Jeff Jonas

Prediction: Channel Consolidation

As we live life, our actions are recorded across countless channels, e.g., text messaging threads versus ATM transactions and so on.  Channel separation is why your bank doesn’t know where you were physically located yesterday and your doctor doesn’t know the contents of your work emails.  While we take channel separation as a given, channel consolidation is the trend and our society is heading in this direction at warp speed.

Channel consolidation is an essential ingredient to improving accuracy in prediction (e.g., your on-line retailer wisely inferring your interests).  For the most part, consumers love this as it makes life more efficient.  Businesses equally like this as better prediction means more efficient operations.  And for all these same reasons the public sector (intelligence and law enforcement to social services) are just as keen to enjoy improvements in prediction as well.

Facebook makes for a great example of channel consolidation.  All your emails, instant messages, status updates, past/present/and future travel, annotated photos, your social circle, memberships, self–expressed interests, and more … all bundled together in one nice little package, under your user account.  Traditionally such life details are expressed on diverse channels – unobservable to any single entity.  No more.  Facebook, with this panoramic view of its users, now likely has a substantially more complete picture of a person than almost any other single entity.

How powerful is this?  Here is one example: if you are a Facebook user maybe you have noticed the increasingly (spooky smart) relevant ads.  I get ads that read “Are you 44, a triathlete, and want abs like this?”  Or a well-timed ad over the summer when I was in Southern California that read: “Are you looking for a triathlete coach in the Orange County area?”  It is so relevant I find it very hard not to click on the ad!  (Be assured I do resist.)

The more sense Facebook makes of users, the better the service, the more folks will find Facebook irreplaceable, the more users will flock to the platform, and last but not least, the more advertisers are willing to pay.  Everyone seems the winner.

Hence, channel consolidation is inevitable primarily because it is irresistible.  [More here]

Consumers actually demand this. For instance, you expect that your healthcare provider will channel consolidate your data (lab work, prescription history, etc.) to properly care for you – or you may sue them for negligence! 

Nonetheless, it takes no leap to realize this very big and very important question: ‘who consolidates which channels and for what purpose?’

Law and policy will inevitably determine which entities can access and commingle which data (channel consolidation) and under what condition.  At the same time, I worry that the technical means to enhance privacy (e.g., Immutable Audit Logs that facilitate accountability and oversight) are not being adopted at an appropriate pace to keep up.

One more interesting tidbit: People often use and then abandon email accounts.  And I bet most of these folks consider all those communications (e.g., associated blog comments) effectively clipped off – like a tail – and no longer of record.  However, if the data lives on, and if there are features in that data that enable channel consolidation (e.g., your name and one or more additional distinguishing features) … then it is quite possible that these bodies could be raised from the dead.  Hummm…

How to prevent channel consolidation and the resurrection of your clipped tails makes for interesting conversation – but that will have to wait.

And on the lighter side: Facebook, by the way, makes use of a fraction of what they know and uses basic algorithms at best.  I realized this when not long ago I commented to my girlfriend about how damn useful the Facebook ads are getting and she pointed out that one of her recent ads stated “Is your boyfriend gay?”  What the hell!  And then a few weeks later I get an ad that says: “Is your girlfriend cheating on you in Vegas?” 

Touché!

The less likely alternative is that Facebook is using most of the data and very smart algorithms … so smart in fact that they have an advertiser intentionally making us suspicious of each other with the intent of soon dishing up a new ad that says something like “Need a private investigator to watch your girlfriend?”

Bastards!

RELATED POSTS:

Six Ticks till Midnight: One Plausible Journey from Here to a Total Surveillance Society

More Data is Better, Proceed With Caution

Puzzling: How Observations Are Accumulated Into Context

Trust Has a Half-Life

Ontology And Why I Am Not Obsessed With This Fancy Little Overrated Word

I don’t really like using overloaded words like this – words that mean many things to many people. Nonetheless, the subject of “ontology” is worthy of brief mention. 

First, one definition from Dictionary.Com:

Ontology -noun- (computer science) A rigorous and exhaustive organization of some knowledge domain that is usually hierarchical and contains all the relevant entities and their relations.

The main problems I have with spending any energy on pre-constructing well thought out ontologies are two fold: 1) change happens; and 2) there is no such thing as a single version of truth.

