Believe me, if this information is all you’ve heard or remembered (if, of course, you’ve seen anything at all); you’ve just caught sight of the first icy pinnacle of the above water iceberg.

As is typical, what gets the most attention is the thing someone made, in this case the neural core chips. In this case, that misses something far more important – the history and progress of a specific research unit within IBM, the Cognitive Computing group, and its chief scientist, Dharmendra Modha. The neural core chip isn’t some one-off research product; it’s a component that researchers decided was necessary to make progress in a massive research program that began in 2006. Funded to the tune of many tens of millions of dollars [most recently$21 million by the Defense Advanced Research Projects Agency (DARPA) for Phase 2 of the Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project], the Cognitive Computing group encompasses the efforts of IBM’s Almaden Research Center, IBM’s T. J. Watson Research Center and five academic institutions (Columbia University; Stanford University, Cornell University; University of California, Merced; and University of Wisconsin, Madison).

The Cognitive Computing group first made news in 2007 with a mouse-scale brain simulation, followed by a rat-scale brain simulation, then in 2008 a cat-scale brain simulation and finally a simulation of a monkey brain. At each step the simulation required a much bigger supercomputer and it became apparent to the researchers that a traditional computer with enough power to achieve a human-scale simulation would require so much energy, it would probably incinerate itself. Yet the human mind doesn’t (usually) incinerate, in fact, it operates rather nicely at about 10 watts, a rather dim bulb. Modha and his team came to realize that if computers were going to achieve human level mental complexity, they too would have to use less energy. This demanded a different model of computing, hardware and software, than the current mainstream (von Neumann) computers. The new model, as expressed by the tiny, low power building blocks of the neural cores is the cognitive computer.

The Big Picture here is the analogy Modha uses, comparing traditional computers with the left side of the human brain and the cognitive computer (composed of neural cores) with the right side of the brain. In this picture you need both sides to make a complete brain. It’s a useful analogy but as usual beware analogies. The cognitive computer, if and when it gets built, will not be a brain hemisphere. For one thing, as neuroscientists will freely admit, we don’t know how the brain works, much less how thinking works. Yes, there has been great progress in understanding the physical and chemical processes of the brain – enough progress so that the knowledge can inspire ideas like the cognitive computer, but there is no one-to-one correspondence between what we currently know about the brain and the ability to design artificial intelligence.

In an important sense, it doesn’t matter. The inspiration may be enough. For example, in the neural cores IBM has created a silicon chip that weaves the function of memory (RAM) into the function of processing (CPU) so close together (45 nanometers) – like neurons and synapses in the brain – so that an enormous amount of energy is saved at each moment of processing (a thousand times less energy than a transaction in a standard computer). The neural core chips run cool, so to speak, and achieve something already accomplished in the brain without needing to be exact copies of the brain. Indeed, there is nothing biological in the neural core chips.

Another crucial element of the cognitive computer that usually is unmentioned in a discussion about the neural chips is the software. Software in this case isn’t the traditional programming – logic, rules, step-wise sequences – of the von Neumann type computer, instead it is the means of interconnecting the operation of the neural cores so that the system learns. Learning is its programming; something like it is for all animals.

It’s impossible to do justice to this research in something short of a book (an e-book anyway). It’s also possible to overlook the possibility that this approach won’t lead to the results the researchers want. In any case, the cognitive computer, which I repeat doesn’t exist yet, does not get to sentient artificial intelligence. It will, however, be a really big step in that direction. Whether the notion of cognitive computers is ever completely realized, the process of developing the notion will teach us a great deal, for we remain the ultimate in learning machines.

I was thinking in terms of the Watson technology specifically, which at the moment is state-of-the-art and costs like it. I reckoned it would take some time, perhaps years, before that technology was adapted for commercial practicality (as in making the price affordable). That wasn’t wrong as far as I went, but the process has already begun. The New York Times, 4 March 2011, article Armies of Expensive Lawyers Replaced by Cheaper Software, outlined the advent of computers and software that can do the searching through legal briefs that used to cost millions and take months, and do it for say $100,000 and in a few weeks.

Computers are getting better at mimicking human reasoning — as viewers of “Jeopardy!” found out when they saw Watson beat its human opponents — and they are claiming work once done by people in high-paying professions. The number of computer chip designers, for example, has largely stagnated because powerful software programs replace the work once done by legions of logic designers and draftsmen.

