I sent the following R&D report to my colleagues, but a few other folks outside Galois have expressed interest in the project, so I’m sharing it more broadly.
Subject: Max: Phase 1 Report
As some of you know, about nine months ago, I started a skunk-works R&D project with Brooke. Autonomous systems are all the rage these days, so we decided to try to create one. First, I have to be honest and say that although I participated in the initial project kickoff, I mostly played a supporting role after that. Brooke did all the major software and hardware development. (If you’ve worked with me on a project, this should sound pretty familiar.) Once Brooke started development, she really threw herself into it. She seemed to be working on things day and night, and it even looked a bit painful at times. She sensed she was getting close to an alpha release a few days ago, and after a final four hour sprint, she made the release at 2:30am on Mothers Day! We are officially out of stealth mode.
We call our project “Machine Awareness X”, or Max for short. The current system is capable of basic knowledge discovery and processing. Now, don’t get too excited; we expect at least a few years before it’s able to do something interesting with the knowledge acquired. I won’t go into the technical details, but the programming model is very biological—think “Game of Life” on steroids. At this point, we’ll have to continue to provide guidance and some rules, but its basically self-programming.
Following a “Johnny Ive” approach to design, Max has basically one notification method. It’s a fairly piercing audio signal used whenever his power supply is running low or there’s a hardware problem. (Frankly, sometimes it seems to just go off for no reason at all.) We designed it to be loud enough to hear across the house, but I wish it had a volume control. Next time! Otherwise, the package is quite attractive, in my opinion, even cute. Unfortunately, at 7lbs. 8oz., the hardware is heavier than even a decade-old laptop, and we expect new versions to require an even larger form factor. Fortunately, we designed the system to be self-propelling, although it’ll take a few years before that hardware is developed (the software isn’t ready for it anyways).
There’s still quite a bit of work to do. Our back-of-the-envelope estimate is that we’ll have to spend just short of two decades caring for Max before he’s fully autonomous. Even more disappointingly, we’re estimating having to spend up to a quarter million (in today’s dollars) in upkeep and maintenance! (Sadly, while others are interested in playing with the system intermittently, nobody seems that interested in joining us as early investors.) Despite all the training we’re planning to provide, the system seems too complicated to guarantee certain behaviors. For example, while more general than an autonomous car, it may take more than 15 years of training before his software is capable of piloting a conventional automobile.
I’m guessing some of you are wondering about commercialization opportunities. The good news: we expect Max to be useful to society (we haven’t found a killer app yet, though) and to generate quite a bit of revenue over its lifetime. The bad news: we don’t expect it to start producing reliable revenue for more than 20 years. What’s more, it has a lot of upkeep expenses that will only grow with time. This might sound like science fiction, but we imagine he might even replicate himself in the distant future, and will likely pour his revenues into his own replicants. In short, we don’t expect to make a dime from the project.
More seriously, we had a kid; mom and baby are doing fine. See you soon.