![]()
Visiting guest. Why not sign in?
|
|
|
Interview with OpenAI FoundersThe Artificial Intelligence Depot has had the opportunity to catch up with both co-founders of the OpenAI project, known as gleo and thornhalo. Alex Champandard asks them both questions about the history of the project, the status of development, and their plans for the future. Alex Champandard: What was the initial incentive to start the project? Gleo & Thornhalo: The short answer is we were bored. :) AC: How many people have worked and are currently working on the project? How much time does OpenAI take up in your week on average? GT: Well, it depends on how you count. Currently there are 3 active developers officially on the project. However, we have had code submitted by two or three other individuals recently which has been a real help in getting some of the fringe work done. In OpenAI's history there have been roughly ten to fifteen people who have contributed code to the project. OpenAI has been hosted on SourceForge since February 25, 2001. AC: Do you ever plan to do this professionally; more to the point, if the opportunity came up, would you seize it? GT: Absolutely. As long as it didn't interfere with the development of OpenAI itself in the public forum. The ideal situation would be if we were funded for OpenAI itself and could develop it to its full potential. AC: I've noticed all the code is in Java. What was the primary reason for using this language? Are there any plans for creating a multi-language library (including C/C++)? GT: We had planned to have a C/C++ version from the beginning and even have a port of the Neural Network and the beginnings of a Genetic Algorithm in C++ but there are very few people who are both qualified and willing to work with the language that we have since all but abandoned that part of the project. AC: You're entire work is released in the public domain under the BSD license, which is one of the least restrictive forms of releasing source code. What was the motivation behind this decision? GT: We don't believe in restrictions. Not to mention if we ever want to use this ourselves for a commercial venture, this is the best option. Of course, that means than any one else is free to do the same. Development of Artificial Intelligence is far more important to us than locking down our code. Personally, I'd like to see the OpenAI community grow in breadth and depth and don't want anything to hinder that. AC: An important part of your work so far has focused on fundamental AI techniques such as Neural Networks, Genetic Algorithms and Finite State Machine. What were the reasons for using these approaches? GT: I think there are three major reasons we started this way. First, we felt there was a need for a comprehensive AI developer's toolkit. Second, it was a way to gain experience in managing and developing an Open Source project of this scale. Third, when we're done we'll have a nice framework to build other technologies. AC: What are the major design criteria involved when developing such "low-level" components? How difficult is it to balance efficiency with flexibility and reliability? GT: There are probably thousands of implementations of each of the base components that we're building. I'd wager that only a few of these are reusable and
even fewer are interoperable. With OpenAI, we'd like to develop a full suite of (eventually) interoperable AI components as its base and see how we can
mix them up to develop something new an innovative. AC: There's a recent screenshot on the site revealing the impressive GUI for the Neural Network module. How important do you consider such a tool, and are interfaces also planned for the other modules? GT: The screenshot is due to the collaborative efforts of Thomas Weber and Peter Hanson and is considered a very important visual tool. It helps people to understand the technology behind the scenes and allows people of varying technical skill to research artificial intelligence. Most importantly, it allows the researcher and developer to visualize what is going on in the network and provides a convenient way of creating and testing a network before integrating it into an application. AC: Together with this, you're also aiming to simplify the creation of agents for games (with your mobile agent library). How is the development on this part of the system guided? GT: Well, the agent system is not intended for games specifically. It's a mobile agent framework that would allow a developer to create any
application that can benefit from distributed computing. AC: I've noticed you released a brand new 'hybrid' package, aiming to bring everything together. Is this a long term goal of OpenAI? GT: I think that combining different aspects of AI can create powerful new tools. 'Hybrid' may not be the best term, but it works for now. Combining such things as neural networks and genetic algorithms you can get amazing results. In the NN-GA case, you can evolve behavior in the absence of a training set. AC: Do you have any plans to produce a 'technology demo' with your work, aiming to publicise the library and its capabilities? GT: Absolutely. We have talked about creating simple applets that would help people visualize the technology and also encourage use of our code by individuals and companies. Some of the benefits of Open Source is you have access to an extremely large developer and testing base and we're anxious to see OpenAI be put to use by third parties. The more people who use and critique our project the better. Of course, the code is free and all we ask is that you let us know what you're using it for and porovide us with as much feedback as you can. AC: I've noticed you're looking for a good logo. How's that going on? GT: We have several submissions already, but more are welcome. We have some good ones, some not-so-good ones and even a few funny ones. We're planning on putting together a page where people can vote on their favorites some time in the near future. In the mean time, people can check out the current submissions on our website. AC: Your web-page mentions you're looking for help. I'm sure there are many people who would love to give you a hand. What type of "candidates" are you looking for? GT: Not to be rude, but we have alot of people come by that are either just learning to code or just learning about AI. This is not what we are looking for. We need some heavy hitters who both know AI and can code. We are also looking for people who can put together documents of all types and people with web development experience. AC: Well, thanks a lot for your time guys; I know you're busy people! Be sure to check out the OpenAI webpage for the latest source code, and lots of other information to get you involved. You can also chat with the people involved in the project using IRC, the #openai channel on server irc.openprojects.net Remember you can visit the Message Store to discuss this interview. Comments are always welcome!. Back to the Artificial Intelligence Depot.
|