[Source: New York Times]

The main point of the article was that computer software, what most people think of as ‘search’ programs but are actually specialized database management applications, are becoming increasingly sophisticated in detecting not only key words but also writing style, signal phrases, and other elements that are useful in doing analysis of legal and business papers. Of course, the computer does much more quickly than humans. What the computer finds still needs to be eyeballed by the meat sacks (people), but as the article implies a lot of jobs are removed in the process. This has broad economic implications.

That aspect of the economic power represented by Watson and its kin was picked up by the New York Times columnist and resident economics Nobel Prize winner, Paul Krugman:

Some years ago, however, the economists David Autor, Frank Levy and Richard Murnane argued that this was the wrong way to think about it. Computers, they pointed out, excel at routine tasks, “cognitive and manual tasks that can be accomplished by following explicit rules.” Therefore, any routine task — a category that includes many white-collar, nonmanual jobs — is in the firing line. Conversely, jobs that can’t be carried out by following explicit rules — a category that includes many kinds of manual labor, from truck drivers to janitors — will tend to grow even in the face of technological progress.

[Source: New York Times: Paul Krugman Blog]

His point, well sharpened, is that many white-collar jobs that form a significant percentage of today’s middle-class are in jeopardy (pardon the expression, this is not a game). Ironically, as he also mentions, it’s the more hands-on (literally) jobs that are thus far outside the range of this kind of computer. That’s the good news, I suppose. The less than good news is that hands-on robots are also rapidly becoming more sophisticated.

During the 1960’s and 70’s it was fashionable to foretell the demise of millions of jobs due to automation. Computers and especially robots were to blame. Actually many jobs were lost, but as was also foretold, many new jobs were created. It wasn’t quite a wash, but at least in the developed countries the loss of employment from automation was far less than that from ‘outsourcing’ – sending the work to countries where labor costs were lower.

Will it be different this time? Yes. This time even outsourced jobs will be lost. For the most part, manufacturing automation has already happened, and where it hasn’t, it will mostly be in the developing countries. In the next ten years or so, the big losers will be many of the relatively good jobs that were exported to countries such as China, India, Thailand and the Philippines. Other losers will be jobs within the middle class of developed countries; managers will find themselves without human staff. This is called the ‘hollowing out of the middle class.’

Krugman thinks the biggest shock of all may be that education isn’t necessarily going to be the golden path to economic success we’ve been led to believe. The kind of educated worker our current education systems produce – even from higher education – are going to be the most vulnerable to replacement either by computers or by more adaptable labor in other countries. His remedies, which I won’t go into here, are not very convincing. But the moral of these articles is that automation of knowledge work is already underwayand we better start considering what, if any, options we have.

In February, 2011 half the population of the United States gathered around TV screens, gorged on fat-foods and swilled beer, while watching a sports contest of epic proportions called the Superbowl. Also in February for three consecutive days several million Americans watched a television game show called Jeopardy! It too was promoted as a contest of epic proportions: Man vs. Machine. It featured two men, Ken Jennings and Brad Rutter, considered the best to have ever played the game, who faced (if that’s the word) a computer called Watson, the latest and greatest Artificial Intelligence created by IBM. A team named the Green Bay Packers won the Superbowl. Watson beat the human players.

Most people living outside of North America know nothing about either epic contest. One of these contests has almost zero impact on anything; the other contest symbolizes something that will relatively soon affect the livelihood of people all over the world.

I’m deliberately pushing the irony button. I hope it’s obvious that the Superbowl and Jeopardy events are not equivalent, although given the levels of promotion (hype); they seemed to be approaching comparable importance. (Of course, by the measure of hype, nothing beats the Superbowl…not even the World Cup.)

In 1997 a computer IBM called “Deep Blue” narrowly defeated the reigning champion of chess, Gary Kasparov. It’s quite likely you’ve heard references to the event because it was considered a significant breakthrough for computer intelligence. In reality there was no intelligence involved; just massive calculation of chess positions (called brute force computing) and a set of rules for playing the game. Big Blue was built exclusively for playing chess, in fact, it was built just for the match with Kasparov and was dismantled shortly thereafter. Nevertheless, the victory had big juju, uh, symbolic power for computers.

