Lunch Talk: Superintelligence

A panel discussion with leading AI experts and business leaders about the challenges and opportunities presented by Superintelligence.

Panelists: Bart Selman (Cornell), David Chalmers (NYU), Elon Musk (Tesla, SpaceX), Jaan Tallinn (CSER/FLI), Nick Bostrom (FHI), Ray Kurzweil (Google), Stuart Russell (Berkeley), Sam Harris, Demis Hassabis (DeepMind).

Overview:
00:00. Yes, No, It’s complicated
03:10. Timescale (Elon at 5:45)
07:07. How to slow it down
14:04. Risks and mitigations (Elon at 32:14)
37:00. Upsides (Elon at 51:18)
Q&A
52:44. Democracy 2.0
54:14. Bad guys
56:43. Democratising AI (Elon)

lunchtalk, intelligence, problem, system,

Stanford CSP BUS 74 [Principles of Invention and Innovation], Session 4 Quiz 1

On July 19, 2016, Bloomberg Technology News reported that Google used its DeepMind AI technology to reduce power consumption in the company's data centers:

In recent months, the Alphabet Inc. unit put a DeepMind AI system to reduce power consumption by manipulating computer servers and related equipment like cooling systems. It uses a similar technique to DeepMind software that taught itself to play Atari video games, Hassabis said in an interview at a recent AI conference in New York.

The system cut power usage in the data centers by several percentage points, "which is a huge saving in terms of cost but, also, great for the environment," he said.

The savings translate into a 15 percent improvement in power usage efficiency, or PUE, Google said in a statement. PUE measures how much electricity Google uses for its computers, versus the supporting infrastructure like cooling systems.

Question 1. Using the system model, name the functional element that DeepMind technology helps to improve directly.
Question 2. Based on what you know about the improved element, describe other functional elements within the same system.

A creativity technique from science fiction

I'm reading Rainbows End, by Vernon Vinge. The novel takes place in a future where security spooks play mind games with social, neurobiological and genetic threats. As usual, I pay attention to nuggets of creative wisdom. Here's one of them:

For Robert Gu, real creativity most often came after a good night's sleep, just as he roused himself to wakefulness. That moment was such a reliable source of inspiration that when he was having problems with writing he would often go the pedestrian route in the evening, stock up his mind with the intransigencies of the moment ... and then the next morning, drowsing, review what he knew. There in the labile freshness of new consciousness, answers would drift into view.

I use a similar technique to get an insight into most difficult problems. During the evening, or several evenings in a row, I do a lot of background work on analyzing the problem, exploring its system aspects, trade-offs, dilemmas, etc. often, when I wake up in the morning, I go over my analysis again and discover a new idea that was not there before.

This experience is consistent with the earlier Lunch Talk video in my blog where neuroscience professor Vincent Walsh recommends to become obsessed with a problem in order to come up with a creative solution. I would add that system analysis techniques really help with getting obsessed in a right way, especially when you have to do it for an inter-disciplinary group of creative people.

Facebook is taking over Google in sourcing the flow of news

Fortune runs an article showing Facebook's influence growing in the news segment:

...it’s clear that search has hit a kind of plateau and isn’t really growing any more as a referral source for media. Meanwhile, Facebook’s influence has “shown it’s on a continued growth trajectory."

Source: Forbes.com (click images to enlarge)

The competition for advertisers' money between Facebook and Google is heating up. We should expect that Facebook will make further inroads into information segments other than news. Although it's too early to pronounce Search dead, its dominance on the web no longer translates directly into the mobile space, especially, when users spend more and more time on social. (Based on system analysis, we anticipated this trend in Scalable Innovation, Chapters 20-22).

It is also somewhat surprising that Twitter is such a non-factor in the race. Despite the "freshness" of their links, they don't have enough users to play the game. Furthermore, unlike the Facebook's, Twitter connections don't have the strength of social relations.

tags: mobile, information, control, google, facebook, twitter, system

Lunch Talk: Why Information and Diversity Grows (Cesar Hidalgo at TED)

MIT professor Cesar Hidalgo considers how to deal with diversity and complexity.

tags: lunchtalk, control, system, science, math, economics

True Detective S2 vs S1 - an inventor perspective

True Detective, Season 2 turned out to be a bit of a disappointment and I wanted to understand why. Reading TV critics and blogs didn't give me much insight beyond the typical "oh, the story was not good" or "oh, the director was not good" or "oh, the pace of the action was too slow", etc. Therefore, I decided to put my inventor hat on and compare the two Seasons as Systems. I applied to both TV series the same system analysis techniques I always use in my invention workshops.

