From Creatures to Androids
May 22, 1998
After 4 years in development, and more than 20 years of previous research the CyberLife team was able to unleash Creatures to the general public on November 11th 1996. Described by Douglas Adams as "more exciting than the discovery of life on Mars" Creatures featured the first practical implementation of biologically modelled and genetically specified neural networks, in an entertainment product. Published by GT Interactive the title was a world wide success and has lead to the upcoming release of Creatures 2.
At the forefront of artificial intelligence, not only in computer games, but in all applications, CyberLife's stated intention is to become the world leader in human-like intelligence and commercially available androids. Current collaborations with the military and other industries has all ready seen them take their first steps along this road. Toby Simpson is the Creative Director for Creatures products at CyberLife and has been the producer for both Creatures and Creatures 2. Toby is well known to many within the games industry for his numerous appearances at major computer game conferences, including CGDC and Develop!.
What were your intentions when creating the first Creatures title? Was it simply an application for the AI technology you were creating, or was it conceived in parallel? To put it bluntly why is it that you do the things that you do?
AI of some description is a critical part of the majority of modern entertainment products, and its quality varies dramatically. Most applications of textbook AI falls over because of the complexity of the problem itself, the result of which is usually a massive knowledge system, of some description, that can only possibly cope with the problems for which it is programmed. With the tendency towards massive virtual reality based products (of which I count Quake as a step along the way) and shared spaces, the ability to create artificial agents that have plausible, self consistent, and believable behavioral responses will be mandatory. Conventional AI is unlikely to yield this. Furthermore, the complexity of emulating a suitable environment in which this entertainment can take place is well beyond current software technology. Our feeling was that the most practical way of solving both of these problems was to copy nature, complete with its in-built complexity management, and a`:h
We felt that given the success of the technology in Creatures, truly adaptive artificial organisms had applications far more reaching. Likewise, as the technology improves and we are able to construct increasingly more realistic organisms, we head towards the holy grail of computing - true artificial intelligence.
How big was the development team for Creatures? Did you have two separate teams, one for the AI, and one for the game, or was everyone pretty much mixing and matching? Who were some of the other key team members and what were their responsibilities?
A surprisingly difficult question to answer! The resource concentration varied dramatically over a period of several years. The entire product is based around the CyberLife philosophy, so everyone - regardless of what they were doing - is involved in the technology. We employ people to use their brains and think - pressing buttons on a keyboard is a necessary side effect. Everyone is encouraged to think of new ways to improve the systems we have. The Creatures 2 development team consists of a lead programmer, two programmers, a lead agent engineer, an agent engineer, two genetic engineers, myself and the majority of our art department. We're also regularly making use of our research and development department and finding ways to integrate their on-going research into new brain models back into product.
How long did it take you to get the AI into a useful and impressive state, and at what point in this cycle did the development of Creatures start? How long would you say the game took from beginning to end, and how is the being shortened for the sequel?
The brain technology was evolved over a period of many years. Work on the underlying systems that eventually made up Creatures started back in 1993, although the genetically specified brain model we used in Creatures was not developed until early 1995. Development of C2 started Q3 1997 and will be completed (on time) early Q3 this year.
At what point did you feel that AI should be the main business of your company, and why?
The decision to make it the mainstay of our business was taken a few years back, and since then the technology has advanced considerably. The infrastructure that allows shared spaces where potentially millions of people can interact together continues to be laid down. We firmly believe that the approach of "modelling reality to get reality" is the only practical way of generating and populating virtual worlds so real that someday you will not be able to tell the difference.
Clearly you see the Internet as being very important to your titles, but thus far it has only been used as a medium for Creatures fans to communicate. Do you have plans to use it as a more integral part of a future title? Surely advanced AI is the natural enemy, or at least alternative, of the Internet? After all, going by many online users you'd be forgiven for mistaken their, usually stock, responses for a cheap AI!
The Internet will be very important for us in the future. Our distributed biological architecture is ideal for solving the incredible complexity problems encountered when considering massive shared spaces. We actually presented a white paper about this at CGDC last year, which you can read in the Documents section at www.cyberlife.co.uk.
CyberLife: Toby Simpson is in the middle-front.
What about 'proper' games. Have you any plans for making a more traditional action or strategy title featuring CyberLife AI? Would you ever 'rent out' your AI to games developed by other companies. After all you have no track record in action titles, so a partnership with another established games developer would seem a possibility?
