Thursday 13 December 2012
Artificial Intelligence is the cross-disciplinary enterprise of trying to do things with a computer which when done by people are said to require intelligence and which computers cannot (yet) do. (The careful reader may notice a certain degree of isomorphism with a popular definition within the extended cognition framework...)
(And comparison shoppers, here is Wikipedia’s current version: “Artificial intelligence (AI) is the intelligence of machines and robots and the branch of computer science that aims to create it.”)
So: Consistently with the bottom-up approach to AI I like to advocate in general, I don’t think allusions to the Turing test or the Singularity or whatever are all that interesting, as far as actual progress is concerned, the cognitive arts advance through innovations which are very small increments from the perspective of AI as a whole but can be quite dramatic for the topical discipline in question.
I do think that the trying (or aim[ing] to create) is an important part of what makes AI AI. Doing arithmetics also requires intelligence but has never been a part of AI since computers could do (and indeed were built to do) it properly from the beginning. And so on the way from notrespondingstilltrying to commercial viability, AI projects start being called computational whatever or whatever technology (hence legal technology). Of course the boundaries are vague and the whole boxological excercise of little use in anything other than turf-wars in academia.
And the fact that the definition refers to human intelligence just serves to illustrate the futility and question-beggitude of definitions for one simple reason: The psychological understanding of human intelligence just adds even more layers of complexity. For example, IQ tests cannot possibly measure human intelligence per se and in general. What they measure instead is a specific indicator known as the g factor (or general intelligence), which has been shown to correlate (reasonably) well with the more specific intelligent abilities.
Even if working on definitions can occasionally serve an useful purpose, personally I think in most cases the more expedient alternative is to follow Justice Potter Stewart in Jacobellis v. Ohio: “I know it when I see it.” For historical reasons, jurisprudence in Finland still has a particular affinity for concepts and definitions not really seen elsewhere to the same degree. I’m planning to address this issue in extenso at some point with the title Begriffsjurisprudenz 2.0. (Spoiler alert: may also offend ontologists.)
Saturday 8 December 2012
Personally I think all this talk about the Singularity is mostly just a distraction (and of course fodder for dystopic science fiction). Actually functioning general-purpose artificial intelligence is not simply just a matter of bytes and CPU cycles or even fully replicating the neural network of a human brain at some instant (because so much of human intelligence depends on neurogenesis and the formation and pruning of connections, processes which only a couple of decades ago were still thought to end by adulthood), and anyway it is so far in the horizon that it is impossible to use as a target. There is still a lot of work to be done in trying to make sense about the actual functioning of human cognition. (The discussion about free will and whether Libet’s experiments show that it doesn’t exist is a good example.) Even if the Singularity does arrive at some point, the interaction of humans and computers at that time will not be something we can easily imagine. (Just compare whatever you are using to read this with a completely character-based interface (your only choice thirty years ago). And I still fondly remember the sound of a good mechanical teleprinter...)
To date, AI has been most successful when trying to solve very difficult but still quite concrete problems with computational methods. My rule of thumb is that when AI starts being useful, it stops being called AI. (Hence I also prefer to talk about (intelligent) legal technology rather than legal AI.) There are many branches of computer science and other computational sciences which started out basic AI research, with language technology as just one good example.
But more importantly, as for the shorter timeframe, I totally agree with Thiel. Computers are much better than people at some tasks and legal technology has great potential for radically transforming the marketplace for legal services (for the better) in the near future. The work we do at Onomatics will hopefully a good example from the more technologically advanced end of the scale, but our domain (trademark law) is just a very small corner of the entire legal system.
All this just reminds me that I should finally get around to writing two blog posts I’ve been thinking about for quite a while, one titled “Why do computers make better lawyers than people” and the other – of course – “Why do people make better lawyers than computers”. Real Soon Now!
- Peter Thiel on The Future of Legal Technology - Notes Essay (by Blake Masters)
- Notes from Peter Thiel’s CS183 Startup class at Stanford (by Blake Masters)
- Syllabus for the Stanford Legal Technology and Informatics course
- Course reader for the same
Thursday 8 November 2012
If you only know usability from real-world usability, or more likely the lack of it (good usability is unobtrusive), here are the standard definitions:
“[Usability refers to] the extent to which a product can be used by specified users to achieve specified goals with effectiveness, efficiency and satisfaction in a specified context of use.” - ISO 9241-11For more information, see The User Experience Professionals’ Association website.