I started the analysis by laying out each story as a system of perspectives. That is, each layer of narration in the series represents a Source of information for the viewers (Scalable Innovation, Section 1). Each Source covers the reality of events on the ground. Paradoxically, it turned out that despite Season 2 has more main characters than Season 1, it also has fewer unique Sources of representation.

In Season 1 we had four key perspectives (2 "real" and 2 "virtual"):
1. Detective Rust Cohle (Matthew McConaughey);
2. Detective Marty Hart (Woody Harrelson);
3. The Official Investigation - a narrative presented by the official police investigation;
4. The Narrator - a director narrative presented by the chronology of events described in an "objective" manner by the video camera and background characters.

The nature of perspectives was also different. All of them were extremely smart but with different flavors. Rust Cohle could be characterized as "weird smart". Marty Hart - "down-to-earth smart". The Investigation - "bureaucracy smart". The Narrator - "visual and story smart". Furthermore, we had variations of each perspective shifted in time and space. In addition to the mystery of the crime, we, as viewers, had to reconcile and process the mysteries of all these Sources that gave us complimentary and conflicting information. The structure of the system provided us with a intricate, intriguing pattern.

Importantly, the system of different perspectives felt natural due to the fact that detectives Cohle and Hart managed to solve their case _because_ they had different perspectives. They also had conflicts _because_ they had different perspectives. Since they broke multiple official rules — and The Narrator shows us how and why — the official investigation perspective provided us with an explanation why a standard bureaucratic police approach to detective work would not solve the mystery. As a result, we had a system of contrasting and explaining Sources that formed a complex but consistent, natural whole.

Finally, the perspectives were not just narrated from a character's point of view. They were SHOWN from that point of view. In short, Season 1 did an excellent job executing the rule "Show, don't tell".

Season 2 had more main characters, but fewer perspectives. Essentially, there was just one perspective - the Narrator, who guided us and the camera through the story. Basically, we had one Source which kept switching microphones and cameras for every character to tell his or her line.
Although the story itself was, arguably, more complicated and somewhat more mysterious, the system of perspectives was no different than in a regular criminal TV piece. As a system, Season 1 turned out to be flat.

Overall, the actors in both Seasons played great, stories were interesting, suspension was adequate for a crime drama, and camera work excellent, especially, the LA aerial shots in Season 2. Unfortunately for Season 2, the script didn't provide a system structure that could support a real thriller of the Season 1 caliber.

tags: system, source, control, entertainment, method

Lunch Talk: Daniel Kahneman on The Machinery of the Mind

Daniel Kahneman, author of Thinking, Fast and Slow, on The Machinery of the Mind. Kahneman is Emeritus Professor of Psychology and Public Affairs at Princeton University and the winner of the 2002 Nobel Prize in Economics.




tags: psychology, creativity, system, lunchtalk, science

Finally, I have my magic formula for analyzing problems and inventing new stuff!

Actor(s) {A}
in Context(s) {E},
using Stuff {S} and incurring Cost(s) {C},
perform Action(s) {P}
produce Result(s) {R},
which counts as Outcome(s) {O}

Several immediate thoughts:
- Dilemma resolution techniques apply to Actor(s) and/or Context(s).
- 10X analysis applies to Cost(s) and/or Result(s).
- The Three Magicians, esp. the second one, move us between Results and Outcomes and provide different levels of Contexts.
- Elements quantize!

There's a lot more but I have to think about how to describe this 6-dimensional innovation space in detail.