While some things are well suited for ontologic declaration like planets, species, currencies, dates and times (a small universe of easy stuff) … many things are not well suited for such pre-defined structural classification (a larger universe of hard stuff) as this leads to non-scalable, non-sustainable, brittle information management systems. No one makes a better case for this than Clay Shirky in this article called: Ontology is Overrated: Categories, Links, and Tags.

My opinionated opinion is therefore: The more difficult the ontologic classification effort, the greater the degree the approach is probably the wrong approach. Conversely, the easier the ontologic classification effort, the more sense this makes to me. Hence, I am leaving ontologies to others as my focus is elsewhere.

What is the preferred model? 

Automated. Untrained. Unguided. Self-organizing.

I think in many cases ontologies are best not pre-defined, more ideally the structures and hierarchies should emerge based on actual use/context. They are not static – they evolve and accumulate over time. They must also support disagreement … which will be sorted out later based on the lens (preferences, biases, etc.) of the viewing entity at a time of need. 

I am hopeful this is the first and last time I have to use the “Ontology” word.

What about the Semantic Web?  DON’T EVEN GET ME STARTED!

 

 

RELATED POSTS:

There Is No Such Thing As A Single Version of Truth

It Turns Out Both Bad Data and a Teaspoon Of Dirt May Be Good For You

Smart Systems Flip-Flop

“Macro Trends: The Privacy and Civil Liberties Consequences … and Comments on Responsible Innovation” – My DHS DPIAC Testimony, September 2008

On Wednesday, September 17th, 2008 I testified before the DHS Data Privacy and Integrity Advisory Committee (DPIAC).  The testimony transcripts were recently posted on the DHS web site.

Transcript Here  (My testimony starts on the bottom of page 9.)

In this testimony I covered a number of topics including:

I. Macro Trends

1. The world is not a more dangerous place – average lifespan continues to increase (More here: The World is Not a More Dangerous Place)

2. Fewer people can create bigger effects faster … in both directions – destruction or value creation (More here: More Death Cheaper in Future, Ludicrous Speed Billionaires)

3. Competition is driving technology adoption – notably the organization who receives the data first has the advantage (More here: Ubiquitous Sensors? You Have Seen Nothing Yet)

4. Data is being created faster than organizations can make sense of it – to this degree organizations are getting dumber (More here: Why Faster Systems Can Make Organizations Dumber Faster)

5. Commingling data for enhanced context will drive a new generation of smarter systems (More here: Smart Systems Flip-Flop)

6. The surveillance society continues to build momentum as consumers find it irresistible (More here: Six Ticks till Midnight: One Plausible Journey from Here to a Total Surveillance Society)

II. Las Vegas: What Can Be Learned

1. Las Vegas is one of the fastest growing cities in the Unites States of America (before the economic crash of course)

2. Casinos have a minimal security and surveillance budget

3. Casinos have a legal obligation, in some circumstances, to determine your identity (e.g., under age gambling, crossing a winnings threshold which necessitates IRS reporting)

4. Public and private watch lists are used – some compulsory, some elective

5. There are a fair number of scams that are tried on casinos – some scam artists are using false identities and disguises (More here: The Vegas Asymmetric Threat, Takin’ Vegas, Be Anyone in Las Vegas, Get Help Creating a Cover Story Here)

6. When employees go bad, this is particularly problematic

7. Most ‘tripwires’ are generated by alert employees who are watching, not by computers

8. An organization called Griffin Investigations provides information sharing of cheaters and advantage players (e.g., card counters) – they system is called Griffin GOLD (My old SRD company build this system)

9. One example of a watch list and insider-threat detection system known as Non-Obvious Relationship Awareness (NORA) was built for the industry (My old SRD company build this system too) (More here: IEEE Spectrum Story: Vegas 911, IEEE Paper: Threat and Fraud Intelligence – Las Vegas Style)

10. A patron can still enter a casino and enjoy a degree of activity without exposing their identity

11. There is no “predictive data mining” to spot unwanted behavior

12. Most vulnerabilities are remedied with processes not additional electronic surveillance

13. Humans are in the security decision loop – computer systems do not make security decisions, rather only promote items of interest

14. Facial recognition technology is not used at doors looking for watch listed persons

III. Towards Responsible Innovation

1. Technologists should more regularly engage the privacy community (More here: Responsible Innovation: Staying Engaged with the Privacy Community) 

2. Information sharing systems must have information attribution (More here: Full Attribution, Don’t Leave Home Without It, Out-bound Record-level Accountability in Information Sharing Systems) 