“Human-centered design is characterised by: the active involvement of users and a clear understanding of user and task requirements; an appropriate allocation of function between users and technology; the iteration of design solutions; multi-disciplinary design.” - ISO 13407
Of course I’ll also take the opportunity to mention my paper titled Software Usability and Legal Informatics (draft paper on SSRN) which I will be presenting later this month at the KnowRight conference. As far as I know, there has been very little earlier scholarship on the topic in legal informatics, but pointers are most welcome. I will be pursuing this line of research further in other articles at least over the next couple of years.
[Update: presentation now available here.]
Monday 5 November 2012
Biometrics are an old acquaintance for data protection law. However, the legal interest has thus far focused on the use of biometrics for identification purposes only, that is, using them as an unique key giving an individual access to something or tying that individual to other, non-biometric personal data. The role of biometric data as (potentially even sensitive) personal data in its own right has yet to received the same kind of attention. Biometrics can also be used for example for personalized outdoor advertising even without positively identifying its individual target. The technological development is extremely fast, and, as with many emerging technologies, law struggles with keeping up to date.Slides here.
One particularly interesting development is the combination of behavioural biometrics with face recognition. Continuous analysis of facial microexpressions based on the Facial Action Coding System (probably most familiar from the TV drama Lie to Me) is being developed for a number of purposes, such as profiling airline passengers and lie detection. For lie detection and `mind-reading' in general, simple optically based biometrics are at least as reliable (ie. not very, at least at this point) as the more widely known fMRI-based and other neuroimaging methods, while being totally noninvasive and thus easy to deploy without the consent or even knowledge of the data subject, and at a fraction of the cost. This type of use of fMRIs and neuroscience in general is already a hot topic in law, but the same questions should be understood more broadly and without commitment to any specific type or level of analysis or any particular technology. Looking for explanations on the neuronal level just confuses the non-specialist completely, thus lending neuroscience explanations their seductive allure.
And so this season’s fashion tip for all paranoiacs is to swap your tinfoil hat for Botox®, as it paralyses the muscles causing facial microexpressions, thus making the technology unreliable. Anti-facial-recognition makeup, which confuses the system by making specific parts of the face undistinguishable, is of course another possibility.
As an example case I used the AVATAR system in pilot use on the US-Mexican border in Nogales, Arizona, only since a couple of months ago. You can find more info on AVATAR here and here. I do feel the need to point out that I did not want to talk specifically about AVATAR, but about the wider privacy implications of face recognition and other biometric technologies in the long term. And to have a bit of fun while doing it, of course.
Somehow I did however manage to briefly mention an issue I seem to return to in every paper, namely the question of judicial (or in this case administrative) decision support. In the EU, the most general regulation of this issue is in the Data Protection Directive (95/46/EC), Article 15, on automated decisions, which can only be warranted by statute or for the purposes of fulfilling a contract. In either case, the system making automated decision must have a safeguard in the form of human supervision with the possibility to override. A similar regulatory scheme is proposed to continue under the forthcoming Data Protection Regulation, this time in Article 20 titled Profiling.
The requirement for human supervision is of course good and necessary, but it is by no means enough by itself. If a system makes correct judgments 90% of the time, people seem to have the tendency to infer that it is correct the other 10% of the time as well. One absurd example of this kind of uncritical sticking to procedure (without machines!) is the Twitter joke trial, which was also taken up by Ray Corrigan in his GikII presentation.
There are different ways to mitigate this. The obvious one is that, especially in a context where the system's decisions are routinely followed, the error rate could not possibly be allowed to be anywhere near 10%. To be sure, rigorous testing protocols are required. One possibility is also to open up the algorithms for review, which can be done ex ante as a part of an authorization protocol, or, especially in an individual case, ex post, or both.
Still, with Big Dada, neither of these is enough of an answer. Creating rigorous tests for real-life systems of this type is easier said than done, and “cheating” on the test by making sure at least all the known test cases work as they should is only common sense. Carving specific requirements for testing in stone is a surefire way to kill all innovation in this field.
ahem), there is considerable reluctance (or at least a hefty price tag) for this kind of openness. In a national security setting it Just Isn't Done. And in any case, the sheer complexity of the task means that access to the algorithm is of no use whatever when you are trying to board a flight but THE COMPUTER SAYS NO.
So what's my answer? Quite simple: require that decision support systems are always constructed to give explicit and, upon request, detailed reasons for their decisions in human-compatible terms, just like in a well-written court decision. This makes it easy for anyone to see if there is something completely off in the inputs or the line of reasoning, and step in and override. It also allows for a more qualitative type of testing, when not only the decision but also its rationale can be included in the evaluation. And when the system does something harmless enough (say evaluates likelihood of confusion for trademarks (smiley)), the rationales can be used for educational purposes.
So is this a case for more regulation? Even if it were, the legislator's track record in this field does not exactly promise any immediate relief. One way to solve this is to let the markets decide, but that requires an educated customer base who knows what to require and why. I guess we'll just have to wait and see.
