- From: Dave Raggett <dsr@w3.org>
- Date: Mon, 13 Nov 2023 11:11:12 +0100
- To: Timothy Holborn <timothy.holborn@gmail.com>
- Cc: public-cogai <public-cogai@w3.org>
- Message-Id: <C36657D1-8FC5-4061-B241-CBA8FB800B71@w3.org>
> On 13 Nov 2023, at 10:37, Timothy Holborn <timothy.holborn@gmail.com> wrote: > > Hi Dave, > > I wondered if you had any thoughts on the potential application of your CogAI work to create a semantic/CogAI filesystem? Yes, and in conjunction with large language models. I see the combination (something I refer to as cognitive databases) as an evolutionary replacement for today’s graph databases, and paving the way to zero-code applications. I see this as enabling collaborative knowledge engineering (see my talk [1]) where the computer and human play complementary roles. The human can ask the computer to perform an analysis of the data, leaving the details to the computer to figure out for itself. The huge hype around generative AI has drawn attention away from the limitations of the current approaches to large language models. I see many opportunities for evolving artificial neural networks inspired by what we know about human cognition. This will enable computers to become very effective partners for human-machine collaborative work. Neural networks and vector spaces are very powerful for representing complex statistical relationships, as a generalisation of symbolic knowledge, and learned and queried by machine. This changes the conception of databases considerably. The impact of this will be dramatic. [1] http://www.w3.org/2023/10/10-Raggett-AI.pdf > > I was looking at the solid code again recently, and was thinking about the implementation structure. > > I haven't done much more thinking about it yet, but thought I'd ask you just in case. > > Whilst my plans for human centric AI stuff, particularly my thoughts on my own implementation, may not be solid, although I want compatibility / backwards compatibility... > > It seemed to me that there's an ability to make distinctions between different agents using multiple "pods" via methods that assume domain ownership. > > Without getting into it further, the basic thought was about semantic file systems which then led to wondering about CogAI file systems... How that might work? > > Tim.H. > > > Dave Raggett <dsr@w3.org>
Received on Monday, 13 November 2023 10:11:27 UTC