684 karma · joined January 18, 2011
Take the electrician example: what happens if an unlicensed electrician kills himself when performing work? Does his family get to collect from your homeowner's insurance? Can they sue you? Are you criminally liable?
What about when you sell the house? Should you have to disclose that someone unlicensed did the wiring? What if you don't and the house ignites in a fire one night caused by improper wiring, killing the future owners, are you responsible? Is the electrician?
And you don't even want to touch the potential for abuse... regulation causes a lot of problems, I completely agree. But it serves to protect the public; we're effectively paying for protection against deception that capitalizes on our lack of knowledge in a specialty. That's worth protecting.
Your opinion that a Wachtell attorney isn't worth >$1,000/hour is immaterial. As a matter of law, what they're charging is not unreasonable, because you're paying for more than the amount of time an attorney spends typing on a word processor. You're paying for the firm's experience and reputation. Moreover, many lawyers would have actual knowledge of the firm's fraud, were it to occur. By not reporting this information, they risk being disbarred and losing their livelihood. I don't generally like to endorse the efficient-market hypothesis, but it appears highly unlikely that there's a giant conspiracy that results in systemic under-reporting and non-enforcement of the rules...
Regarding your second, the fact that you pay your company's taxes is irrelevant. "You" can be substituted by any other shareholder or partner; the shareholders/partners are not the corporation. The corporation is its own, independent legal entity. Because it's fictitious, it cannot make decisions, which means (as a matter of law) it's incompetent to stand trial.
Some jurisdictions may have rules deviating from this to allow corporate pro se representation by 100% shareholders in small claims matters, but they would be the exception.
In terms of "institutionalized fraud," you are totally wrong. The legal profession is one of the most heavily-regulated professions that exists today. The duties imposed on lawyers--which are legal obligations to act or refrain from acting--should not be taken lightly. For example, Rule 8.3 of the Model Rules of Professional Conduct[1] (adopted in whole or in part by 49 states) provides:
(a) A lawyer who knows that another lawyer
has committed a violation of the Rules
of Professional Conduct that raises a
substantial question as to that
lawyer's honesty, trustworthiness or
fitness as a lawyer in other respects,
*shall* inform the appropriate
professional authority.
(emphasis added). Note the word "shall:" it means that a lawyer who has actual knowledge that another lawyer (regardless of jurisdiction!) violated the rules MUST report the violation to the appropriate authority.The reason I explain this rule in particular is because of Rule 1.5, which provides in excruciating detail a lawyer's ability to collect fees from a client.[2] Everything you said is wrong.
Here's the short, plain English version of Rule 1.5: the lawyer must communicate the fee arrangement to the client before the start of representation (except when charging an ongoing client the same rate in subsequent matters). There's a blanket prohibition against "excessive fees" and "minimum fees," and a set of factors used to determine whether a fee is reasonable.
Reasonable fees are judged by time and labor, the issues' difficulty/novelty, the lawyer's experience/reputation/abilities, the nature and length of the lawyer-client relationship, whether the lawyer must turn away other cases, whether it's a fixed or contingent fee basis, and the customarily charged fees. When dealing with contingent fee arrangements, they MUST be reasonable and signed by the client, but may not be used in criminal cases. Similarly in criminal cases, a lawyer may not ask for incentive fees, and is proscribed from taking a percentage of publication rights (i.e. Casey Anthony's lawyer asking for a percentage of any subsequent book deal in exchange for his time) until after all appeals have been exhausted. Contingent fees also may not be used in domestic relations.
When dealing with division of fees in law firms, it's not actually a "division of fees" in the legal sense. The lawyers are salaried employees of the law firm, which is hired to represent the client. Again, when you hire a law firm, you are not hiring a single lawyer; you're hiring the entire firm. Division of fees occurs when a single billing client is covering the fees of two or more lawyers who are not in the same firm. When this happens, division is permitted as long as it's proportional to the services each lawyer (or firm) provides to the client, and the client must agree to this allotment in writing.
Further in this thread, you announced that you should have a right to represent your company in litigation. That's absolutely ridiculous, and in criminal matters in the United States, is dissonant with the Sixth Amendment. In such cases, the corporation has a right to assistance from counsel. Because corporations are fictitious entities incapable of self-representation and thus incompetent, corporate pro se representation is impossible.
But, there are good reasons why a client shouldn't have a right to non-lawyer representation. The Model Rules of Professional Conduct attach an enormous burden to lawyers by compelling them to disclose to the authorities another lawyer's violation therefrom. Without strong industry self-regulation in this form, there would be no way to protect clients from invidious representation.
---
[1] http://www.americanbar.org/groups/professional_responsibilit...
[2] http://www.americanbar.org/groups/professional_responsibilit...
The problem facing people who intend to work with data that does not yet exist becomes one of feature selection: what data matters and how do we use it? For NLP tasks, does stemming matter? What about part-of-speech tagging? Some classification problems are not linearly separable, which makes certain kernel methods impossible without using (and knowing to use) the kernel trick.
