4. Future Plans:

We are proposing here an ambitious computer design project where neither the hardware nor the software is fully determined so they have to be developed together.

4.1. Preliminary Software Development.

We suggest the most promising software project might lie at a level somewhere between the low-level Network of Networks attractor networks and associatively linked networks. Therefore we suggest initial cognitive application software might use the semantic nets with integrated symbolic and perceptual components in the expanded nodes that we described above. The PI and collaborators have worked with closely related parallel models (neural nets) and their cognitive applications for decades. Preliminary software development is underway but needs to be enhanced.

Significant amounts of hardware oriented simulator software will have to be written to assist the hardware development.

4.2. Preliminary Hardware Development.

Parallel systems of the type we propose have the unfortunate property of needing to be large to work well. But it is still possible to suggest a viable path for hardware development that, with sufficient funding, will lead to a cortex power machine, to work in conjunction with the software effort.

• Phase 1. Implement a small version of the architecture using off-the-shelf FPGA development boards. This will be mainly a software effort where we will try to figure out how to use the FPGA board and the controlling computer. Some preliminary support for hardware is requested for this phase.

• Phase 2. Next, build a medium sized version of the Phase 1 architecture. At this level we fabricate our own hardware with special attention to miniaturization, packaging and power consumption.

• Phase 3. Last, scale-up the results of Phase 2 by a factor of 100. At this point, we would use an external vendor to fabricate the hardware and we do assembly and testing.

The time course and ambitiousness of hardware development will depend largely on level of support. Phases 2 and 3 are expensive. Phase 1 could get underway with only Brown support.

4.3. Project Governance and Academic Activities.

Our initial Ersatz Brain Project team has met weekly for over a year. We have recruited several new members in recent months, including one current part time graduate student and one recent Brown Ph.D. who are currently employed in the local software industry, working on problems with a strong cognitive science component.

If funded, we plan to continue and expand our meetings, attempt to find appropriate student projects, and write lots and lots of proposals to both governmental and private sources.

In spring term, 2003-4 we held a seminar course (CG 186 Section 1, Foundations of Brain-Like Computing) that explored some of the issues involved in this project in a more traditional academic forum.

Our major task with Brown support will be to write grant proposals. Potential government funding sources have been identified at DARPA and at NSF. Proposals to these agencies are being planned and written at this time. In the private sector, some communication has been held with Microsoft and IBM, as well with other companies in the southern New England area.