In some ways the Watson victory in Jeopardy! is similar. More than a few people who watched the show were left with the impression that Watson has some kind of intelligence. It doesn’t, no more than Big Blue did; but Watson is quite a different computer than Big Blue. Where Big Blue calculated millions of chess moves, essentially mathematical calculations, Watson draws on an enormous database of facts (Jeopardy trivia), which it handles much like a specialized search engine. Where Big Blue was programmed with the relatively simple rules of chess, Watson has a much more sophisticated program to evaluate the facts coughed-up by the search engine, determine the most likely answer, and then play by the rules of Jeopardy. Watson also has something Big Blue didn’t have at all – the ability to understand questions in plain language (English, in this case).

It’s this capability, called natural language processing that gets Watson in the game and sets it apart from most other advanced computing. To play Jeopardy it had to understand the questions posed by Alex Trebek, the game’s host. Jeopardy questions are notorious for containing wordplay, ambiguity, wit and double meaning. They’re meant to be tricky. So not only did Watson need to understand plain English (through what’s called a Conversational User Interface), but it had to deal with the language peculiarities of Jeopardy. Of course, the developers of Watson’s programming were able to concentrate on the typical wording, cues, and phrases found in Jeopardy. They did well enough that Watson could correctly interpret most of the questions, but notably it did worst on those questions which were meant to be tricky.

Watson is not perfect and using 2600 processors in 90 IBM servers (not to mention the army of people who set this all up) – it’s way too expensive to be commercially useful for general applications. Nevertheless, even though putting the three elements together – natural language understanding of Jeopardy questions, searching a huge trivia database, and operating within the rules of the game – did not produce general intelligence; what it did is demonstrate for the first time to a mass audience an Artificial Expert.

Expert systems have been around a while, decades in fact. One way or another they capture some specific body of knowledge, nominally something an expert would know, and make it available on demand. With its ability to understand questions and formulate relevant answers in normal language, Watson takes the expert system to another level.

Let’s be clear about this, this Artificial Expert. It can understand questions in normal language. It can look up answers to complex questions in split seconds. It can select the best available information and provide the answer in normal language. It can sound like a human being. What else do you think a machine like Watson can do besides answer Jeopardy questions?

Let’s be specific. Do you take orders over the phone? Are you employed as a receptionist? Are you one of the people who answer questions on a help line? You will probably be out of a job when future generations of Watson become less expensive. How long do think that will take? Three years? Ten years?

Robotics and computer programs are already replacing manual labor and various kinds of information systems (yes, even the terrible phone menus count). As an Artificial Expert, a Watson-like computer will be capable of replacing or augmenting a wide range of jobs in the information industry. Most of these are considered relatively safe white-collar jobs. They’re not that safe, never have been. Most knowledge workers have ‘good enough’ jobs. They don’t really have to be experts; they just need to be good enough to answer more or less routine questions in relatively well defined areas of knowledge. Watson can already do this. Many other computers will be capable within this decade.

Still skeptical? Consider outsourcing – that’s where jobs are transferred to countries where the labor costs are lower. Outsourcing has already eliminated many knowledge worker jobs from some countries. Think of Artificial Experts as the ultimate outsource.

I’m piling on, but it’s necessary: In a modern economy, knowledge workers account for between 30 and 50 percent of the work force. Over the years – and we’re not talking decades – Artificial Experts will replace increasing percentages of this type of worker. The question is: What kind of job will replaced knowledge workers find?

I’m being deliberately blunt because complacency about the future of computer power is not harmless. We need to start considering that future. What happened in the Jeopardy game is not far away from happening to a lot of people’s jobs.

Nor does it help to pull the denialist stunt. You’re going to read and hear from people a lot of “Watson was interesting but technically no big deal.” Technically, they’re right. Substantively, they’ve got it wrong. It’s a bit like the Apple iPad. You can pick that tablet computer apart in all kinds of ways, and make all kinds of disparaging remarks. Much of the criticism is warranted; but it doesn’t matter. The Apple iPad is not only the most successful new computer in a decade or two, but it’s also driving the tablet computer into becoming a standard format for the future. IBM is determined to do something like that with the technologies behind Watson.

Granted, those technologies are in a different league than the Apple iPad. Watson-style computers are a long way from the consumer market. IBM will go first where the big money is available – medical applications, corporate systems and the like. It will take years to improve the price/performance ratio to make the technologies a commercial success. The important thing is that where IBM goes, so will at least a dozen other companies. It does not take a crystal ball to see that Artificial Expert computing is potentially a big, spreading money tree.

