Jeff Jonas

Your Movements Speak for Themselves: Space-Time Travel Data is Analytic Super-Food!

It doesn’t matter who you say you are!  Where you are (space), when you’re there (time), and your movements over time (travel) are closer to the truth.

I’ve seen a lot of data in my life, and I’d like to think I have a decent grip on what can be accomplished with data and analytics.  However, I recently stumbled upon some facts that have radically reshaped my understanding of the world we are living in.  What I thought was years away is already here! Our toes are dangling over the edge of a very different future.

Now, before you get all worked up, remember: You have helped create this, most folks love this, and most will continue to eat this up despite the obvious consequences.

Mobile devices in America are generating something like 600 billion geo-spatially tagged transactions per day.  Every call, text message, email and data transfer handled by your mobile device creates a transaction with your space-time coordinate (to roughly 60 meters accuracy if there are three cell towers in range), whether you have GPS or not.  Got a Blackberry?  Every few minutes, it sends a heartbeat, creating a transaction whether you are using the phone or not.  If the device is GPS-enabled and you’re using a location-based service your location is accurate to somewhere between 10 and 30 meters.  Using Wi-Fi?  It is accurate below10 meters.

Fancy.

It should be no surprise that all this data lives in the coffers of the cell providers.  Lots of people know that.  What is new, at least to me, is that this data is being provided to third parties that are leveraging specially designed analytics to make sense of our space-time-travel data. 

With the data out and specialized analytics emerging, this infant industry is already doing some pretty amazing work. Your space-time-travel data makes where you live and where you work self-evident, and it reveals your most frequent, periodic, infrequent and rare destinations. 

The data reveals the number of co-workers that join you Thursdays after work for a beer, and roughly where you all go. It knows where these same co-workers call home, and just exactly what kind of neighborhood they come from (e.g., average income, average home price) … information certainly useful to attentive direct marketing folks.

Large space-time data sets combined with advanced analytics enable a degree of understanding, discovery, and prediction that may be hard for many people to fully appreciate. Better prediction means a more efficient enterprise and nifty consumer services. 

Cellular companies are now receiving essential insight about their customers (e.g., to better understand and predict customer churn).  Major retailers can now better understand changes in consumer behavior (e.g., how far their customers are traveling on average this month compared to previous months).  Consumers are benefiting by getting real-time traffic information so they can avoid congested roads.  (I have a colleague that thinks he is saving two to four hours a week in commute time due to this service!)

Tip o’ the iceberg.

I can barely get my mind around the ramifications. My concept about what comes next shifts almost daily now.  A government not so keen on free speech could use such data to see a crowd converging towards a protest site and respond before the swarm takes form –  detected and preempted, this protest never happens.  Or worse, it could be used to understand and then undermine any political opponent.

A stalker might be questioned just days after he starts and before his victim is personally aware of it – detection previously beyond human capacity.  Maybe it’s not a crime in this case, and it turns out to be just a private investigator with poor tradecraft hired by a suspicious husband.

Such a surveillance intensive future is inevitable, irreversible and as I have said before here … irresistible.

Why?  Companies must be competitive to survive and consumers have quite the appetite for almost anything that optimizes their life, especially if it’s cheap or free.  For example:

Tuesday afternoon your [free] Gmail account advises you that your buddy Ken is going to be 15 minutes late to the pool hall this coming Thursday, unless he leaves work 15 minutes early … which he has only done twice in seven years.  Brilliant!

Your Starbucks drink of choice (a grande vanilla soy latte in my case) is handed to you the instant you pull up, and you did not call ahead nor did they ask.  Priceless!

When powerful analytics commingle space-time-travel data with tertiary data, the world we live in will fundamentally change.  Organizations and citizens alike will operate with substantially more efficiency.  There will be less carbon emissions, increased longevity, and fewer deaths.

I think people should know about this imminent new age we are marching into.

[Theatrical pause.  Breathe.]

Now I’m going to step back and address some questions you may have, using the good news/bad news format.

Good news: The space-time-travel collected by the cellular network carriers is de-identified when provided to these third parties for privacy reasons in that it does not include your name, address, phone number, etc.; rather, unique identifiers are assigned to transactions from the same device so that trends can be measured.

Bad news: If you were to provide your home, work and one other address (e.g., gym, school) in most cases, with just these data points, you are re-identified.  With just a few days of space-time-travel activity, your top three or four more frequently visited destinations become self-evident, and without a whole hell of a lot of effort you could be re-identified through a tertiary data set like a credit header.

Good news: There is so much data being produced, a lot of transactions are tossed aside, are sampled and summarized to make the computational effort feasible.  Historical data also falls off the back of the wagon (ages off the system) rather quickly.

Bad news: The competitive nature of this emerging business model will likely require these organizations to make more sense of more data faster.  Cloud computing and new classes of algorithms will make it possible to keep more transaction detail, keep it longer, and commingle it with other large and very interesting secondary data sets (e.g., phone books and property records).

Good news: So far there are only a handful of companies already entrusted with this data.