We have no plans to "rent out" our AI, mostly because it is a way of thinking rather than a bolt-on black box. Software has to be designed and developed in a different way, and the traditional approach does not lend itself to having CyberLife technology inserted as an afterthought. The term "proper games" is an interesting one, as proper games in the current sense are not going to be around forever. Once you can create massive realities and populate them with hundreds of thousands of real and artificial people, you're talking some serious entertainment. However, we are working on Creatures products that use the systems we have in a more "traditional entertainment" sort of way.
What do you see as the big successes and failures of Creatures? Do you feel you got the balance right between accessibility and gameplay, and the technological accomplishments of the AI? What are the main areas that you are looking to improve in Creatures 2?
Creatures was definitely a tough one, as there was so much invention involved. It was hard to know how users would receive the product, and how they would interact with it. The reality was both surprising and gratifying. Almost out of nowhere, a huge Internet community appeared. We now know of over 250 web sites, there is a newsgroup (alt.games.creatures), and our own web site (www.creatures.co.uk) takes about 500,000 hits a week. The product itself has spawned a book (www.sybex.com), a CDROM add-on called the Life Kit (now available to buy on-line) and a Genetics editor. Considering the complex and specialised nature of the Genetics Editor, we were somewhat surprised when we sold well over 1000 copies over our web site in under three months.
The Creatures architecture is highly expandable. The engine itself knows nothing about any of the agents, creatures or environment, meaning that it is possible to add or remove new systems without having to change the program in any way. This led to a number of enterprising players taking apart our genetic code and in some cases studying it in enormous detail. As well as thousands of Creatures on the web you can download (some genetically "doctored", some natural), there are thousands of new agents you can get - from toys and plants to teaching machines and flying machines. We've also released many free object packs ourselves.
Perhaps the biggest luxury for us was being able to watch what our customers liked, and then "retro-fix" things by releasing new agents and Creatures. We were able to learn a considerable amount from observation and seeing what others were doing about the possibilities for Creatures. For Creatures 2, we were really able to "go to town" on new functionality in the product, knowing the kinds of areas that required attention, as well as being able to use the very latest artificial organism technology that we have developed. The primary things that are different in C2 are: Complete working eco-system, including weather, seasons and a functioning food-chain. The plants seed and spread around the environment if conditions allow. The result of this is a very "real" feeling world that adapts around the player. In addition, the world is completely new and twice the size of the Creatures environment. We're also displaying it in 16 bit colour which makes a dramatic difference in presentation quality.
New brain technology and creature physiology improvements. We believe that the new brain structures have the potential to double the perceived intelligence of the Norns. We've also added genetically specified organs, so that their actual internal makeup is increasingly more biologically plausible. We found quite early on that although we didn't fully understand some of the "why's" behind some of the behaviour that emerged, the more detailed we made their internals, the more real their behaviour seemed to become. The introduction of organs also helps with identifying cause of illness or death - as you can see that it was "kidney failure", or whatever. A Creatures 2 genome has about twice the number of genes as those in Creatures 1. Creatures are also now able to process considerably more sensory information, and are thus able to form more complex relationships between cause and effect. The Norns themselves now have more facial expressions, and can communicate their primary drives - reducing the gap between the user and their creatures. Norns in C2 can now use sentences such as "Nicky very hungry, get food".
There is also an all new user interface and applet design. One of the criticisms we had was that Creatures could be very difficult to get into initially, and that it could take as long as an hour or two for the product to "click" with people and start making sense. We wanted to reduce this gap and allow for more "instant gratification" so that we could broaden the product appeal. Most of this comes from the improved Norn intelligence and communication. In addition, though, we've made the user interface more intuitive, and staggered the rate at which we unravel the more complex features of the product so it is not too overwhelming. We have added lots of groovy new toys and other agents as well. From the genetic splicing machine to a tomato gun, with numerous plants and animals that really grow and seed. It is quite enchanting to watch the butterflies and other insects rush around looking for flowers. The aim has been to make it more fun.
Do you still intend CyberLife to be the "Hoover of androids"? If so presumably you don't see yourself as a games developer at all? What are the companies short and long term goals, and what partners are you trying to court in order to achieve this? Wasn't there some mention of working with the MOD ([UK] Ministry Of Defence)?
That is right, we are actually doing some work for the MOD. It is in the field of UAVs (Unmanned Air Vehicles) and explores the possibilities of putting artificial organisms into UAVs as pilots. Our long term goal is "to put human intelligence inside computers by the year 2020." Astonishingly enough, we call that our 20-20 vision. In the meanwhile, we are working on projects that take us closer towards that goal. Creatures is an excellent example, as we get to put this technology into consumer's hands, and see the results of this quite quickly. There have been millions of Norns out there now! Creatures also acts as a vehicle for us to develop the realism of the environment and organisms themselves, as well as improving our CyberLife based eco-system stuff. Some time in the next three years, we'll have a 3D version of this technology - which is really exciting.