3. Data destruction requires careful planning and execution (More here: Decommissioning Data: Destruction of Accountability) 

4. Limit data transfers and use indexes to make information discoverable (More here: Discoverability: The First Information Sharing Principle) 

5. If data changes in a source system replicate this change through the information eco system (More here: Data Tethering: Managing the Echo) 

6. Where possible obfuscate data (More here: To Anonymize or Not Anonymize, That is the Question) 

7. Build high assurance accountability into systems (More here: Immutable Audit Logs (IAL’s), Found: An Immutable Audit Log) 

8. Data mining is not always good or bad, it depends on the circumstances (More here: Effective Counter-Terrorism and the Limited Role of Predictive Data Mining, Data Mining, Predicate Triage and NSA Domestic Surveillance) 

9. Link analysis, especially predicate-based, has some usefulness – although it is also wise to ‘prune’ early (More here: Hunting Bad Guys, Phone Records, and a Few Good Dead Men, Predicate-based Link Analysis: A Post 9/11 Analysis (1+1= 13), Sometimes a Big Picture is Worth a 1,000 False Positives) 

10. Low fidelity watch list entries (identities with few attributes, e.g., name only) are problematic (More here: Precision in TSA’s Terrorist Watch List, Comments on the TSA No-Fly and Selectee Watch List Process)

IV. Closing Statements

1. There are going to be more sensors, more data.  This data will be commingled for greater accuracy to serve consumers and to protect countries.  What data is collected and when … will be the debate.  Once data has been collected, the holder has the obligation to make sense of it.

2. The most fundamental principle I have synthesized from my many conversations with folks in the privacy advocacy community is this one point: “Avoid consumer surprise.”  (More here: Where Possible … Avoid Consumer Surprise)

3. Finding professional bad actors involves the detection of weak signal.  The computational remedy for weak signal involves observing and commingling transactions that are unanticipated by the bad actors.

4. Hence the tension.

[End of Testimony] 

I wish I would have also made this point:   

I am concerned.  If organizations don’t start implementing more privacy-enhancing technology I think research and productization in this area will decrease in coming years.  This would be a bad thing. 

RELATED POSTS: 

My Testimony Before The DHS Data Privacy and Integrity Advisory Committee (2005) 

RELATED PAPERS: 

Giannino Bassetti Foundation: Jeff Ubios

Transparency, Privacy and Responsibility: An Interview with Jeff Jonas 

Heritage Foundation: Paul Rosenzweig/Jeff Jonas

Correcting False Positives: Redress and the Watch List Conundrum 

Cato Foundation: Jeff Jonas/Jim Harper

Effective Counterterrorism and the Limited Role of Predictive Data Mining  

Steptoe & Johnson: Stewart Baker

Anonymization, Data-Matching and Privacy: A Case Study  

IEEE Security and Privacy: Jeff Jonas

Threat and Fraud Intelligence: Las Vegas Style 

Puzzling: How Observations Are Accumulated Into Context

If the goal is to make substantially more sense of data – the only way forward is CONTEXT ACCUMULATION.  

This is so important … I have been feverishly looking for a better way to explain in plain English how real-time, streaming, contextualization works.  The very best analogy I have come up with to date is that of assembling jigsaw puzzles.  The parallels are uncanny.

Puzzling

A pile of puzzle pieces are in a box.  You cannot look at the picture on the cover of the box, as that would be cheating and not like the real world.  

You grab a piece out of the box and look at your table space.  It is the first piece.  So it really does not matter where you place it.  The second piece is now in hand.  Does this relate to the first piece?  Probably not.  So it is placed stand alone elsewhere on the table.  This was an assertion. You decided this second piece is not related to the first piece.  

Soon there are many free standing pieces scattered across the table space - none at this point have been associated (snapped together).  Now you have the next piece in hand, eager to find its mate.  Do you attempt to physically match it with every possible piece in what would be a very brute force and time-consuming manner?  No.  You notice the piece in hand has at least one discriminating feature, some red and white on one of the puzzle edges.  Glancing over the table space you look for pieces with a similar distinguishing feature.  You find three such candidates.  Your attention is now narrowly focused on just these three pieces.

Comparing the piece in hand to the each candidate, you are assessing confidence.  And at the end of this process you have come to a decision point: a new assertion.  It is either (A) a match, (B) not a match but possibly a member of a family of pieces that hopefully will converge as new puzzle pieces arrive, or (C) at this time, has no apparent relation whatsoever to any other pieces.