Also, we live in a world abundant with high-tech startups, while Christinsen's "Innovator's Dilemma" was formulated for a world with sparse startups. Today's successful high-tech companies — Google, FB, Apple, etc — feast on this abundance.

tags: invention, innovation, system, model, dilemma, 10x

Solving Detection problems with the System Model

Since the publication of Scalable Innovation, I've had many discussions about the System Model with readers and students. Intuitively, they think that the left-to-right dimension corresponds to Space-Time. That is, the Packaged Payload moves from the Source to the Tool via the Distribution.
Although this intuition is correct, it's not the only one we can use in the system model. Importantly, we can think about the second Time axis — the vertical one — that applies to a particular system element. In this case, the element becomes dynamic, e.g. changes over the time or moves in space.

In other words, the system model covers processes that involve repeated transactions and/or evolution of a particular physical element over time. For example, the Sources in the picture above represent the same physical Source at different points of time. This approach works really well for solving Detection problems because it allows us to identify an element based on its behavior. That is, we can extract Aboutness by controlling and/or interacting with the element.

Since this is not intuitively clear, I probably need to develop examples that explain this use of the System Model.

tags: model, system, detection, book

Invention of the Day: the Tea Bag

In our book, Scalable Innovation, Max Shtein and I introduce the concept of Packaged Payload, an element of the system that encapsulates an essential ingredient — mass, energy, information — that moves within the system. The Packaged Payload is critically important for the functioning of the system.

Paradoxically, most people don't see it in their everyday lives because engineers do a good job at hiding the functionality. For example, we can't see AC electricity because it's securely insulated within the wires. Also, we can't see data packages because they are transmitted over wireless connections. We can't see ocean shipping containers either because we buy products in retail, not in bulk.

Explaining the Packaged Payload to students and inventors can be a challenge; therefore, Max Shtein and I are always on the lookout for good examples. Today Max sent me several pictures — a Packaged Payload galore, as he called it — that make the concept easier to grasp. For example, in the picture above you can see chocolate milk and tea packaged in single-shot bags.

Remarkably, the tea bag was invented more than 100 years ago (US Patent 723, 287), but it got popular relatively recently when a new system of fast-food establishments, e.g. McDonald's restaurants, Starbucks Coffee shops, and others became a common place.

US Patent 723, 287, issued March, 1903.

The tea bag represents the Packaged Payload in a food distribution system. Similarly, many other food items are available for one-time use. All of them are standardized for mass production, delivery, and dispensation (see below).

Thank you, Max!

tags: packaged payload, distribution, system, example

Why Tesla gives away its patents to copycats?

Wired reports on the latest in patent news:

Tesla CEO Elon Musk announced today that his company will not “initiate patent lawsuits against anyone who, in good faith, wants to use our technology.” In plain English, that means that if other car companies want to produce electric cars, they can use Tesla’s technology to do it, and, in turn, advance Musk’s sustainability vision.

What's the significance of that?

I looked up Tesla's patent portfolio on the US PTO website this afternoon: 156 issued US patents and nothing of great importance there. In comparison, BMW and Toyota have thousands of patents. In a patent fight against its direct competitors Tesla has a slim to none chance to win. Therefore, giving away a weak patent portfolio is not a big loss for Tesla. On the other hand, if the company succeeds inducing the competitors to give up their patents, that would be great! Besides, Tesla promises to gives its patents to those who want "to use our technology" only. Interesting. This brings our attention to Tesla's new business model.

Recently, the company announced that it is going to build a huge battery-making plant in the US. For this project to be successful, Tesla needs economies of scale: a lot of electric cars made by those who use Tesla batteries and electric drive technology. Selling batteries to a potentially huge market would be more profitable than trying to enforce weak patents in a small market. Giving away the patents is a shrewd PR move by Elon Musk. This reminds me of an ancient Chinese stratagem called "Tossing out a brick to get a jade gem." It means "Bait someone by making him believe he gains something or just make him react to it ("toss out a brick") and obtain something valuable from him in return ("get a jade gem")."

===
In the system model terms (see our book Scalable Innovation), Tesla intends to make money on the battery, i.e. the Packaged Payload, while encouraging others to build more electric cars, i.e. the Tools.

Invention skills are very similar to wisdom

Today I learned that:

In recent years wisdom has received much scientific attention in psychology. It can be defined as expertise in coping with difficult or unsolvable life problems.