So was the androids bit just a metaphor, or are you really planning to get us R2-D2 by the year 2020? If so, presumably you're going to need some pretty heavy weight partners, in the hardware business at least?
Who knows! Once we have the power to produce human level intelligence inside a machine, the chance of putting that into an autonomous robot unit will be far too tempting to resist. We're already experimenting with many autonomous hardware units powered by CyberLife brains, partially in connection with our defence work.
What do you think of the standard of AI in computer games, and in other products in general? Where do Tamagotchi fit into all of this - the craze seemed to hit the UK at about the same time as Creatures, and managed to appear both very similar and incredibly different, all at the same time. Do you see Tamagotchi as a fad, or an indication of an industry to be?
For my own safety, I'll avoid risking idle speculation on the future of everyone else! My view on this is that comparing Tamagotchi to Creatures is a bit like comparing a microwave to a refrigerator. They're both appliances in the kitchen, but that is where the similarity ends. Their target market was very different to ours, and they made no attempt to actually create life. As I've mentioned, I feel that AI in computer games varies dramatically between "ok" and "awful". I don't think I've seen a product where I can sit down and say "yes, that Artificial Intelligence was fantastic", although I'm sure I will be proved wrong some time!
One would assume that Bandai, and whomever, will see true artificial life as a logical next step for the Tamagotchi style product, and could conceivably become competitors in your market. Will you in turn be attempting to create a product aimed specifically at the Tamagotchi market, or are you going to leave all of that to Bandai and the vagaries of fashion?
Complex Artificial Life in consumer products like Tamagotchi is probably a long way away, mostly because of the CPU power required to model biological structures accurately enough to get the required emergent behaviour. Likewise, research and development takes time - it is difficult to just throw money at it and produce miracles. Let's face it - it takes a woman 9 months to have a baby. Assigning another woman to the task isn't going to make the baby appear any faster.
Taking the science of AI as a whole who do you see as your main competitors now and in the future? Who at the moment do you feel is doing comparable work, and how do you plan to stay one step ahead of them?
There are lots of organisations doing work that is similar in parts to some of the stuff we are doing. However, as far as we are aware, we're the only people creating artificial organisms based around genetically specified biologically inspired structures - and using them in industrial, defence and entertainment applications. These organisms are going to become more and more intelligent as the months and years roll on, and as a result of which their range of potential applications will widen.
What other applications are you planning for CyberLife, beyond games and the Creatures franchise? Do you foresee a point were you will out-evolve the games industry and multimedia products, or will these always be a part of your core business?
Whilst Creatures continues to be a huge success, we will continue to develop products along those lines. To have the chance to develop detailed environments, advanced creatures, and then get them into hundreds of thousands of people's hands is an incredible opportunity for us to forward the development of the technology. Working with reality is what we, as humans, are really good at - we already know how to deal with it. Our trick is just to put that reality into a machine, and capitalise on the knowledge and experience the user already has as a result of their real life. Creatures products of the near future (within the next decade) could be enormously intricate evolving 3D worlds where hundreds of people can get together, and where the artificial players will be able to pass "informal Turing tests".
It's an obvious question really, and Hollywood's answered it many times before, but what sort of dangers do you see in true human-like artificial intelligence? It's hard to believe, but presumably you're reaching the stage where this becomes a serious question. Will we be seeing the three laws of Robotics, or other similar safeguards, at some point?
We'll deal with this when it happens! People tend to be frightened by things that they do not understand - it was not that long ago that many people thought that you couldn't travel faster than 30mph because all the air would rush out and you'd suffocate. It is very easy, thanks to Hollywood, to assume that any intelligent machine would be dangerous. I believe that this is a very short-sighted view, as there is no reason to believe that an artificial intelligence would be as vicious and violent as we are, and furthermore, the potential improvement to our quality of life is immeasurable.
Isn't that what the mad scientist guy always says in the movies - right before he gets his spleen ripped out by a rampaging robot? A more serious question and pointed question then: Who'd win in a stand up fight between SkyNET and HAL 9000?
SkyNET (and that is a shame, as HAL is a nicer person), unless HAL could figure a neat way of reassembling himself from a zillion pieces of stuff around Jupiter!
Toby Simpson, thank you very much.