You only connect two pieces when you are sure.  You would never get out a hammer and force the piece.  In this regard you are favoring the false negative – only connecting the pieces if you have a high degree of certainty.

Let’s say this latest piece finds a match.  You are now sitting there with a new puzzle unit (two pieces).  Looking this new unit over you ask yourself, “Now that I know this, are there some other puzzle pieces that can now be associated to this unit?”  This involves the same process: candidates, confidence, and assertion.  There comes a point where an assertion or set of assertions are made and now it is time to move on to the next piece.

Some pieces produce remarkable epiphanies.  You grab the next piece, which appears to be just some chunk of grass - obviously no big deal.  But wait … you discover this innocuous piece connects the windmill scene to the alligator scene!  This innocent little new piece turned out to be the glue.

You can change the approach, operate on a hunch and become curious.  Sometimes you decide there is an opportunity to resolve some uncertainty.  For example, you have a cluster of pieces all appearing to be related – each having some red and white expressed.  With this new interest in mind you don’t grab the next random piece but rather shuffle through the pieces in the box looking specifically for red and white pieces.  The goal: with the right few pieces this whole portion of the puzzle may resolve.

Luckily you have only been snapping pieces together when you are sure.  Otherwise, your puzzle would be not only a mess but more importantly it would not be evolving towards any degree of clarity.  But no matter how careful you have been … every now and then you grab the next piece out of the box and the home you find for it causes you to realize that two earlier piece should not have been connected at all.  It catches you off guard for a second, but upon closer inspection you realize you stuck these pieces together inappropriately.  Shame, shame.  But luckily, this past error (a false positive) was discovered and corrected – thanks to the new observations.

The puzzle you are working on today happens to be a pretty big puzzle with what appears to be thousands of pieces.  And without the cover of the box you are unsure of its final size (e.g., 1’x 1’ or 3’x 3’, or bigger).  One thing is for sure: you have to leave some space between pieces that do not connect; therefore, you need a table bigger than the final puzzle size.  Notably, after a great deal of work, there is a point when the in-progress puzzle reaches a maximum of required workspace.  After this tipping point, new pieces have a higher likelihood of finding mates and consolidation than not.

As the working space of the puzzle begins to collapse, not only does context become richer, but the computational effort of figuring out where the next piece belongs becomes more efficient despite the fact there are more pieces on the table than ever.  Assertions become faster and more certain. So much so … those last few pieces are as fast and easy as the first few pieces!  [I have seen this behavior in one of my systems … an absolutely phenomenal event with extraordinary ramifications!]

But is it really this easy?  No.

There may be more than one puzzle in the box, some puzzles having nothing to do with others.  There may be duplicate pieces, pieces that disagree with each other, and missing pieces.  Some pieces may have been shredded and are now unusable.  Other pieces are mislabeled and/or are exceptionally well crafted lies.  

Nonetheless, you will never know what you know ... unless you contextualize your observations!

On a Slightly More Technical Level:

1. When I speak of Persistent Context, this is synonymous with the ”work-in progress puzzle” where the net sum of all previous observation and assertions co-exist.

2. Semantic reconciliation (e.g., identity resolution) is but one of several steps of contextualization processing, albeit one of the most important ones that asserts ”same” or ”not same.”

3. Contextualizing observations is entirely dependent on one’s ability to extract and classify features from the observations.  Feature extractors fall woefully short today.  I’ve hinted at what I think will fix this in previous posts and more about this later.

4. Using new observations to correct earlier assertions is an essential property I have been referring to as Sequence Neutrality.  When systems favor the false negative, sequence neutrality most frequently discovers false negatives while the discovery of previous false positives are far and few between.

5. Non training-based, context accumulating systems with sequence neutrality have this behavior: the puzzle can be assembled first pass, without brute force, where the computation cost of the last pieces are as easy as the first pieces, while having no knowledge of what the picture looks like before hand, and regardless of the order in which the pieces are received.

This is what I am working on these days.  It is very exciting and I enjoy talking about such things.  Feel free to comment or email me questions.  I answer every email I get.