Monday 24 September 2012
And as for myself (Anna), I have been busy with a start-up company specialized in intelligent legal technology in the field of trademark law that I co-founded with some other people earlier this year. (Less vague version coming up in October...) The company is called Onomatics and the beta version of our first product went live today. The product is an intelligent trademark analysis system for the time being usable as a trademark database but more to come later, and it is largely based on the work I have done on MOSONG starting over ten years ago. You can read more about Onomatics on the Onomatics blog (design still under construction...), on this blog I’ll try to keep the pitching (as opposed to bitching) to a minimum and focus here on the researchy stuff instead.
Still, this move has also resulted in something of a shift in my own research interests. Now, for obvious reasons, I’m paying a lot more attention to the (local) start-up scene in general and (worldwide) legal start-ups in particular. A related question in this context is the commercialization of the fruits of academic basic research in general, and as someone who has earlier worked for ages at a company founded for that specific purpose and still continuously struggling with the whole idea, I certainly might have a thing or two to say about it and some authority to stand behind my words as well.
Another more indirectly related topic is Usability. My continuous participation in the design process at Onomatics together with having to use the alternatives provided by OHIM, USPTO et al on a daily basis has been something of an eye-opener as to how crufty those services really are. (BTW, my favourite piece of this kind of legacy cruft is the Logout button in TESS.) Coincidentally, Usability and software procurement in general have now also become hot political topics in Finland with the plans to replace all the medical record and other related data systems in one ginormous EUR 1.8bn (budgeted) project, and just in time for the upcoming municipal elections. The cost of bad usability in the health sector has already been pointed out by the National Audit Office as well. Although I have mostly just followed the local debate, these questions are certainly relevant all over the world, and I would certainly appreciate any pointers to earlier work in this field in a legal informatics or regulatory context.
Sunday 24 June 2012
1. The two essential components of the registration of a trade mark are (a) the sign and (b) the goods and services which that sign is to designate. Each of those components makes it possible to define the precise subject-matter of the protection conferred by the registered trade mark on its proprietor.What this means in practice is that the same word(s) can be a registered trademark for several different owners at the same time, as long as the products are different enough. High-reputation marks are a bit different in that they receive a broader scope of protection because anyone else trying to register, say, ROLLS ROYCE for anything whatsoever has certainly chosen the mark to try to gain unfair advantage of the earlier brand’s reputation. Even if many of the trademarks you can recall might be high-reputation marks (and certainly all the most valuable brands are, which is kind of the whole point), the vast majority of trademarks in the registries are not high-reputation marks, and thus must coexist with similar marks of others’ if necessary. In the EU trademark registry at OHIM (Office for Harmonization on the Internal Market (Trade Marks and Designs)), at the top of the list (apart from single-character renditions of picture marks) there are right now 90 different entries (granted/pending/rejected applications) for FUSION, also 90 for ECLIPSE, and 89 for EVOLUTION.
(from Advocate General Bot's Opinion or draft judgement)
In order for the registrars (and everyone else) to keep track of the different products for which marks are registered, they are classified into 45 different classes according to the Nice Classification, 34 classes for tangible goods and 11 (up by three quite recently) for services. The classes are quite uneven in their scope, ranging from
Class 9. Scientific, nautical, surveying, photographic, cinematographic, optical, weighing, measuring, signalling, checking (supervision), life-saving and teaching apparatus and instruments; apparatus and instruments for conducting, switching, transforming, accumulating, regulating or controlling electricity; apparatus for recording, transmission or reproduction of sound or images; magnetic data carriers, recording discs; compact discs, DVDs and other digital recording media; mechanisms for coin-operated apparatus; cash registers, calculating machines, data processing equipment, computers; computer software; fire-extinguishing apparatus.to
Class 23. Yarns and threads, for textile use.The classification also contains an Alphabetical list of over 8000 more specific product descriptions divided into the individual classes (637 in class 9; 22 in class 23). Overall, in my opinion, the Nice Classification stands out as a relic of a bygone era designed to fully leverage the state-of-the-art computing power of index card technology. (Using words like ‘gutta-percha’ in the class headings doesn’t exactly help.) Of course, fifty-odd years ago the registries were also much more manageable in size.
In EU (at OHIM and in most of the Member States), it has been possible to register a trademark not only for a detailed list of specific goods and services (obviously covering only those) but also for all goods and services in an entire class by using the class heading as a product description. This is different from eg the US, where only specific products must be used, and trademark protection normally only extends to those products for which the mark is actually used in trade, even if the original registration also covers additional products.