“Who’s afraid of Watson?” is not entirely a rhetorical question, but Watson is not Hal. I hope that’s clear. It is not to be feared like the psychotic autonomous thinking machine depicted in Stanley Kubrick’s 2001. Watson does not have intelligence. Many future generations of Watson will not have general intelligence. In practical terms, however, it doesn’t matter. Watson is showing us – right now – just how much even limited computer intelligence will compete with the knowledge of human beings. Maybe we should be just a little more anxious.

It turns out that research on swarm intelligence has many potential applications that are pursued by a diverse set of interests including the military, space agencies, robotics companies, and nanotechnology research. Why?

As the article emphasizes, whatever ‘intelligence’ is – and we still haven’t really crystallized a definition, much less figured out how to create it artificially – it’s very complicated. Perhaps someday we may untangle all the aspects. Meanwhile, the demand for things, including many of our machines, to work in an organized and ‘intelligent’ fashion has also outstripped the ability to provide for it. It seems like a natural avenue of approach to look at making many simple things operating with simple rules perform that old trick of the whole being greater than the parts – in short, using swarm intelligence.

It’s well worth the time to read (or at least scan) the Wikipedia entry on swarm intelligence. It may be something of a revelation to learn how much this field has expanded in the last decade or two, for example:

- Ant colony optimization (ACO), algorithms that use ant behavior to simulate complex environmental and search problems (also applied to bees)
- Particle swarm optimization (PSO), examines the behavior of communicating particles to solve various geometrical and dynamic problems
- Stochastic diffusion search (SDS), using simple agents to converge to solutions of complex problems

There are other branches of SI (yes, Swarm Intelligence), but you probably get the idea. Swarm intelligence comes in many variations, and it’s useful. Some of the research is abstract in the extreme, beloved mostly by computer scientists and mathematicians. Some of the research is driven by immediate practical needs, such as controlling a swarm of nano-robots. From time to time an application or a variant model of swarm intelligence pops into view, but usually it represents yet another gradual advance of a field that continually experiments with the border between simple and complex. And that’s where the Economist article finds the value in applying SI to AI.

But anyone who is really interested in the question of artificial intelligence cannot help but go back to the human mind and wonder what is going on there—and there are those who think that, far from being an illusion of intelligence, what Dr Dorigo and his fellows have stumbled across may be a good analogue of the process that underlies the real thing.

For example, according to Vito Trianni of the Institute of Cognitive Sciences and Technologies, in Rome, the way bees select nesting sites is strikingly like what happens in the brain. Scout bees explore an area in search of suitable sites. When they discover a good location, they return to the nest and perform a waggle dance (similar to the one used to indicate patches of nectar-rich flowers) to recruit other scouts. The higher the perceived quality of the site, the longer the dance and the stronger the recruitment, until enough scouts have been recruited and the rest of the swarm follows. Substitute nerve cells for bees and electric activity for waggle dances, and you have a good description of what happens when a stimulus produces a response in the brain.

[Source: The Economist]

I was reminded of this by an article in The Economist Droning on: How to build ethical understanding into pilotless warplanes.

The notion explained in the article is to give drones the ability to analyze not only the tactical situation, but also the ethical. Ethical?

Ethical as in – The Laws of War and Rules of Engagement – these are the codes by which modern warfare is supposed to be conducted. If a robot-drone, that is, its computer programming can be made to analyze a situation according to the Laws of War and Rules of Engagement; it can be said to have ‘an ethical governor for constraining lethal action in an autonomous system.’ This happens to be the title of a 2009 paper by Ronald C. Arkin (Georgia Institute of Technology, USA) and team GVU Technical Report. (The work behind the Economist article.)

Note the phrase ‘constraining lethal action.’ That means the drone’s analysis will contribute in some meaningful way to the decision of using lethal force, or not. The assumption is that a drone, flying over a target, can ‘see’ more clearly the objects in the target area, assemble information from a variety of sources (including ground observations), and combine this with a built in set of operational rules to determine if the target is appropriate. All in ‘real time,’ as computer techies would say, meaning most of this happens on the spot and in seconds.