Wisdom is a multidimensional psychological construct, including dimensions like
(1) change of perspective,
(2) empathy,
(3) perception and acceptance of emotions,
(4) emotional serenity,
(5) factual knowledge,
(6) procedural knowledge,
(7) contextualism,
(8) value relativism,
(9) uncertainty acceptance,
(10) long-term perspective,
(11) distance from oneself and
(12) reduction in level of aspiration.

Wisdom can be learned and taught (source: DOI: 10.1159/000321580)

Practically, all items on the list apply, except #12. Inventors must have aspirations to change the world.

Facebook wireless wake-up patent (US 8,644,892)

Today, US Patent Office awarded Facebook a patent on a wireless device with a passive RFID tag that can trigger different power modes. In one scenario, when your iPhone is in sleep mode it receives a wake-up call from an RFID reader, powers up the main battery, and transfers the data from the tag to the device.

Again, the easiest way to explain the patent is through the train analogy we used in Scalable Innovation (Chapter 3). 

Imagine that instead of wireless devices and Radio Frequency signals you are running a train station operation. You also have a telegraph machine that allows you to receive and read telegrams from neighboring stations. It's early in the morning; no major load-unload processes are in progress; the only half-awake person in the building is one Thomas Alva Edison, your trainee telegraph operator.

Suddenly, his telegraph machine starts chattering and he receives a telegram from a neighboring station that a big train is departing toward you. Mr. Edison reads the telegram, wakes up your station crew, and reads the contents of the telegram to the station manager.

In the Facebook patent, the wireless device is your train station in wake up or sleep mode. The RFID tag is Edison with his telegraph apparatus. First, he can receive a telegram that no trains are coming and send everybody home. Then, the tag receives a wake-up signal from an RFID reader (the neighboring station) and transfers the contents of the signal (the telegram) to the main memory with a processor (the manager), which is configured to run a pre-defined program. That's it. The rest of the wording in the patent is for obfuscation purposes.

The invention fits the Telegram before the Train invention pattern we consider in detail in Chapter 25. 

tags: patent, packaged, payload, control, system, example, facebook

Amazon patents - Content Management

Amazon continues its push into content management technology. Their  US 8,639,817 patent issued on January 27, 2014) is the latest in a series that covers delivery of digital media.

The patent applies (among other things) to delivering ads based on the original content. In their terminology, a first set of users consumes the "real" content, while a second set gets [relevant] ads. For example, Claim 2 reads:

In claim 3, they continue using the anticipatory approach we found earlier in their other patents, which cover delivery of physical goods.

With physical goods, Amazon describes a scenario where the system
1) routes packages to a general geographical location in anticipation of demand;
2) re-routes packages to a specific address, based on a customer order.

With virtual goods, Amazon patents a scenario where the system
1) delivers content to a content delivery network in anticipation of content demand;
2) delivers content to a specific user device, based on user requests or targeting logic.

In system terms, Amazon creates a smart Distribution network, which sits in between the content providers and users. We model the arrangement in Scalable Innovation, Chapter 25. Anticipating Control Problems. Because Amazon collects a lot of information about both content (Packaged Payload), users (Tool), and providers (Source), it has the ability to determine and anticipate consumption patterns. The patents are a strong indication that business value migrates from the Tool -- Source axis, to the Distribution -- Control axis.

Similarly, Facebook, Google, Twitter, NSA, and others sit between users and content providers (e.g. other users). Remarkably, Amazon doesn't cover social networking scenarios in their patents. Vice versa, Facebook doesn't talk about content management in their patents.

tags: patent, system, aboutness, distribution, control, business, value, amazon, facebook

Streamternet - a new term for describing our post-web world.

I've been struggling for a while to find a name for the new, post-web reality of the Internet. In Scalable Innovation (Section 3), we explain why we think that the web is dead, but we don't use any new word to mark the new reality. 

The core idea is that instead of sending files, we now deal with streams, e.g. video or update messages: Youtube, Twitter, Facebook, High-Frequency Trading, etc. In system terms, we see a dramatic change in the Packaged Payload and the intensity of its flow. 