Bad news.  It may not be good news that only a few companies do this.  If only one company can monitor the consumer foot traffic of all Nordstrom stores in near real time,  this would be an unfair advantage in terms of predetermining its financial condition before anyone else.  As I learned from countless conversations with my friends at the ACLU, very powerful tools in the hands of a few is not often a good idea without one hell of a lot of oversight and accountability. And even then, this is no panacea.

Good news: Some of the organizations holding space-time-travel data are fully aware of the privacy consequences and are offering consumers the ability to opt-out – meaning, if they get a transaction about you it will be permanently removed from the system and all future correlation.

Bad news: If by chance a snapshot of sufficient detail had been sold off to another party before the opt-out request, then the toothpaste is out of the tube.  Data tends to replicate, more about this here.

Good news: Not any old mom and pop operation can get into this business.

Bad news: That won’t be true for long.  Suppose an aspiring entrepreneur makes a compelling proposition to a number of parties holding space-time-travel data.  Anticipating free analytics and a cut of the future action, the parties work a deal. For computing power this entrepreneur simply hops onto Amazon’s EC2 cloud and partners with a data aggregator to get some tertiary data and what do they have?  An ultra-sexy prediction engine.

Good news: People tend to appreciate location-based services, which is why they are opting in.

Bad news: Sensitive information about people is no longer under their own control.  As well, a number of well held secrets (e.g., your hideout) evaporate overnight.

Good news: If you want to escape the consequences of having your space-time-travel being graphed by others, here are some options that come to mind:

(a) Stop using mobile devices;

(b) Use multiple devices e.g., use one device only at work, and only a land line at home – all mobile devices being off at all other times (never moving around with a device on) – being sure these mobile devices are registered to someone other than you – and if you need to use some kind of device while on the move or at other locations. see (c) below;

(c) Unregistered, cash-purchased, disposable devices – used once then discarded (or recycled!) – although in some cases you can use the device a few times, but you better let some fancy software (which I may have to invent) advise you what is safe usage and what is not.

(d) If you can figure out locations on earth where only one cell tower exists (and you are not moving between towers and never using GPS or Wi-Fi) you will probably live safely under the radar – unless you are a way bad mofo and others know it, in which case, you are ‘going down’ anyway because there are more tricks (expensive) which will be levied against you.

Bad news: Few are willing to be this inconvenienced.  And if only a handful of innocent, clean living folks go to this same effort that the bad guys MUST employ … well crap, that in itself may be considered by some to be signal.

Net Net: My guess is most consumers don’t fully realize how their space-time-travel data is accumulating and congealing.  I hope consumers come to appreciate how all of these nice conveniences of life are delivered. And I hope they will continue to enjoy these while they make better informed decisions, especially with respect to their privacy.

However, without a feedback loop consumers may never fully appreciate what can be gleaned from their space-time-travel trail. Therefore, one way to enlighten the consumer would involve holders of space-time-travel data to permit an owner of a mobile device the ability to also see what they can see:

(a) The top 10 places you spend the most time (e.g., 1. a home address, 2. a work address, 3. a secondary work facility address, 4. your kids school address, 5. your gym address, and so on);

(b) The top three most predictable places you will be at a specific time when on the move (e.g., Vegas on the 215 freeway passing the Rainbow exit on Thursdays 6:07 - 6:21pm -- 57% of the time);

(c) The first name and first letter of the last name of the top 20 people that you regularly meet-up with (turns out to be wife, kids, best friends, and co-workers – and hopefully in that order!)

(d) The best three predictions of where you will be for more than one hour (in one place) over the next month, not counting home or work.

I think Google’s Android and Latitude products might be able to move on something like this first.  It would then be cool if other holders of space-time-travel data followed.

On the subject of privacy and civil liberties consequences, privacy by design is essential.  And for those of you with ideas in the area of policy or technology, I would be most appreciative if you would share these thoughts with me … sooner rather than later.

I will continue sharing perspectives about these ideas and the apparent consequences with my many friends in the privacy community, the defense/intelligence community, and media.  (Surprisingly, their feedback so far has been quite similar.)  I am also speaking with the organizations amassing and analyzing this space-time-travel data to learn more about what is possible.  From the perspective of the analytic engines I create, this space-time-travel data looks like “super food.”

USEFUL REFERENCE:

Internet Society: Wireless Location Privacy: Las and Policy in the US, EU, and Japan

 

RELATED POSTS:

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

Prediction: Channel Consolidation

More Data is Better, Proceed with Caution

How to Use a "Glue Gun" to Catch a Liar

Asserting Context: A Prerequisite for Smart, Sensemaking Systems

What this means: If you cannot count and associate like things and remember them, it is damn unlikely your higher-level prediction system is going to produce accurate answers.

Trajectory and velocity are required for many forms of prediction.  Example: Is the customer becoming a better customer and if so, how fast?  How fast is the voter registration roll growing?  Is the swine flu headed this way or that way and how fast?  Is on-line sentiment around your prized brand on the rise or fall?

If you think you have five customers moving slowly on an unremarkable vector when in fact these five customers are all the same person – you might be missing the fact this customer is moving on a specific vector with meaningful velocity (e.g., becoming a much better or worse customer).

