On Neuroplasticity, the Extended Mind and the Intelligence Explosion
This posting is a reply to this response by Daniel Estrada to my paper The Coming Social Singularity.
Mr. Estrada argues that my basic position requires a strong differentiation between the technological and cultural. This is, however, not what I have intended to convey. My paper rather concerns an argument for the comparison of the plausibility of Vernor Vinge’s AI (artificial intelligence) and IA (intelligence amplification) hypotheses. In other words, I do agree with much that Mr. Estrada writes. We are, in many senses, “tools all the way down”. As what comes to the nature of the mind, it is in a very profound sense extended to begin with. If an AI were forthcoming, it would in many senses contribute as an extended resource to the human mind.
My claim in the paper is not so much intended as the comparison of the intrinsic nature of a biological mind to a simulated mind (which, as I think Mr. Estrada rightly points out, cannot justly be separated), but rather the plausibility of whether an IA or AI explosion will take place sooner: in other words, where the focal point of the intelligence explosion will be: in the network itself (IA), or in identifiable components of it (AI).
The problem with the plausibility of the AI hypothesis is not that it would be impossible or somehow IA-incompatible. It is rather that we are not very likely to reach it before an IA explosion takes place. In addition to the complexity of the nervous system that can be postulated on the grounds of the Stanford experiment, the integration of nervous and extended processes is of a far higher order than in a simple sensory coupling or a feedback loop. The nervous system is dynamic to a far greater degree than any known computational system as is demonstrated by the massive literature on neuroplasticity. The nervous system does not compute – synaptic connections grow and shrink. The brain is not a machine. It is a garden.
In the light of what we now know about brain function, in the nervous system the hardware and the software are intrinsically intertwined. In other words, the brain is not a static processing and memory system where information is stored, but rather a dynamic feedback mechanism that *produces* information by creating complex enough connections. Once you add to this the ability to augment these connections by using the environment, there is a very profound sense in which human intelligence differs dramatically from what has been postulated as machine intelligence. Using Searle as an example was simply to show that there are some dramatic difficulties in attributing intelligence to a machine (whereas attributing intelligence to a human-machine coupling is by no means problematic).
Incidentally, this is not to say there could not be an intelligent machine. I do not subscribe to the fundamental Searlean assumption that this would be philosophically impossible. Quite the opposite: if a machine is constructed that for all purposes acts like an intelligent agent, it should be treated as an intelligent agent, even if this behavior came about in the way of complex enough computation. But this is not at all the point I am trying to make in the paper.
What I am arguing is that while an AI explosion may as well be on the way, it is not very likely to happen very soon. But once real-time networking of human beings is achieved (which should happen in a few years now), the IA explosion will take place. I have no doubts that this will also contribute to the AI explosion as well, whatever that will mean then, which will in turn augment the capacity of the IA system and so on and so forth.
To sum up, none of this is to say either that human intelligence and machine intelligence should be intrinsically separated, or even that the simulation of intelligence were impossible. It is just to say that the intelligence explosion that involves real-time networking of existing nervous systems of human beings is somewhat likelier to happen sooner than a significant enough advance in computing technology.
Why Apple’s iCloud may be one of the biggest computing revolutions up to date
Like TUAW said, you don’t build a really big data storage facility in North Carolina just to stream music. On the contrary, tomorrow’s iCloud publication may be one of the biggest revolutions in computing up to date.
It looks, in fact, like Apple is going for over the air realtime sync of all apps and media over iOS and OSx devices. In effect this is no different than the present iTunes wired sync – they have had most of the backend for ages. But in going wireless, the difference is huge.
An elegant and transparent sync solution will bring the usability us nerds have already drawn from Dropbox to the mainstream. Think about keeping all your iDevices and Macs in real-time sync, docs and apps included.
This will mean that your entire desktop and file system goes with you wherever you go, completely transparently. This would be a revolution in cloud computing. You write on iPad. Edit on iPhone. Finish on a Mac. No syncing, sending or updating required.
Of course, they could just release a watered down Spotify. Tomorrow we’ll find out.
What the Extended Mind Does Well – And What It Doesn’t
What EM Does Well
Declarative Memory
It is relatively easy to dig up trivia and tidbits if you have a good enough archiving system and/or search engine. With biological memory, the information must be relatively significant to be remembered.
Volitional Recollection
Directly related to the above: it is difficult to volitionally remember many things, whereas digging them up from an archive is easy.
Information Management
Again, directly related to the above: information management is massively easier with pen and paper and libraries, not to speak of the digital realm. Furthermore, with the advent of ubiquitous connectivity, we can push the digital retrieval response times close to spontaneous recollection, which will no doubt produce interesting results.
Ubiquitous Availability
That is, of course, unless things crash or break apart. But digital technologies enable us an increasingly available access to EM capacities, whereas biological mental capacities are available variantly.
Task Management
Externalizing information works particularly well for tasks and other repetitive declarative information.
Organizing
Directly related to task and information management.
Generating Randomness
This should be rather obvious; the biological mind cannot produce genuine randomness. A program can.
