The Konrad Artificial Consciousness Project: Historical Milestones
Copyright Notice: This material was written and published in Wales by Derek J. Smith (Chartered Engineer). It forms part of a multifile e-learning resource, and subject only to acknowledging Derek J. Smith's rights under international copyright law to be identified as author may be freely downloaded and printed off in single complete copies solely for the purposes of private study and/or review. Commercial exploitation rights are reserved. The remote hyperlinks have been selected for the academic appropriacy of their contents; they were free of offensive and litigious content when selected, and will be periodically checked to have remained so. Copyright © 2011-2018, Derek J. Smith.
First published online 09:00 GMT 15th March 2011. This version [2.0 - copyright] 09:00 BST 10th July 2018.
PAST VERSIONS AND PROJECT MILESTONES
The March 2008 prototype of the program [the Konrad 1 first working prototype system] delivered the following functionality .....
(1) To allow and record the construction of a long-term memory semantic network, on demand, from a suitably prepared input file.
(2) To allow and record the interrogation of said semantic network, on demand, from a suitably prepared input file.
(3) To implement an arbitrary semantic "interlingua" such that conceptual processing might be encoded semantically, and thereby separated from the phonetic encoding used in surface speaking and hearing. [For an introduction to the interlingua notion in the history of Machine Translation, see "Short-Term Memory Subtypes in Computing and Artificial Intelligence (Part 4)" (scroll to Section 4.1).]
(4) To respond in simulated sound to a simulated auditory input. [Note that the value of using an interlingua was emphasised by arranging for the input to be in English and the output in Welsh.]
(5) To demonstrate the feasibility of "linearising" the distributed content of the network into a serial "deep linguistic" form. [For an introduction to the problem of sequential read-out from what is ultimately a spatial long-term memory engram, see "Motor Hierarchy".]
(6) To produce a command-by-command audit trail of activity within both working-storage and the DBMS's internal currency tables, during applications (1) thru (5).
Konrad 1 was announced by press release [see Press Release (historic pdf)].
Konrad 2.1 and 2.2
Konrad 1 did not simulate any of the structures of the biological nervous system, merely their function. The database Schema was therefore updated during the remainder of 2008 to add the following functionality .....
(7) To simulate the separate anatomical processing modules recognised by cognitive science, to organise these into the most life-like real-time processing hierarchy, and to distribute the different subtasks of cognition around this hierarchy.
(8) To distribute long-term memory content to the most appropriate module within the hierarchy.
(9) To simulate short-term neurotransmission between, and medium-term neuro-sensitisation within, the resulting memory network.
(10) To respond in simulated speech to a complex but static visual scene.
Konrad 2.1 installed these enhancements, and in January 2009 Konrad 2.2 began to address the philosophical problems of visual perception by adding the following functionality .....
(11) To demonstrate how component elements of a visual scene can be permuted and re-permuted for propositional cognition until a satisfactory holistic understanding of that scene emerges.
Taking Meinong's (e.g., 1902) Theory of the Objektiv as its vehicle, Konrad 2.2 was demonstrated in conference at Edinburgh University in April 2009 [link to Conference PowerPoint on homepage].
Konrad 2.3 and 2.4
Versions 2.3 and 2.4 then started the process of cross-indexing the anatomical and functional aspects of cognition, that is to say, of relating mental activity to brain activity. This called for considerable extra detail to be included in the output print. Again taking Meinong's Theory of the Objektiv as its vehicle, Konrad 2.4 demonstrated this more detailed processing log in conference at Oxford University in September 2009.
Konrad 2.5 and 2.6
Versions 2.5 and 2.6 then added a multi-threaded attentional control system to support "multi-tasking". The following functionality was added .....
(12) To recognise input in any of five sensory modalities, namely visual (text and non-text), auditory (speech and non-speech), and proprioceptive.
(13) To output in any of five motor modalities, namely auditory (speech and non-speech) and kinetic (postural, locomotion, and semiotic), and to do so simultaneously wherever possible.
(14) To simulate the basic cyclicality of motor output described by Craik's (1948) theory of the "discontinuous operator". [For an introduction to this problem, see Section 4 of "Motor Programming".]
(15) To engage in time-extensive motor behaviour, using the resultant feedback to make output adjustments where necessary.
(16) To detect task-exceptional input, and to "interrupt" ongoing motor behaviour where necessary.
Konrad 2.6 was demonstrated at the Institute of Ergonomics and Human Factors Conference at Keele University in April 2010. The system cycled at about 6 to 7 times a second - reassuringly close to that of the biological system - and carried out 50 or so primitive database operations per cycle.
Konrad 2.7 then added two of the system's most significant areas of functionality to date, namely .....
(17) To recycle output speech covertly, that is to say, in silent form as "inner speech", and to regard this self-generated stream of input as one's own thoughts being resubmitted for validation.
(18) To support the sort of "second order" propositionality required for a "Theory of Mind" [Wikipedia briefing], and the social cognition which this mental faculty allows.
Konrad 2.7 was demonstrated at the International Control Room Design Conference at Eurosites République, Paris, in October 2010 [see Press Release].
After the Paris ICOCO conference Konrad 2.8 targeted the age-old philosophical problem of qualia - perceptual quality, the hope being that by simulating the aesthetic experiences of liking and disliking something, insights would emerge into the nature of the consciousness doing the experiencing. Specifically, Konrad 2.8 was upgraded so as to respond "emotionally" to a work of visual art. The following core functionality was added ...
(19) To process a visual scene both (a) rapidly for its gist, that is to say, its flashbulb narrative, and (b) more slowly for its constituent detail.
(20) To scan both these input streams TWICE, once at mid-level for biologically important past associations, and then again at top-level for semantic meaning. This dual scanning allows a subconscious "nice-nasty" flag to be set up prior to detailed semantic analysis, thereby setting the emotional tone during the latter process.
(21) To monitor throughput in selected main modules, not least Module #21 - the propositional truth processor.
Konrad 2.8 was demonstrated at the Wrexham Science Festival in July 2012, further documented later that year in one of the University of Zurich's prestigious ShanghAI Lectures, demonstrated again at the CA World '13 conference in Las Vegas in April 2013 [CA Technologies provide the hosting CA IDMSTM database platform, of course], and then again at the EVA2014 conference in London in July 2014.
A major overhaul of the system then took place, reflecting growing industry demand for Big Data and IoT applications. Konrad 5.0 delivered the following core functionality ...
(22) To replace conventional input and output files with reporting streams over Microsoft MQ middleware.
(23) To apply the software's proven capacity for "second-order" propositionality [see (18) above] to the real-time analysis of rapid IoT datastreams.
In May 2017 Konrad 5.0 was shortlisted for Computing's Big Data and IoT Excellence awards.