There seems to be a movement afoot - long overdue - to embody the foundations of intelligence and philosophy. It is partially old news, the heroic AI of the 50's, 60's and 70's has been discounted despite our benefiting from the results of that research in today's mainstream technology: neural nets, machine speech recognition, handwriting recognition, natural language translation, cars that drive themselves, and all kind of other low-level intelligent (unconscious?) activities simulated by programs.
I think that it is the Minsky flavour of AI (now re branded cognitive science), the one grounded in logic, seems to have fallen out of favour -since the fruits of that work did not rise up to the expectation of the "I" in AI; an AI based only on logic, a formal game with symbols that pretended to be words. How can that be smart without the observer having to make all kinds of allowances and assumptions?
This stuff was subject to sometimes acrimonious debate for a while, until technology and big data made many of the arguments moot.
The good news is that there is now a strong movement towards the recognition of the role of the perceptual environment and its embodiment in animals like us as part of the equation.
This realization may be due to the Web, indirectly, since it should be now be obvious that the seeming "intelligence" of the Google search would not exist without the "environment" of all the Web sites, which are of course interactions with people's brains (in fact there was a very simple AI program that used "Google distance" as a way to measure metaphorical strength between concepts. For example "Castro" and "cigar" scored pretty high on that scale).
So now there is a movement, a possibly serendipitous program of research coming together to ground AI in the real-world of experience, and some of the people that I think are leading this effort align themselves along a somewhat elegant but odd symmetry of ideas.
Let me explain.
There is
David Gelernter, the computer scientist and columnist who wrote "The Muse in the Machine" and "Mirror worlds". These books made me want to go back and look at AI again. In fact I ran back. His arguments are so cogent, so seemingly simple and common-sensical that I almost missed their radical novelty. He states that data and the links between data is the key to intelligence. These links are often emotional gradients, in the sense that logic is not always what binds ideas; it is similar emotional affect!
The brain associates memories based on the closeness of the emotions they evoke, the sensory input parallels that occurred as the memory is created. The whole model, not just the idea...how you felt, how cold or hot or tired or dizzy you were. This links data to environment so elegantly that it is a mystery to me why no-one has picked up on this earlier. He breaks the syntax/semantics dichotomy very deeply. Also, Gelernter has very conservative political views - right of centre. I am noting this now so as to make my strange symmetry argument a bit further down. Bear with me.
There is George Lakoff, the cognitive linguist. His work with Johnson on embodied philosophy and with Nunez on the origins of math, takes the empirical evidence around the links between our brain wiring and the perception of the world and uses it as a foundation for philosophy.
Basic brain mechanisms are considered to be foundational and ideas grow as hierarchies of metaphor (i.e. links).
Metaphor becomes a key mechanism for cognition. Similarly to Gelernter, Lakoff et al propound that metaphors are based on evolutionary and environmental constraints, not some platonic logic ideal. No such thing as the separation of mind from the world of things. These guys are existentialist scientists. Finally! The rampant positivism and Platonism of the English speaking world, the Whitehead/Russell school, is giving ground. Merleau-Ponty and Husserl are now more visible in the world of scientific cognition.
Lakoff is an ex-student of Chomsky. He seems to disagree with the flavour of Chomsky' thesis of hard-wired syntax in the brain. He believes syntax and semantics to be more of a spectrum of complexity in cognitive biology than a differentiator between humans and other animals. Also he is a
non-neo-con, a strong and vocal opponent of right wing politics as it is played in the US today, and although I hesitate to slot him with Chomsky on the political front, they are on the same side of a hypothetical centre line.
So we have right and a left wing thinkers on cognition who fundamentally disagree in the political realm, the realm of people relations, nations, influence, and power. That is a healthy and good thing. This lends support to the notion that there is a grounding in "objective" reality for this thinking on cognition, it is much more than opinion. (A slim argument, I know, but hey, this is my blog, I can say what I want.)
Aside from the political thing, I want to continue to talk about is the cognition thing. There is another thinker/scientist that seems to be breaking new ground along those same lines.
He is a neuro-scientist, one who gets his hands dirty opening up craniums, working with brain surgeons. William H. Calvin. I read two books of his: "Conversations with Neil's Brain" and "How Brains Think, Evolving Intelligence Then and Now". His ideas centre around the notion that cognition is an initially subconscious Darwinian process among neural bundles that remember perceptions, integrate them and ultimately model the environment perceived by the body, and that the successful "species" of memories/emotions (let's call them cognitions) are the ones that we call conscious, the ones that rise to the top of the noise of cognition, to differentiate themselves as our inner voices.
His research fundamentally agrees with the models put forth by Gelernter and Lakoff. Like Lakoff, he believes that most cognition in the brain is fomented through unconscious processes. Like Gelernter he finds that memories are linked through non-logical processes. He shares with Lakoff a deep knowledge of Chomsky's ideas on innate language ability and like Lakoff disputes the simplicity of these models, without denying the human brain's ability to parse language. He also talks of metaphor, not surprisingly, since all this is supported by empirical evidence at the chemical and behavioural level.
So I am very happy that this stuff is happening. I tended to get hot under the collar when I read "classic" texts by Minsky or Pinker on intelligence. They seemed simplistic and smug. I was always perplexed by the lack of attention given to Grey Walter's ideas which preceded but anticipated Gelernter's. I wanted 'The Muse in the Machine" to be discussed more, but to my chagrin, Penrose's strange books on quantum brain hypotheses got more press.