Calculation
Once again, rather obvious: all rule-following is massively easier to an algorithm-driven program than to a human being.
Collective Thinking
This is only beginning to emerge, but we can do more and more together with the aid of EM technologies, whereas in biological connectivity, we are limited to very small groups.
Social Networking
Directly related to the above: real-time social networks are relatively small, whereas a digital network can consist of hundreds of active participants.
What EM Does Not Do Well
Creative Thinking
Machines do not as for now think creatively. Furthermore, while EM can augment creativity (think mind maps), it does not alone produce creative thought.
Emotions
This is actually more relevant to AI than EM; it is also arguable that EM can be used to induce and direct emotions. But once again, it is a subtle interplay between the biological mind and EM.
Reflection
It is very hard to think what would EM reflection even mean. Reflection is quite directly related to the biological mind, while of course it may involve EM components.
Metacognition
It seems metacognition is hard for both BM and EM. Perhaps a solution will emerge later? Thinking about thinking is not a very easy skill to learn, it appears.
Humor, analogies and irony
These require a human interpreter, and do not have an intrinsic EM dimension to them.
Evaluating Information
This is a field where EM will no doubt soon catch up. Nonetheless, right now automatic evaluation of information is still very elementary and gives a very varying mileage.
What the Mind Does Well – and What it Doesn’t
In our latest session we sat down to think over what the biological mind does well, and what it does not. Likewise, we considered what the extended mind does well and what it does not. Here are the results we brainstormed; coming up next week, the respective EM ones.
What the Mind Does Well
Creative Thinking
We still have to build a machine that is capable of genuine creativity. Whether this is a question of complexity, hidden variables or something we do not even yet understand is an open question. Nonetheless, biological mind is by far superior in creativity compared to technology.
Intuitive (Aschematic) Thinking
Same as above: machine intelligence is still for the most part schematic thinking, whereas intuitive thinking, at least according to some researchers such as Djiksterhuis and Nordgren, is aschematic.
Reflection
Machine intelligence is paradigm-constrained, whereas human intelligence can reposition and view things from various perspectives. Also relevant to empathy.
Semantic Processing
EM is catching up here, but humans are still superior in understanding meaning.
Irony
I think this one will take too long to explain.
Humor
Directly relevant to the two of the above. Also to the first item: whether this is a question of complexity, or of something deeper is still an open question.
Analogies
Same as above.
Imagination
Do androids dream of electric sheep? This leads to a can of worms of a question with respect to AI and EM, that is to say, can even the most complex of machines have phenomenal consciousness?
Association
Here too, technology is catching up fast, but biological mind still prevails.
Beliefs
Similar question as imagination.
Dogmas
Only an agent can have dogmas (i.e. axiomatic beliefs). Does this require a biological mind?
Image recognition
This is similar to semantic recognition: machines still have some way to go, but they are catching up.
Things the Mind Does Not Do Well
Tedious Tasks
We tend to get bored quickly with repetitive tasks.
Massive Information Storage
What did you have for lunch a month ago?
Trivial Declarative Just-In-Case Recall
What is the tenth digit of pi?
Volitional Recollection
See massive information storage.
Metacognition
What do you think about what you think about right now?
Calculation
8433953 x 234235?
Task Management
This is an interesting tangent to EM in terms of information processing. Tasks consist of declarative memory items, and they are hard to recall volitionally.
Information Management
Like David Allen put it, a brain is a great place to have ideas, but lousy to store them in. The memory constraints apply to any management of large amounts of non-consolidated information, for example raw data.
Cognitive Multitasking
Here, the constraints of the working memory (the magical number seven) make it hard to focus on several processes at the same time.
Thinking by Negations
The biological mind seems to have hard time grasping the word no. Try not to think of the pink elephant.
Rational Thinking
Whether we like good old Aristotle or not, we are not really very rational animals. Human decision making seems driven by a huge number of cognitive biases and other effectors that have nothing to do with rational inference.
Analysis
Directly related to the above. Also, even the most rigorous mind must commit to some axioms and make intuitive decisions on choosing rules of inference. Pure rational analysis just does not seem to be cut out for the human mind.
Next week, a similar breakdown of what the extended mind does and does not do well.
How the Social Singularity Makes Creative Collaboration Possible
by Petro Poutanen
The collaborative capacity of the Web might bring about a new era of human intelligence: the social singularity. The social singularity refers to collective human intelligence enabled by a huge number of interconnected individuals. Imagine the possibilities of an enormous information pool that the millions of web users comprise. Some examples are Wikipedia and more recently Aardvark that provides an extended social network for answering people’s unique questions. Also firms are seeking ways to benefit from the collaborative capacity. For example, a firm called InnoCentive provides a common platform for companies looking for solutions and people willing to solve companies’ problems. What is important here is that people are contributing creative outcomes without centrally planned organizations. So the question goes: How the emergence of social singularity makes collaborative creativity possible?