Fundamentally, the iPhone and Google Glass are Streamternet devices. Their zoomable interfaces allow us to zoom in and out of the stream of information, and see it, e.g. in 2D, 3D, or 4D. The difference between the web and Streamternet is that time flows differently in them. That is, the intensity of data transactions is 100X higher on the Streamternet.

web, streamternet, system, packaged, payload, tool, 10X

Facebook patents secure upgrade of a wireless mobile device.

Facebook got a nice patent (US 8,631,239) that covers a secure software upgrade of a wireless mobile device. According to the specification, the system uses a public key to authenticate the software delivered over the air (OTA).

Wireless connections are notoriously unsafe and prone to hacker interception. The Facebook solution enables a service provider to perform a reliable upgrade over an unreliable channel. It's highly likely that in the future most software upgrades, especially in the enterprise environment, will be done using this approach - simple and powerful!

Unfortunately,  the patent itself has an important flaw: it does not define the term "endpoint", which figures prominently in claim 1. Moreover, in Fig 1B it uses a different term "System Front End (120)."

As I noted several times before, the company's quality control over their patenting process seems to be spotty, at best. A simple document search would allow them to spot and fix the definition problem.

1. A method comprising, by one or more computing systems: executing software from a first partition of system memory; requesting an over-the-air (OTA) software update from an endpoint; receiving a manifest for the OTA update; downloading a payload pursuant to the manifest; installing the payload into a second partition of system memory; and rebooting, pursuant to the manifest, to the second partition of system memory, wherein rebooting to the second partition of system memory comprises authenticating a bootloader signature with a bootloader public key.

Brief system analysis: the manifest represents the "Aboutness"; encrypted software update - Packaged Payload; device  - Tool; a process that runs on the device to verify authenticity - Control; endpoint - Source; over-the-air channel - Distribution. Overall, it's a textbook example of system composition (Scalable Innovation, Chapter 2). To solve the problem, the inventors use Separation in Space - one of the key TRIZ principles.

Model-wise, it is quite similar to my patent US 7,529,806. They have a different payload, but the aboutness is managed and created for the same purpose. I should use the Facebook patent as a system analysis homework assignment in BUS 74 this summer.

In view of the Nortel patent and invention principles listed above, the Facebook patent can be attacked as "obvious."

tags: patent, invention, innovation, security, mobile, enterprise, system, model, aboutness

Lab Notebook: Emotions, Metaphors, and Creativity.

When people talk about their emotions they often use metaphors because it is difficult to express how you feel in words. In everyday life we solve this language problem by using common experiences with objects as metaphors. In the 1980s, UC Berkeley psychologists Lakoff and Kovecses studied the connection between emotions and metaphors. The metaphor of a fluid in a container (hot, cold, boiling, flowing etc.) turned out to be associated with all major emotions: anger, fear, happiness, sadness, love, lust, pride, shame, and others. We are so used to dealing with water and other liquids that when something unknown and unfamiliar is presented as "water" it becomes easily accessible to our minds.

Here's a great example how a difficult concept can be explained to a lay audience: Robert Shockley, the co-inventor of the transistor, shows the work of his new incredible electronic device as water flowing over the dam. In the picture, charged elements (electrons and holes) become "water," while the electro-magnetic field that controls the flow of the current is shown as a dam barrier that can be raised or lowered to control the flow of "water."

Remarkably, the metaphors of UP and DOWN are also frequently used to describe emotions, e.g. "I feel a bit down today," or "Lighten up!".

Max and I should try to use the common metaphors discovered by Lackoff and Koveses to explain how our system model works. In Scalable Innovation (Chapter 3) I use "train" to explain a difficult patent, but water would probably work even better.

tags: creativity, psychology, emotion, system

Lab Notebook: a computer security disaster waiting to happen.

Legacy computing systems are notoriously difficult to retrofit for new threats. Recently, Target got hacked, with in a possible loss of 110 million user records (including mine). Now, the banks have discovered that their good old ATMs need to be upgraded to new software to be ready for new hacking dangers. According to Bloomberg Newsweek,

When ATMs were introduced more than 40 years ago, they were considered advanced technology. Today, not so much.
Inside every ATM casing is a computer, and like all such devices, each one runs on an OS. Microsoft’s 12-year-old Windows XP dominates the ATM market, powering more than 95 percent of the world’s machines and a similar percentage in the U.S., according to Robert Johnston, a marketing director at NCR (NCR), the largest ATM supplier in the U.S.