MUST READS:

Algorithms At Dead-End: Cannot Squeeze Knowledge Out Of A Pixel

Context: A Must-Have and Thoughts on Getting Some …

Photosynth / Infosynth

Smart Systems Flip-Flop

Federated Discovery vs. Persistent Context – Enterprise Intelligence Requires the Later

To Know Semantic Reconciliation is to Love Semantic Reconciliation

 

OTHER RELATED POSTS:

The Fast Last Puzzle Piece

Entity Resolution Systems vs. Match Merge/Merge Purge/List De-duplication Systems

Accumulating Context: Now or Never 

Big Breakthrough in Performance: Tuning Tips for Incremental Learning Systems

Streaming Analytics vs. Perpetual Analytics (Advantages of Windowless Thinking)

Enterprise Intelligence – My Presentation at the third Annual Web 2.0 Summit

The Triadic Continuum

It’s All About the Librarian! New Paradigms in Enterprise Discovery and Awareness

Intelligent Organizations – Assembling Context and The Proof is in the Chimp!

The Phone Call is Coming From Your House! Context is King

More Data is Better, Proceed With Caution

How to Use a Glue Gun to Catch a Liar

It Turns Out Both Bad Data and a Teaspoon of Dirt May Be Good For You

There Is No Such Thing As A Single Version of Truth

Sequence Neutrality in Information Systems

The Dissenting Numerati

There is a new book out called The Numerati by Stephen Baker.  In a nutshell, this book describes machines that are making sense of more and more data about people. As a result, these machines are making more and more decisions for people.  The “Numerati” are the people who build these machines and their algorithms.

This book brings to light our journey: machines measuring and directing people and how, as this continues to evolve, it will lead to some pretty interesting consequences … good and bad.

Anyway, Baker names names in this book such as Rayid Ghani (Accenture), Josh Gotbaum (Spotlight Analysis), James Schatz (National Security Agency), Eric Dishman (Intel) and a handful of others as Numerati.

I have a bit of a cameo in the book too (pages 131-37, 140-41, 150-53 and a few other mentions).  Also, in this NPR interview (staring at 13:30) Baker describes me as “the dissenting Numerati” which, I still think is a good thing.

And after careful inspection of the references to my work, I must say the author he nailed it.  I love that – as there is nothing worse than something untrue making its way into print.  

Baker did a fantastic job.

RELATED LINKS:

Transparency, Privacy and Responsibility: An Interview with Jeff Jonas

Effective Counterterrorism and the Limited Role of Predictive Data Mining

RELATED POSTS:

Ubiquitous Sensors? You Have Seen Nothing Yet

Responsible Innovation: Designing for Human Rights

Responsible Innovation: Some Things are Best Left Un-invented

Responsible Innovation: Staying Engaged with the Privacy Community

The Fast Last Puzzle Piece

The notion that the more data, the slower the system – ain’t always true.

My favorite way to explain this very important phenomenon involves the familiar process of assembling a jigsaw puzzle.

The first piece you take out of the box and place on the work surface requires very little computational effort.  The second and third pieces require almost equally insignificant mental effort.  Then as the number of pieces on the table grows the effort to determine where the next piece goes increases as well.  But there is a tipping point where the effort to determine where to place the next piece gets easier and easier … despite the fact the number of puzzle pieces on the table continues to grow.

Well isn’t it interesting, although obvious, that those last few puzzle pieces take nearly as little effort as the first few!

I have witnessed this.   

This has a slew of ramifications.

This does not apply to all domains.  This behavior requires: (a) observations from the same universe; (b) observations with enough features to enable contextualization; (c) observations in which these features can be extracted, enhanced and classified; (d) sufficient saturation of the observational space; and (e) enough smarts to stitch these puzzle pieces together.

Context accumulating systems, fed appropriate observations, can be expected to have this behavior.

 

RELATED POSTS:

More Data is Better, Proceed With Caution

Context: A Must-Have and Thoughts on Getting Some …

To Know Semantic Reconciliation is to Love Semantic Reconciliation

Big Breakthrough in Performance: Tuning Tips for Incremental Learning Systems

 

When Risk Assessment is the Risk

It is risky business to use risk assessment systems which produce alarms at a pace faster than the organization can evaluate and handle them.  

The problem often takes one of these two forms: giant batch reports with so many pages it is an almost impossible task for analysts to manage each scored risk item; or real-time alerts that pile up in queues whereby analysts pick off the top while the backlog keeps growing.

All of these potential alarms lying around in reports or queues – each possibly a smoking gun – really must be viewed as potential evidence that could be used against you if stuff goes wrong.  These unhandled alarms each evidence that the organization did not take a risk seriously and evidence that suggests someone was incompetent or negligent for not amassing the necessary resources to evaluate each and every alarm in a reasonable timeframe.  Judgment day arrives after something bad happens when this incriminating – or even just suspect – evidence comes to light during forensic analysis.  “Look the bad actor is right here on page 32,912 … had you appropriately staffed, this disaster would have never happened!”