Momentarily, we are lacking a common, coherent theory of network-based collaboration but having multiple terms for the phenomenon (crowdsourcing, collective intelligence, open innovation, the wisdom of the crowd – to mention but a few). Clearly, we are talking about some kind of “self-organizing” – an uncoordinated behavior of a large mass that produces something coherent and cogent collectively – yet we don’t know if there is a single logic behind the different kinds of self-organizing systems, such as ant colonies and human brains.
I have tried to figure out how to describe a system of collaborative creativity in online. What happens when people solve problems collaboratively? I have come to think about this as a system of creativity. The famous systems model of creativity suggested by Mihaly Csikzentmihalyi is constituted of three parts: the cultural domain, the field of experts and the individual. For creativity to emerge, the individual must produce a novel variation of cultural information, which is then subsequently selected by the field for inclusion in the cultural domain. Thus, creativity is the product of all constituent parts in the system and emerges from the interplay of them.
What would this model look like in the collaborative online environment? First of all, the creator (individual) and the evaluator (the field of experts) can be the same person. A person participating in a project of programming might contribute the project with a single line of code and simultaneously, by that very same contribution, act as a peer to another contributor by suggesting a modification to the code made by that participant. So, in a collaborative field anyone can be an expert and a creator at the same time. Secondly, we have knowledge that on the one hand belongs to the cultural background of a participant, and on the other hand to the field of continuous negotiation.
This perspective opens up a dual nature of both the contributor and the knowledge in a collaborative field: the contributor as a “creator” and “modifier”, and the knowledge as a “cultural” and “shared/negotiated”. And when thinking about this collaboration on the systemic level, it is the group in collaboration that decides whether a variation produced by an individual (on the basis of the collective work) is selected or not. The picture below illustrates that dynamics.
Figure 1. A model for Creative Collaboration in Online Environment. The process starts when a problem or a need for change occurs. An individual contributor proposes a solution to it in a form of a variation drawn from the cultural knowledge. Subsequently, other participants work as peers to the proposed variation and give feedback to participants. Then variation is modified, if necessary, and finally selected. After the requisite amount of iteration, the final solution is moved to the area of cultural knowledge.
Functions for the Extended Mind
A cornerstone of the extended mind hypothesis is to look at the mind from the point of view of cognitive functions. What are such functions that the EM should encompass? Here’s a quick taxonomy we are planning to use for our fall EM classes.
1. Seek
Looking up information is an obvious internet EM function. Services like Google and Wikipedia augment our access to new data significantly compared to offline access to data.
It is an interesting question what this does to the concept of knowledge. If information is accessible on the internet as fast as if it was memorized, is that information already knowledge? This would at first glance seem to follow from the original EM hypothesis. So do we already know what is on the internet? Opinions vary in our group for the time being.
2. Sort
Another function, especially for EM tech, is sorting out the massive amounts of information we can access. By using various web services, such as iGoogle and StumbleUpon, we can create interfaces that produce only relevant and interesting information.
I am looking forward to working augmented reality solutions that would also bear some sorting function for offline data. It would, for example, be fabulous to be able to replace advertisements with inspiring information, à la Vernor Vinge’s novel Rainbow’s End.
3. Store
Evernote has branded itself as your extended memory, and that is precisely what it does: extends your capacity to recall information. Accessing information on Evernote is massively different than accessing information on Google or Wikipedia, since that information is already processed by you.
In other words, information on Evernote has massively more significant semantic load than some piece of data you look up on Wikipedia. By storing information in notebooks and cloud services, we can expand our available reservoirs of useful information. We are already actually very close to never forgetting.
What happens when online cloud storage is coupled with smart semantics, and perhaps some augmented reality integration?
We live in interesting times.
4. Share
Finally, the cornerstone of swarm intelligence and social singularity (not to speak of distributing funny cat videos) is sharing. Social media is already making a huge impact on how information is accessed and processed. What happens, when we are able to share information more or less real-time?
There are still hurdles to cross. But nonetheless, we are advancing at an amazing rate at the time being. And we are looking at a very interesting future.
To recap: we live in interesting times.
Social Sherlock
We have been looking at various angles as to how something like a social singularity could be brought about. In our latest think tank session, we came up with an idea: how about if we pit a swarm intelligence against an AI? More specifically, how well could a swarm do against IBM’s now famous Watson supercomputer? In February, Watson beat two of the most prominent human players in Jeopardy – an unprecedented feat for a computer.
Basically pitting a swarm against Watson would require some kind of a real-time crowdsourcing platform, which would also involve features like specified in the earlier EM tech postings. In other words, the platform should be able to differentiate participants’ credibility, and it should also be able to differentiate between a huge pool of submitted data to select best contenders for answers. It appears that these functions can be implemented with relatively simple semantics that draws from the social network itself.
Effectively, there could be three phases that this kind of a project would require from the social community:
I) Open source development of the platform itself.
II) Calibration of the platform and practice of the participants.
III) Participation in the swarm itself.
For the time being, “Social Sherlock” remains pure brainstorming, but we will see whether it can be developed in further think tank sessions.