For banks, investing into a massive upgrade of an old computer system would be a waste of money because customers are switching to mobile payments. We can deposit checks, pay bills, and invest with a smartphone. The only thing we can't get from the smartphone is beer physical cash. A lot of startups and established companies, including Square and PayPal, are going after this market. The success of BitCoin is also a strong indicator that physical banking is dying.

If the banks don't invest in a rigorous new security system, old ATM networks will become a juicy target for hackers. Like any other parasites, hackers love weak targets: newborns and elderly, a legacy ATM system being the latter. A computer security disaster with a major ATM network will speed up the adoption of digital currencies immensely. A great financial play would be to buy a lot of bitcoins, then hack an ATM system just to scare people into using the new technology.

tags: innovation, deontic, payload, system, money, business

Google + Nest GSV = the Web is Dead

Google announced that they are buying Nest GSV, the smart thermostat company, for $3.2B. The acquisition extends Google's push into the Internet of Things (IoT), including robotics, automatic cars, etc. Gartner's 2013 Hype Cycle report puts IoT firmly at the top of the hype chart.

Recently, Cisco added to the hype with an estimate for the IoT market of $19T by 2020. Since Google missed badly on Social Networking, the company is eager not to miss on the next big thing. With the web going away (Scalable Innovation, Chapter 20), Google needs new massive sources of data streams to process; otherwise, all their data-crunching technology could become worthless. The Internet of Things seems to fit the requirements. Although the valuations are of hype-size proportions, today's web advertisement (search) produces enough cash to finance the future S-curve.

When I worked in research on IoT concepts in the early 2000s, it was too early. We did get a number of good patents, e.g. US7,620,703, US7,257,839, US7,092,861, US7,069,345, US6,838,986, out of that work, but the inventions didn't become mass-market innovations back then. Even today, it's still too early for the mass market. Interesting questions to consider would be, What does it mean to be too early? What problems do we need to solve, so that an innovation in this particular technology market becomes real?


tags: invention, innovation, hype, google, system, synthesis, web

Facebook US Patent 8,627,506: a blunder or strategic omission?

The vagueness of Facebook patent claims keeps surprising me. Take, for example, their latest one: US 8,627,506 "Providing privacy settings for applications associated with a user profile," (Inventors: Nico Vera, James Wang, Arieh Steinberg, Chris Kelly, and Adam D'Angelo).

The patent is supposed to cover a transfer of user private information to third party apps based on friendship relationships (social graph) in a social networking system. I wanted to use the invention to illustrate the concept of "aboutness" in our system model (Scalable Innovation, Chapter 5).

Even a brief system analysis of the claims shows that the third party app does not provide any definite information about the second user to the social networking system, i.e. a key "aboutness" element is simply missing. Only in claim 4 we find a vague statement about a second user "who is connected to the [first] user in the social networking system." We don't know the nature of the connection, nor the degree of connectedness. Maybe she is a direct connection, or maybe she is one of the billion people on Facebook. Who knows...

Since the social networking system doesn't know much about the second user, it can either give out all private data or no data at all. In short, according to the patent, where third party apps are concerned privacy is non-existent; the apps are entitled to receive the first user's entire social graph. Using this graph, they can fish for other users' social graphs, and so on.

Let's give the patent the benefit of the doubt and assume that it tries to cover a minimally useful configuration with no privacy. Then, there should be at least one dependent claim that describes what information about the second user is required to determine the amount of private data transferred to a third-party app. Unfortunately, no such claim exists in the patent.

Some people believe that such vagueness — they often confuse it with broadness — is harmless. But is it? Imagine that a patent troll looks up Facebook patent applications when they are just published and files a patent application that covers a scenario with a more specific privacy information exchange. When the troll gets its patent issued, it can sue Facebook for damages because in a real social networking system specific "aboutness" for the second user has to be exchanged to determine privacy boundaries. As a result, Facebook is going to be rightfully punished for sloppiness and vagueness in its patent portfolio.

Complaining about trolls and software patents is easy. Getting your patent house in order is more difficult.

tags: system, aboutness, patent, invention, social, networking