In the end, I think my reply here is tautological: ML is too complex to be transformed into a set of APIs a la Google Maps and Google Search.
Sure, prediction APIs could arise that give detailed use cases for each algorithm, but then there's a problem with the fringe cases: you might not know that two pieces of data are so heavily correlated that they completely shatter a conditional independence assumption, for example.
As a hacker who originally subscribed to the belief that a thorough understanding of machine learning was overkill, it is without hesitation that I admit being 100% wrong. The truth of the matter is that when it's done properly, artificial intelligence and machine learning ought to be inextricably linked with your core business processes.
Contracts of adhesion prevent us from negotiating which services we receive at what price. We can either accept Facebook's terms or not use the service. Period. Usually, this isn't a problem, but Facebook started off small. It grew into a service with network effects strong enough to actually raise a viable argument that we agree to the terms of service under duress and therefore never manifest assent.
It's a unique legal situation: generally, people offended by a company's practices can boycott. However, Facebook has turned us into the product, and while we can leave, our friends might not. The underlying purpose served by Facebook requires its community, a community that grew before establishing the invidious privacy practices.
The technological difficulty of building a Facebook replacement is compounded by challenge of convincing not just one user to switch, but enough of them to guarantee a tipping point. Not discounting the technological difficulty, I think more of the focus should be on an "exit" plan from Facebook. Google+ nailed the technology, but completely blew their Facebook evacuation plan.
More specifically, the presence of an all-knowing middleman/service undermines trust; no matter what, we will never trust Facebook as much as the members for whom we maintain a profile. Facebook wants us to completely open our private lives to a corporation notorious for its lack of transparency and openness. Facebook is entirely self-serving, as evidenced by its metastasis into even more of our lives. Masquerading as a selfless company innocently trying to connect people, Facebook continues to shove invidious technologies down our throats for one reason only: $$$.
That said, I truly hope that its FTC settlement portends a turning point for the social web.
This data is just too dangerous, even in spite of innocuous intentions. Data breaches happen, National Security Letters remain threats, and targeting ads based on a user's current location will tempt anyone. You will fill your users with confidence if you remove your ability to group the data points.
Moreover, you shouldn't discount the importance of historical barometric data. At some point, it might no longer be practical to store the data in its most atomic state, but maybe rolled up to a periodic snapshot (maybe hourly) of some geographic region.
What data would you be looking to store for longer than a month?
I'm not sure how you're storing the data, but I would suggest to store only the GPS coordinates and barometer reading, normalizing both to have the same approximation error.
In that regard, users would receive the service in exchange for something of value to Facebook--information. Their current Statement of Rights and Responsibilities requires users to grant Facebook a "non-exclusive, transferable, sub-licensable, royalty-free, worldwide license to use any IP content that you post on or in connection with Facebook."[2] To me, this is unconscionable, since the service you receive remains constant, while what you 'pay' (the consideration) varies with your level of usage. While one could reasonably argue that the quality of service you receive increases with respect to your level of interaction (i.e., you can't chat with friends until you connect with friends), there's a certain level of unforeseeability inherent in Facebook's intended use at the time you disclose that information (such as making a friend request, wall post, liking a page, etc.)
There are some even more troubling issues, as well. Interacting with third-party services via the application platform (by installing apps) gives those services access to your information. Once they have it, you're bound by their terms of use and privacy policies. I'd love to see the exhaustion rule/first-sale doctrine[3] discussed vis-a-vis data privacy, since I'm unsure whether it would apply to "data." We saw this discussed tangentially by the web sites attempting to create marketplaces to sell used MP3s[4]--the first-sale doctrine allows you to resell used books, used CDs, etc., but can it allow service providers to sell its copy of "used" data? What about the putative buyers of "used" data--are they bound by Facebook's Special Provisions Applicable to Developers/Operators of Applications and Websites? All of these are questions with answers that could have an extremely disruptive impact on Facebook's entire profit model.
[1] http://www.telegraph.co.uk/technology/facebook/8917836/Faceb...
[2] https://www.facebook.com/terms.php
[3] http://en.wikipedia.org/wiki/First-sale_doctrine
[4] http://www.techdirt.com/articles/20110218/12432213167/yet-an...
This troubles me, because information being collected and 'anonymized' is somewhat of a misnomer. I think many would contest any assertion that data capable of disassembly into its constituent parts (assuming each constituent part to be a unique combination of variables represented in the aggregate) comports with anonymity. I'm generally not one to praise governmental regulatory agencies for their technological prowess, but the U.S. Department of Health & Human Services really "got it" with HIPAA's de-identification standard.
