was known about artificial life in the early 1990’s and crammed it into a game which could run on a home computer (I’m not sure, but I think that was on the first Pentium processor in our house).
About 2 years ago I found out that Steve has produces other work since Creatures. I think the discovery was as a result of explaining the original game to new friends and googling it to find some images. Apart from developing follow-ups to Creatures, in an attempt at making a development studio, Steve moved on to the world of robotics. He created an android called Lucy, which he describes in his book Growing up with Lucy.
I have to confess that this book is somewhat more rushed and not as easy a read. He is somewhat open about this in the text itself. He seems to have gotten a grant to develop Lucy and ran out of time towards the end. But if you have an interest in simulated biology, robotics, or learning systems then the book is well worth a read.
Lucy is a small android who’s brain (simulated) is designed around some of the most advanced concepts in neuroscience today. I write that in 2018, knowing that he carried out the work in the early 2000’s. The design chimes with my own instincts that Generalised AI, as we see it in Humans, is really the co-operation between multiple specialised modules and the tailored routing of information between them. (Maybe I’ll explain some day why I think that this means that a singularity is unlikely to occur, particularly using current approaches.)
Lucy has an auditory cortex, a visual cortex, speech producing centres and a motor cortex. The design of each of these regions is largely still mirrored by what we know about biology today; nothing major has changed in our understanding. But, it is unusual to find even a neuroscientist with such detailed knowledge of each of these disparate systems.
Admittedly, the achievements of Lucy in learning are somewhat limited by the 12 month limit on development and testing time. But, frankly, they are still astounding. I can only imagine what it would mean if some modern hardware and a few minor software upgrades were applied to Lucy today. Sadly, this work does not seem to have made any impact in the circles of Computational Neuroscience which I inhabited for so long.
According to his books, Steve’s main occupation is as a teacher. He seems to be a pretty regular guy, of his generation, who was born a little to early to take up a ‘proper’ career as a computer scientist. To my mind, though, he has had more impact than any number of scientists. His work is directly in keeping, both in line of thought and in terms of when it was done, with that of many of the founders of the Santa Fe Institute where they study Complex and Adaptive Systems.
I don’t find much from Steve online nowadays. But his project Grandroids showed its last update in 2017 which, let’s face it, is not that long ago in terms of most people’s blog updates.
Mostly, I am already very happy with the influence that Steve Grand has had on my life. His work exposed me to key advanced technologies very early in my own development. And, his ideas about intelligence needing to be learned rather than pre-programmed agree with my own. But I would also be very happy to hear any updates, on what he has worked on since Lucy, from anybody out there.