Why do I speak of this?  Well, I get a chance to see some of the most advanced sensemaking systems being created and tested around the world.  And the way I determine (in about five minutes) whether they have half-a-chance of ever delivering high value is this quick and dirty assessment: Can they count discrete objects?

Take a hypothetical biosurveillance system on the West Coast of the US, which is supposed to observe the trends of a future influenza outbreak – say a new swine flu mutation.  Such a system might, for example, use newsfeeds and other available data like blogs to count incidents and locations over time.  How accurate could a system make predictions if San Francisco, San Fran, SF and the Bay Area were tallied as discreet regions?  If the system cannot tally geographically there might appear to be five cities each with mild volumes – when in fact, it is one dense region with moderate volumes.

Counting like entities (Semantic Reconciliation) is fundamental to the measurement of trajectory and velocity.

Nonetheless, seems a lot of organizations want to work on prediction first and then circle back to cover” basic counting” later.  Well, let me just say … counting is a first order activity.  Get this wrong and your prediction system will certainly miss obvious predictions, it will lead astray all downstream processes e.g., staff who are taking these predictions as inputs.  Such systems will also fail to scale.

And, to the extent an organization is in the ”we want to detect weak signals” business … counting becomes even that much more important.

Smart systems, prediction systems, sensemaking systems, situational awareness systems, incremental learning systems … whatever one calls these things … must first be able to form an opinion (aka make assertions) about context (aka count and associate) … if they are to be relevant.

On a more technical note:

1. While I am referring to determining when two entities are the same entities (e.g., same location) it is equally important to be able to assert classification (e.g., same kind of car) and relationships (e.g., employed by).

2. For some time now, I have been thinking of technologies that compute when entities are the same, similar and related as “Assertion Engines.”  Note, I do not think about these as “Estimation Engines” which would compute and persist probabilities between each and every observation collected.  Assertion-based systems take very high-grade probabilities and call it “so” … leaving some room for maybes … and calling the rest “not so.”  This, I believe, is essential to scalability.

3. Assertion engines that deal with context (determining if an entity is either the same, similar, or related) seem to misbehave if one does not favor the false negative (confident beyond a reasonable doubt).  This means one must decide it is true – and remember that it is true.  Example: Donald Conrad at 123 Main Street IS Don Conrad at 123 Main Street.  Unless of course, new evidence becomes available (at a later point in time) which might bring this into question, e.g., learning there is a Jr. and Sr. Conrad living at 123 Main Street.  This is why smart systems flip flop – changing their mind about the past.

RELATED POSTS: 

To Know Semantic Reconciliation is to Love Semantic Reconciliation

Sequence Neutrality in Information Systems

Smart Systems Flip-Flop

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

Accumulating Context: Now or Never 

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

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

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

Big Breakthrough in Performance: Tuning Tips for Incremental Learning Systems

 

Data Finds Data

O’Reilly Media has just released an excellent book called: Beautiful Data: The Stories Behind Elegant Data Solutions.  Lisa Sokol and I wrote chapter 7 entitled “Data Finds Data.” 

Here is the introduction:

Next-generation “Smart” information management systems will not rely on users dreaming up smart questions to ask computers; rather, they will automatically determine if new observations reveal something of sufficient interest to warrant some reaction, e.g., sending an automatic notification to a user or a system about an opportunity or risk.

An organization can only be as smart as the sum of its perceptions. These perceptions come in the form of observations—observations collected across the various enterprise systems, such as customer enrollment systems, financial accounting systems, and payroll systems. With each new transaction an organization learns something. It is at the moment something is learned that there exists an opportunity, in fact an obligation, to make some sense of what this new piece of data means and respond appropriately. For example, does the address change on the customer record now reveal that this customer is connected to one of your top 50 customers? If an organization cannot evaluate how new data points relate to its historical data holding in real time, the organization will miss opportunities for action.

When the “data can find the data,” there exists an opportunity for the insight to find the user.

How data finds data is a statement about discoverability, the degree to which previous information can be located and correlated with the new data. Discoverability requires the ability to recall related historical data so that an arriving piece of data can find its place, similar to the way each jigsaw puzzle piece is assessed relative to a work-in-progress puzzle. Each new puzzle piece incrementally builds upon what is knowable, at each given point in time relative to the evolving puzzle picture. Often new pieces, although important to building out the bigger picture, do not themselves bring new critical information. (On the other hand, some pieces may change the shape of the puzzle in a way that warrants ringing the bell—finding that one piece that connects the palm tree scene to the alligator scene.) It is at this moment in time, when the new puzzle piece presents the opportunity to reshape the picture, that discoveries are made. Real-time discovery replaces the need for users to think up and pose the right question at just the right time.

Organizations that are unable to switch to the “data finds data” paradigm will be less competitive and less effective.

The chapter in PDF is here: Data Finds Data (Provided here under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License)

 

RELATED POSTS:

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

To Know Semantic Reconciliation is to Love Semantic Reconciliation

Accumulating Context: Now or Never 

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

Sequence Neutrality in Information Systems

 

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