HHS understands the importance of aggregating healthcare data and conducting statistical research, and does not let HIPAA preclude this from occurring. Instead, HIPAA outlines a "safe harbor" approach that limits a covered entity's criminal/civil liability in the event of a breach if and only if the covered entity removes 18 identifiers and has no actual knowledge that the remaining information could identify the individual. These identifiers include names, dates, geolocational codes covering populations less than 20,000, etc. Alternatively, covered entities may opt to use a 'statistical' approach by hiring a qualified statistician (or other scientific expert) who can use acceptable analytic techniques to conclude that the risk of identifying the person from the disclosed information is very, very small.[1]
A safe harbor approach to large-scale data privacy would be absolutely wonderful. Using statistics to prove the anonymity of data being collected/used by Facebook is nonsensical, since by default, we've given them a non-exclusive, transferable, sub-licensable, royalty-free, worldwide license to 'use' anything we submit to them. Contracts of adhesion, in my opinion, are more of the problem since there's no good way to 'make change' for the information that you submit. For example, someone very active on Facebook might arguably be 'worth more' to the company than someone who isn't, but both receive the same product.
EDIT: A safe harbor approach to large-scale data privacy isn't even unknown. See COPPA[2], for example.
[1] http://ecfr.gpoaccess.gov/cgi/t/text/text-idx?c=ecfr&rgn... 45 CFR 164.514(b)(1)--(2).
[2] http://en.wikipedia.org/wiki/Childrens_Online_Privacy_Protec...
Article 56 of the United Nations Convention on the Law of the Sea [UNCLOS] provides states with sovereign rights to exploit the resources contained within the EEG. If Blueseed wants to connect to undersea cables (or provide another mainland direct link), the ships become artificial islands because of the connection to the seabed. Because the provisions of Art. 60 of UNCLOS apply mutatis mantandis to artificial islands, it might not matter whether the start-up is an Art. 56 installation or a ship exercising its "high seas" freedoms. This protection extends to up to 200 nautical miles from the coast, so to be "truly" free, you'd have to go past the EEZ and likely operate an offshore airport for ferrying your employees.
As for the point of entry issues, I'm afraid I don't know anything about that. They'd likely be rendered moot by the need for an airport, though.
While 24 nautical miles is the upper boundary per international treaties, coastal states are free to adopt a contiguous zone less than 24 miles (such as to respect other coastal states' boundaries). In 1999, however, the United States announced that it was extending its contiguous zone to 24 nautical miles to protect against the infringements mentioned above.[2]
[1] http://www.un.org/depts/los/convention_agreements/texts/uncl... Sec. 4, Art. 33.
Of course, in practice, implementing an algorithm that performs well in experiments will only be slightly polished before putting it into production. Even in spite of the relative low cost of EC2 hosting, developers trying to maximize their ROI might find that the equation balances in their favor by maximizing efficiency on each core and decreasing the amount of memory that can be used. I admittedly don't know a lot about scaling experimental code into enterprise-level code, but I do know that at some point, it must reduce to a cost-benefit equation, where the costs of EC2 hosting is compared against the costs of development/debugging time in languages like C++. This, of course, reduces to the specifics of your business needs.
These are all choices that the user has to make. For something as seemingly simple as a decision tree, you can see why some knowledge is required before embarking on any machine learning mission.
Understanding your needs (or your company's needs) is where people with PhDs make their money. Machine learning isn't a panacea, and we won't be seeing a one-size-fits-all approach for awhile. Even though data has become more accessible, it might be noisy, incomplete, streaming, partially labeled, etc. This is why understanding exactly what you're trying to model with these algorithms is crucial and why "just applying" them is impractical at best and misleading at worst.
For what it's worth, the relatively recent KSR v. Teleflex case gives us the current opinion of the Supreme Court in non-obviousness battles. In that case, they slid the scale back to favor the Graham[1] test for non-obviousness, rather than the controversial teaching-suggesting-motivation test, which tries to determine whether something in the prior art suggests the desirability of combining the claimed subject matter, and thus its obviousness. Now, the Graham test (which was a 1966 decision) allows for empirical evidence of non-obviousness, to wit: commercial success, long-felt but unsolved needs, and failure of others to invent the patented subject matter.
All of these are issues that need to be litigated in federal court, an exceptionally expensive endeavor that no corporation will take without a very strong incentive.
[1]: http://en.wikipedia.org/wiki/Graham_et_al._v._John_Deere_Co.....
Their patent pledge[2] promises that they will not assert their rights against any open-source or home-grown initiative, which is why Stanford et al. are able to get away with hosting their own content in this format.
It's a harsh reality that by itself makes the case for patent reform.
[1]: http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sec...
[2]: http://www.blackboard.com/about-bb/patents/patent-pledge.asp...
That's just it, though: they don't get legal protection for publishing the idea. They get legal protection for patenting it, which requires that the idea be more than an idea--it has to be an invention. The USPTO is charged with the task of ensuring that the invention is novel, non-obvious, useful, and adequately reduced to practice. Basically, they're giving you an exclusive right for giving someone "instructions" on how to practice your invention. Whether the USPTO does a good job at this is obviously the issue, and my answer is that they don't, which comports with your overall thesis.