I have a few suggestions for those organizations dealing with this problem already and those that would like to avoid this dilemma in the first place.

1. Favor the false negative – wherever possible don’t err on the side of alarm.  While, obviously, you can’t ignore what appears to be a real, serious threat, err on the side of innocent behavior.  Let new observations catch the false negatives.  More about this here: How to Use a Glue Gun to Catch a Liar.

2. How one calculates relevance and when to register an alarm is crucial.  On the technology side the objective is: the very next item in an alarm queue is the most important item for review.  And in this model, no matter what … there is no reason to produce more alarms than there are available resources (e.g., analysts, systems) to deal with them.  Therefore, especially on day one, configure your risk assessment engine to produce alarms appropriate to your individualized risk, staffing, and ability to respond. Then, as your resources have increasing bandwidth, consider increasing the alarm sensitivity.

3. When it comes to addressing the unresolved ambiguity that percolates under the alarm sensitivity, use tertiary data for automated disposition.  Instead of having an army of analysts wrestling each “maybe” to the ground it makes more sense to locate what data is needed and let that data disambiguate these low grade risks en masse.  Most risk ambiguity will be released as harmless, while a far smaller percentage will promote up to an actionable alarm.  Example, if the phone book holds the evidence necessary to resolve some risk ambiguity (i.e., to learn the middle names are different thus concluding two identities are unrelated) – and a person would normally look here – then maybe the system can look there on its own. Lots of ambiguity can be solved with lots of data – but the question is what data and with what legal and policy justification?

4. The most “bulletproof” approach involves ... lawyers.  In this case, one would contract with a law firm with expertise in information security law to bring expert engineers aboard for the analysis, technical testing and findings, and the ensuing debate which results in new corporate policy.  All ‘discovery’ – including of negative findings that could be used against an organization in the future are then protected by attorney-client privilege and, therefore, are most likely to be protected from disclosure.  And by the way, this also creates an “advice of counsel” defense to any future litigation.  If you want to go this route I’ve got a few names, just ask.

My advice: Rarely is it a good idea to ask for a detailed report of every possible risk.  So be careful what you ask for.

OTHER RELATED PAPERS:

Correcting False Positives: Redress and the Watch List Conundrum

RELATED POSTS:

Consequences of False Positives in Government Surveillance Systems

Sometimes a Big Picture is Worth a 1,000 False Positives

 

FOO Camp 2008 – How to Beat Google! (At Search)

Three fellow Foo-Campers and I held an open meeting session at this year’s Foo Camp entitled “How to Beat Google!”

 

Here are the highlights of what I wanted to say about beating Google at search.  Note: I actually didn’t get the chance to say these things at the session because it spun out in a different direction.  But, had I had my way, I would have made these observations.

 

1. It’s not about search.  I am convinced the future is going to be less about search and more about location-aware, context-sensitive, relevance detection … published to the user in real-time. This will happen via some form of Perpetual Analytics whereby, “The data will find the data and the relevance will find the user.”  In the future users will benefit more regularly from pushed intelligence than they will benefit from pulled intelligence (search).

 

2. ‘Pixel’ sorting has natural limits.  Google page ranking is like a high-speed document sorter – each document being somewhat like a pixel.  Herein lies the problem, there are limits as to how smart algorithms can be when treating each document discreetly because there is a limit to the  knowledge you can squeeze out of a pixel.  Material improvement to search is going to require extracting document content and placing this content into context first: pixels to pictures.  For example, search for a common city name and you’ll get, say, 100k entries containing documents with such a city name.  In five years, the same search might produce 200k document entries even though there are exactly the same number of cities with that name in the world.  With information in context, the search engine would know there are only three such cities and have the documents organized based on this grouping.  The point being, as indexed documents grow, search engines which simply rank documents are at risk of producing less relevant, or diluted, results over time.  There is a stark difference between an ordered list of puzzle pieces and a work-in-progress puzzle with a high degree of pre-assembly and organization. 

 

Is it just me or are search results getting a little less perfect as time marches forward?

