News
August 31, 2022
Demo I
First demo:
More will be posted soon.
August 16, 2020
Update
This update is long overdue: work on Attentron is continuing, with slow but steady progress being made.
The code is in C, not Python. No estimate can yet be given of a completion date, but there is reason to be optimistic.
May 30, 2019
Work done so far
When this website was started in September 2018, the following note was displayed:
Hybrid primitive, molecular-to-cellular proof of concept, first of three exhibits, will be shown here before end of Q1 2019
The idea being that the equivalent of a unicellular organism, named Critter, would be placed in a simulated 2D environment, demonstrating the emergent mind as Critter learnt to find food and avoid danger. The fact that Critter would learn and adapt in a completely generic way, with nothing programmed in except for the capacity for pain and hunger would serve as proof of the emergent mind. Given enough development time, even pain and hunger could have been abstracted away at the level of Critter’s cellular structures – hence the reference to “mollecular-to-cellular” hybrid primitive in the quote.
By the start of programming in March 2019 (a quarter late due to lack of time), it was clear that emergence of sentience in a machine is the easiest part of the project. There is irony in this insight, because if you had asked me even 2 years ago I would have insisted that sentience could never be emergent in a machine. To the contrary, not only is machine sentience possible, but its emergence is primarily a scheduling problem, dependent on the quality and “genericity” of the perceptual and cognitive components. It was for this reason that priorities changed in March and work started on the most difficult part of the whole, the visual system.
After two months of programming, the Attentron visual system is about one quarter complete. It has been deployed in a client project, with excellent results. However, Attentron was adapted to the task with parameters of shapes that had to be recognized hardcoded instead of learnt by the system. The learning capability is the next stage of development on Attentron.
Note that this project is a little different from standard AI/ML applications. Neural nets are not used and are unlikely to be. Accuracy is not a priority (though it was in the client project), but an ability to think, reflect and correct errors is the foremost objective. If a sentient slug is the result, that will be a success. Accuracy, speed and super-intelligence are a matter of optimization and relatively easy to achieve once the rest is in place.
I conclude this first update with something I like to say to anyone willing to listen: if you’re not working on Attentron, you’re wasting your time. This project is priority number one for those smart enough to understand the implications. Get on board and let’s get this completed!