Unfortunately, the background to this is the issue of ‘collateral damage’ – the killing of bystanders and other innocent people. Repeatedly, attacks using drones have also killed civilians, sometimes egregiously, meaning there was no valid military target present. That’s a critical point. When the drone uses lethal force on a building or vehicle, if it is a civilian asset, then it must be shown that it was ‘commandeered’ or taken over by people who are valid targets (i.e. terrorists, enemy army, etc.). This kind of ‘in the field’ distinction has profound legal and moral implications – war crimes. It would be helpful if the drones could distinguish friend, foe, and civilian. It would be even more helpful if the drone could evaluate probabilities of violation (or compliance) with the Laws of War or the Rules of Engagement. According to Dr. Arkin’s work, this is possible. He proposes giving the drone (through its software) an “Ethical Architecture.”

This would be a very complex rule-based system. (A rule-based system is an old field in artificial intelligence programming.)

An interesting set of conditions are involved:

- The drone’s optics must be good enough to distinguish major characteristics of human figures in its view, as in clothing, colors, body position.
- The database of known assets (which already exists) must be accurate on the ownership for each building in the area of the drone’s interest.
- Any ground observation fed to the drone needs to be accurate, particularly when it comes to the presence of civilians.
- The communications with the drone must be reliable and efficient in real-time, even when a drone’s controller may be thousands of miles away.
- The ability of the drone’s programming to accurately evaluate the situation must be quite good.
- The programming of the rules (the algorithms that evaluate what rule goes with what situation) needs to be quite good, nearly perfect.
- The links of command and control (between the drone, its control personnel, and officer in charge) must be effective in real-time.

None of these conditions are impossible, even for current technology. In fact, some of them are met on a routine basis. However, most of them have points of weakness, which in the fog of war tend to become points of breakdown. Mistakes mean loss of civilian life.

Most everything a combat mission robotic encounters will involve this sort of problem. We already have drones. Ground based robotic weapons are undergoing trials. Most of them have addressed some of the ethical control issues; but nothing like Dr. Arkin’s ‘ethical architecture’ is currently in use.

The ethical robot is an old topic, going back even farther than Isaac Asimov’s famous Three Laws of Robotics. The military use of robots, which by the way is not limited to the United States military, puts the issues in high relief. Since it is well known that human soldiers make mistakes against the Laws of War and the Rules of Engagement, what are the standards for robots? While theoretically the robot’s ability to analyze and assess a situation is faster and more comprehensive (and not clouded by emotion), the elements that go into that process are inherently unreliable. We do not have artificial intelligence, yet. At a fundamental level, a robot will have large areas of incompetence added to any flaws in its working environment (like a bad communications connection).

Nevertheless, war conducted remotely by robotic machines is the trend. Theoretically, between air power, rockets, artillery, and robotics – a military action (the actual fighting) could be conducted entirely soldiers at a distance. This has been a topic of discussion for a long time. Meanwhile, the technology becomes more sophisticated. Even Dr. Arkin’s ethical architecture is a possible piece of programming. The pressure to use more robotics (as long as the budget allows) and the pressure to keep robotics under control is not going anywhere but up.

This month (Jan. 2010) MIT (Massachusetts Institute of Technology, Cambridge, Mass. USA) kicks off a five year, five-million dollar (at least) research orgy called the Mind Machine Project (MMP). Machine Mind = Artificial intelligence (AI). Now there’s a concept. Oh wait. We tried that, what? Fifty years ago? The nineties were big on AI. Last year? Still trying, obviously. It’s like the old joke about a doctor’s practice – “Oh, he still hasn’t got it right!” I’m not sure if the Mind Machine Project is a resurrection of AI, an exhumation, or a back-to-the-laboratory Dr. Frankenstein operation. Probably all of that and more.

MIT has assembled a lot of famous natural intelligence to participate in the project: Newton Howard, Marvin Minsky, David Patrick, Eliot Gershenfeld, and Ed Boyden (all pioneers or well-known practitioners of AI). This could be wonderful – a lot of creativity, big thinking, and very smart people. It could be awful – huge egos stepping all over each other. The betting line should be on ‘messy’ – good things will come out of it, but perhaps as inceptions of ideas that will find more fruitful ground elsewhere.