 

3. Custom crafted lenses.  One way of producing better search results than Google would involve using personal/local context to enhance ranking.  I think of the Google search as a lens through which we look for ordered Internet content.  Today, for the most part, this lens is a general-purpose lens … meaning my results and your results are similar if not the same.  One way to materially advance search is going to involve custom crafted lenses, meaning your results are tailored to your life.  For example, if searching for souk (market area) in Dubai, while at the same time your calendar has entries about an upcoming trip to Dubai, involving a stay at the super luxurious Burj Al Arab hotel, then the first result might be Souk Madinat Jumeirah the closest souk to the hotel.

 

4. The query is the data.   Google keeps the data about documents in its indexes and the queries elsewhere. While this is how most systems operate, I say the better model is to store queries (some if not all) in the same place as the data – as if it is data – because it is data.  Some data points to documents (URLs) and other data (queries) point to users or sessions in which the question was asked.  What is better about this is (a) the ability to instantaneously discover new content related to yesterday’s questions; (b) the ability to instantaneously detect when the same questions are being asked (because the queries find the queries); and (c) when data and queries are commingled in the same data space, the result scales very well.

 

5. Bad news: Time is of the essence.  Google clearly wants to cement users to the brand, making it just too costly for consumers to switch search engines.  As such, it will be easier to compete with Google today than tomorrow.  So hurry.

 

6. Good news: There is little chance Google can deliver data finds data and relevance finds the user at ingestion speeds (item 1) and information in context (item 2) any time soon.  Two reasons: (a) they would have to figure out how to do this and do this at scale and this is not trivial (b) it would require substantial re-engineering, I suspect.

 

7. Names matter.  Google’s name has such a nice ring to it, so much so that I think that if someone develops a moderately better mousetrap for search they will still lose unless their brand name flows off the tongue just as smoothly.

 

Nothing against Google, of course.  I use it every day.  It’s a great product.  Nonetheless, I think competition is healthy and I would not mind seeing other content providers enjoying some of that “search” market share.

 

On a side note: I wonder if Google has a formal historian.  Google is an amazing success story and I hope future generations will be able to read and enjoy the back-story.

 

RELATED POSTS:

Why Faster Systems Can Make Organizations Dumber Faster

It’s All About the Librarian! New Paradigms in Enterprise Discovery and Awareness
You Won’t Have to Ask -- Data Will Find Data and Relevance Will Find the User
What Came First, the Query or the Data?
What Do You Know? Introducing Perpetual Analytics

 

Smart Systems Flip-Flop

Certainty often shifts with observations over time. And this is good.

The smartest an organization can be is directly related to the net sum of its perceptions. These perceptions come in the form of observations – observations collected across the various enterprise sensors (e.g., transactional/monitoring systems).

The second most important factor for smarts is the ability to determine how these observations relate to each other – for lack of a better word – contextualization.

But smarts requires much more than just available data and good correlation. Two additional critical elements of smart systems are:

1. An ability to make assertions based on new data points

2. An ability to use new data points to reverse earlier assertions

If everything you ever learned was held in your head as a probability, your ability to think quickly on your feet would be drastically reduced. Deciding that two objects are the same (aka Semantic Reconciliation) is an example of an assertion. Like human thinking, smart systems make assertions based on previous experience.

Smart systems also have to be able to undo earlier assertions made in error. If a new observation is in fact evidence that invalidates earlier assertions, these earlier incorrect assertions must be corrected (there are some caveats, more on this at another time).

Once presented with compelling new data, systems that cannot flip-flop on previous certainties … are dumb. The same goes for humans.

I thought my youngest son’s birthday was the 5th of the month for the first five years of his life – celebrated his birthday on the 5th, filled in forms using this date, and so on. This went on until I saw his birth certificate where to my surprise I discovered his birthday was really on the 2nd of the month! Having to explain the change of date to him and others made for an interesting conversation. When revealed this to my boy, I used the good news and bad news model – the good news being he was a little older than we thought – kids love that. Should I someday get evidence the birth certificate was in error … well then … I’ll have to revise the ground-truth yet again.

With this point of view, when politicians accuse each other of “flip-flopping,” I always laugh to myself. As it is the person who refuses to change his/her point of view – despite ANY and ALL new evidence presented – that scares me the most.

When I speak of Sequence Neutrality I am referring to exactly this characteristic whereby new data points remedy earlier assertions. Basically, had one known the new data point earlier, would the outcome of any earlier assertions be different or were there assertions that could have been made that were not? While this is hard to do on real-time transactional streams at high data rates, nonetheless this is essential behavior and hence something I have spent years thinking about and working on.