Let's get back to empirical research, let's continue the program of work that will ultimately bring philosophy and psychology back into science.The foundational thinkers list above should also include Turing. Even Penrose mentions his "other work". He did mathematical research on molecular biology (plants mostly) and how stem cells can become complex structures. His mathematically-based hypotheses on these matters were validated experimentally just this year in the UK. Ultimately this theoretical work may prove to explain some of the brain mechanisms underlying the work Calvin and others are doing.
There are conceptual layers to all this analysis, from the bottom-up: the molecular, the cellular, the systems (bundles of neurons evolving), and the cognitive and behavioral, leading to metaphor, logic and science, including physics, on to philosophy, morality and politics. The gaps needs to be filled so that we have a continuum of understanding. A New Science in the Vico sense. Something we made and understood that we made, but that is grounded in experience.
(I just found an
earlier post that was wishing this stuff would happen, and so here we are.)
IM/IT systems must align against the business organization (people and processes) to work well. Management must decide who must budge before a new system is put in place in either camp. When deciding whether to build or buy a system, this effect must be considered. Neglecting to factor-in the cost of the organizational change can cancel the perceived economic benefit of using commercial off-the-shelf software (COTS).
Organizations usually have to change to fit the COTS assumptions, especially for large enterprises with large systems such as enterprise resource planning (ERP). This can also apply to core business systems, since they differentiate the company from others. Think of mobile phone billing for example.
If COTS is chosen (for ERP for example), it is likely that the finance and personnel departments will have to change their processes significantly. There are many case studies that show that CEO support is required for such migrations to succeed, since it becomes a governance issue. It has become the common wisdom now.
Replacing custom-built software with COTS is a form of devolution. Size matters in this kind of decision because the “cost of re-organizing versus the benefit of saving on development costs” equation must be juggled. The bigger the org the more the cost of adopting COTS since the org usually has to change. The converse holds as well.
So where does the argument that COTS is preferable come from? Why does upper management often fear custom work? Why do they see it a a dependence on their technical staff, and why should this be preferable to a dependence on vendors? One possibility is that the pattern emerged from the computer hardware world? COTS for hardware usually makes sense for non-manufacturing sectors since the costs of custom hardware are prohibitive and it is generally ridiculous to suggest that one should build computers from scratch. Again, there are notable exceptions, Google apparently configures boxes quite heavily to suit their needs.
The COTS for software approach is not so obvious with the advent of open source and the devolution from mainframes towards lower cost systems, the proliferation of high-level development tools and the commoditization of the software stack. For example networking and OS are now mostly standardized, and much of the higher level layers are becoming standard - Mail/Web, database and application servers. Even development environment possibilities. The database-to-presentation layers are now the target of service-oriented architecture (SOA) driven commoditization, but the vendors are all looking for their own lock-in (i.e. avoiding commoditization) while claiming interoperability to get the sale.
For example vendor A’s ESB makes integration with vendor B’s apps much more difficult than with vendor A’s. Interfaces are easy to import but hard to export. Such pitfalls can be avoided by a vendor diversification strategy across the application layer that is coupled with good cost control and use of custom solutions where appropriate. Thankfully ESB’s are not a prerequisite for SOA, but this is not generally well understood.
So IT planning strategies need to revisit the seemingly popular notion that custom is always a liability. I think that if cost control is the business objective, then it should be decoupled from the “customization vs. COTS” argument. There is no direct mapping. The organizational size attribute has to be factored in, amongst other things.
Another ideology-prone minefield is standardization. Standards should really be orthogonal to this discussion, but are related in strange ways. Standards are sometimes perceived as a factoring exercise to reduce the proliferation of types of solutions and products in an IT department. This can work very well with hardware. Where standards are problematic is when they prescribe major system software COTS without taking into account organization structure (again). Why prescribe COTS through standards? Well COTS is good, right? And less COTS variety should be even better. Hummm.
Vendors claim and often try to provide the flexibility in their products to allow (some) business process mapping and legacy integration but usually at the cost of configuration complexity or alternately through costly professional service customization.
The benefits of these approaches over custom development need to be weighed carefully; there are plenty of case studies that show success with either approach and the cost-benefit is not at all straightforward, especially when you consider upgrade cycles and other dependencies.
If product/portfolio standards are adopted by an organization, then it must also adopt compatible organizational and operational standards, since software is really an extension of organizational processes and COTS will force an organization into a process relationship with the vendor. It is a form of outsourcing.
This can ultimately lead to having the hired-guns, the management change consultants and the vendor IT/IM professional services too tightly coupled, resulting in them having a profound influence on organization structure and operations. Are organizations prepared for this kind of loss of control in exchange for perceived but generally illusory cost benefits?
I think that one of the sub texts of the discussion is at root ideological - outsource development versus do it in-house. Yes, IT departments do have much influence and are complex things to manage, and require expertise, but control cannot always be gained through outsourcing or by buying COTS, unless you are very small or have very simple needs. Horror stories abound. Control and accountability concerns need to be well understood and documented and the business objective must be stated explicitly and decoupled from the implementation. There are no silver bullets, only smart management.