I can’t help feeling some of this project is atonement for sins, well, perhaps not atonement but possibly secular restoration to good repute despite past sins of the intellect and methodology. Among the many fields of technology, seldom have so many predicted so much and produced so little. I am not saying that AI research produced nothing. Speech recognition and endless telephone menus came out of AI research. Clever use of people’s expertise came out of Expert Systems (a dead branch of AI). There are thousands of useful algorithms and subroutines floating around the world of software that owe their heritage to AI research. However, I am saying that compared to the predictions, “…machines will be capable, within twenty years, of doing any work a man can do.” (Herbert Simon, 1965), the actual products of AI were nowhere nearly so spectacular.

So here we go, again. Time for a rethink. Time for a reinvention of AI. Why not? Any concept that’s been kicking around for half a century with questionable success is a candidate for a rethink. Ah, but this is science and technology. AI was conceived in a state of almost pure ignorance (the 1950’s). Since then the fields of study such as cognitive psychology, neurology, and supercomputing have made great strides. We’ve come a long way, baby. Nobody is going to make the kind of silly predictions that were made back then…

One of the projects being developed by the group is a form of assistive technology they call a brain co-processor. This system, also referred to as a cognitive assistive system, would initially be aimed at people suffering from cognitive disorders such as Alzheimer’s disease. The concept is that it would monitor people’s activities and brain functions, determine when they needed help, and provide exactly the right bit of helpful information — for example, the name of a person who just entered the room, and information about when the patient last saw that person — at just the right time.

[Source: Dr. Dobb’s Journal]

Oh. It’s a mind reading project, among other things. Of course, this is just what was reflected through the media. I’m assuming the people behind the ideas have much more sophisticated knowledge of the challenges involved. Most of them should have, they’ve been through this kind of hype before, a couple of times. Trouble is, as they say, the line between thinking big and blowing it out the ass is right down the spine.

The Mind Machine Project confronts the same issue that dogged AI from the beginning – we don’t know how the brain works, which means we don’t really know what intelligence is. We know some things at the gross physical level about which areas of the brain are involved with some kinds of mental activity. We know much less about how the brain works at the molecular level. Consequently, whatever we think we know about ‘the mind,’ which is all the things the brain does taken together, is mostly based on informed speculation – though some would say based on a lot of imagination, or worse, psychoanalysis and philosophy.

Any project that tries to synthesize or create intelligence with computers based at all on how the brain works (much less ‘the mind’) is also working on partial information, a lot of speculation, and perhaps no small amount of fantasy. That’s why some argue, at least for now, that it’s futile to attempt analogous imitation of the brain. What we should be doing is simply creating ‘intelligence’ of various kinds, however we define it, with whatever techniques our current level of technology can permit.

Others say, ‘Let the experimentation begin.” It’s the process that counts. By addressing the issues, both current and past, confronted by Artificial Intelligence, we should at least be able to hammer out some sort of workable framework toward computational intelligence. This will inevitably be composed of several approaches. As it is, the project recognizes these areas of research: The nature of mind (What is it?), memory (how does it work?), and body (combining computer science with physical science). By patching together what we do know about the brain with what we know can be done with computers – throw in some interesting new paradigms such as quantum computing and parallel processing – and who knows what insights about reproducing intelligence through computing may pop out? Of course, the project also predicts usable hardware and software.

It will be worthwhile to follow the stories out of the Mind Machine Project. Look for answers to questions like these: How much of the project seems to be focused on imitating the brain? How much of the project is taking what computer technology can do and looking for pathways to some kind of intelligence? There’s a dynamic here, which could be focusing, if it is recognized; or fractionating, if it is not.

In a way, the Mind Machine Project is a sumptuous feast of intellectual endeavor, the table well set with funding, intelligence, and purpose. However, in the ensuing Bacchanal, the diners may forget what they’re consuming. Other researchers, leaner of means but keen of eye, should be like wise old ravens watching the table for choice scraps that fall to the ground or go unguarded on the table top. Some of the tid-bits could be more productive than the main course.

The key to the value of the Mind Machine Project may be its transparency. The more fully described – including failures, dumb stuff, wild ideas, and disagreements – the more those watching from the outside may learn. To get the details like a fly on the wall (or its equivalent micro surveillance device) would be useful. It’s easy to suspect that with the MMP some of the liveliest ideas, including the pro and con discussion, will take place impromptu in the halls, at local food and watering holes, and even the small meetings. The technical details will also, for some, be critical. If details are hidden because it is hoped there will be future patents and commercial benefit – the MMP will not do well for the broader AI community.