Incremental learning systems must have a high degree of sequence neutral processing, or there is little chance they will be smart.

On a more technical note: As most systems are not smart, sequence neutral, flip-flopping systems … downstream systems getting fed data (e.g., data marts and case management systems) have not been designed for the ever-changing recontextualization of previous reports. A new breed of systems – with different design parameters – are going to be needed to take full advantage of upstream smart systems.

RELATED POSTS:

Big Breakthrough in Performance: Tuning Tips for Incremental Learning Systems

Context: A Must-Have and Thoughts on Getting Some …

Accumulating Context: Now or Never

USC School of Cinematic Arts, "Imagine the World in 2050"

In April, 2008 five IBM scientists including myself appeared at the USC Film School on a panel discussing what the world would look like in 2050. Picture here.

The general topic was “putting the science back in science fiction.”

We each had five minutes to articulate how technologies might affect us in the year 2050. The rest of the session was dedicated to Q&A. I presented a newly crafted hand-drawn PowerPoint deck entitled “Macro Trends and What They Mean for 2050.” 

In summary, I made these observations about the future:

1. Big advances are generally not the result of any single technology but rather the result of combinations of technologies.

2. Good news: The world continues to become less dangerous. Today, we live longer than any time in the history of man – and this trend will continue. More here: The World is Not a More Dangerous Place.

2050 Prediction: Your doctor is 102 and this is not weird.

3.0 Fewer people can create much more, much faster … and more easily.

3.1. The bad news is faster death: Today, it takes less than 50 people and less than $100,000 to manufacture – and infest humans with – the reanimated 1918 Spanish Influenza virus. Some estimates place the death toll of such an event at 160 million people. That is an awful lot of death for relatively little cost and effort. More here: More Death Cheaper in Future.

3.2. The good news is faster wealth: On the other side of the coin, wealth can now be created faster than ever. For example, Marc Zuckerberg of FaceBook went from zero to over a billion in net worth in under three years. More here: Ludicrous Speed Billionaires.

2050 Prediction: Your 14-year-old neighbor makes $10B from their bedroom.

4.1. Surveillance societies are not only inevitable, irreversible … but more importantly they are irresistible! You love location-based services on your phone and you will love RFID chips in your sunglasses so you can find them if you lose them. More here: Six Ticks till Midnight: One Plausible Journey from Here to a Total Surveillance Society. 

4.2. Sensors become ubiquitous … not due to governments … rather this is caused by commercial enterprises as they compete for consumers who are eager to adopt any and all technologies that help them optimize their lives. More here: Ubiquitous Sensors? You Have Seen Nothing Yet.

4.3 Ubiquitous sensors result in countless piles of data. But to make real sense of this data it must be commingled. As such, these piles of data will have a tendency to converge into one pile of data. More here: More Data is Better, Proceed with Caution.

5. When disparate information is stitched together context emerges – much in the same way individual puzzle pieces are difficult to make sense of in isolation. Point being, the more puzzle pieces that can be associated, the greater the overall picture emerges. This stitched-together data will be stored in a pre-assembled form, in very large databases, ready for use. But unlike today, the enormous power of such “information in context” will not be in the hands of a few privileged (e.g., multi-billion dollar organizations) … in the future, such capabilities will live in the “network cloud” and be available to the masses. With such, the future will not be like Tom Cruise in the movie “Minority Report” where he stands in front large screens and searches for data using his hands to navigate. Nope, that would be old school.

2050 Prediction: Collective intelligence will locate what you need to know … and tell you!

To bring this prediction to life I presented the following example: The collective intelligence system in the network cloud will know the precise geo-coordinates of where you are physically standing right now. As well, imagine if it is also amassing data related to the current activities of migratory birds. While at one moment these two data sets may have no nexus … the moment a sensor presents some wind speed data to the collective intelligence system … this new data point being the puzzle piece that connects your location data and the birds’ migratory data. The cloud then will immediately push real-time relevant insight to you, the consumer. In this case, you are told to “jump to the right one foot.” Seconds later, to the your delight, a bird poop falls from the sky … splats on the ground … exactly where you were standing! 

6. When collective intelligence serves you and your doctor … you are going to love it. But when it serves the police looking at you … you are going to hate it. 

7. And that is the truth about the future … it’s going to be love/hate. 

Here is a related news story:

C|Net – Imagining the Tech World in 2050

Here is a related 7 minute YouTube video:

Imaging the World of 2050