Another thing to remember is that a trademark must be able to distinguish the owner’s product from those of others’. Because of this, a trademark cannot just be descriptive and quite simply name the product by its common name. Trademark degeneration, or a trademark becoming a generic name for a certain type of product, can mean that the trademark is no longer protected. (This has happened for example for ASPIRIN in the US, whereas in Europe Bayer’s registration is still valid.)
In the IP TRANSLATOR case, the CIPA applied for registration of IP TRANSLATOR as a UK trademark for ‘Education; providing of training; entertainment; sporting and cultural activities’ in class 41, which also happens to be the class heading for that class. The UK Registrar refused the application based on OHIM’s classification guidelines (Communication 4/03) on the grounds that IP TRANSLATOR is descriptive for translation services, which are also a part of class 41, even though they certainly do not naturally fall under any of the categories listed in the class heading in normal parlance. Of course there was an appeal, and this is where the ECJ was asked for a prejudicial opinion. And thus the ECJ opined:
[...]trade marks must be interpreted as meaning that it requires the goods and services for which the protection of the trade mark is sought to be identified by the applicant with sufficient clarity and precision to enable the competent authorities and economic operators, on that basis alone, to determine the extent of the protection conferred by the trade mark.In response to this OHIM issued a new Communication 2/12 to supersede the earlier 4/03, requiring for future and pending applications one extra bit of information for each class of goods and services using complete class headings, namely whether it is intended to cover the entire class or not. This alone would only address the issue raised in last paragraph of the ruling, and ignore the first two altogether. Furthermore, the second paragraph in fine read together with first one quite simply means that not all class headings are created equal. The ones that are sufficiently clear and precise standing by themselves (like class 23) are fine, but others (like class 9) are not. This would of course bring registration practices more in sync between the EU and the US, which some might argue is a Good Thing. Actually this is even spelled out quite clearly earlier in the judgement:
Directive 2008/95 [...] does not preclude the use of the general indications of the class headings [...] to identify the goods and services for which the protection of the trade mark is sought, provided that such identification is sufficiently clear and precise.
An applicant for a national trade mark who uses all the general indications of a particular class heading [...] to identify the goods or services for which the protection of the trade mark is sought must specify whether its application for registration is intended to cover all the goods or services included in the alphabetical list of that class or only some of those goods or services. If the application concerns only some of those goods or services, the applicant is required to specify which of the goods or services in that class are intended to be covered.
54 In that connection, it must be observed that some of the general indications in the class headings of the Nice Classification are, in themselves, sufficiently clear and precise to allow the competent authorities to determine the scope of the protection conferred by the trade mark, while others are not such as to meet that requirement where they are too general and cover goods or services which are too variable to be compatible with the trade mark’s function as an indication of origin.In response to this, OHIM is, for the time being, reviewing all incoming and pending applications using class headings on a case-by-case basis, while also trying to make up its mind as to which class headings are still okay, together with the local registrars. Even if registrations such as the class heading for class 9 are no longer allowed, the ones registered up to this point will of course still remain in force unchanged.
55 It is therefore for the competent authorities to make an assessment on a case-by-case basis, according to the goods or services for which the applicant seeks the protection conferred by a trade mark, in order to determine whether those indications meet the requirements of clarity and precision.
Of course my interpretation is far from being the only one around. From a trademark attorney’s job security perspective, this one by Niamh Hall as seen on IPKat would probably be the best one I have come across so far. Not only would class heading registrations be allowed for all classes, but they would only cover all products listed for that class in the Alphabetical list of the Nice Classification (and perhaps not even the latest one if the mark is older), and not just all the products that would eventually end up in that particular class, thus potentially excluding over 85% of the individual products listed in EuroClass, the product classification system used by OHIM and an increasing number of local registrars. From the perspective of legal certainty and predictability, this wouldn’t exactly be optimal.
Sunday 6 May 2012
What was so intriguing about that slide was that apart from a small number of outliers, almost all of the systems evaluated were at the bottom left quadrant, with both precision and recall (typically well) below 50%. Based on my expectations from my previous line of work, this was quite shocking, and had I known this earlier, I would certainly have worded a paper or two a bit differently as far as using e-discovery as an example is concerned. All the same, I suppose even at these figures e-discovery already outperforms all the alternatives, but there is certainly still considerable room for improvement.
What I was accustomed to was that a marketable product should deliver well above 90% in both columns or else the users would just stop using it. Whether they are 93% or 98% is not all that important, because that kind of variation is mostly just noise and depends on how well suited the test materials happen to be for that system. In particular, if you use a manually annotated test corpus as a gold standard, in a commercial scenario, you can really only use it once, because whatever is left between the actual performance and 100% are what we in the business call bugs, and they should be dealt with unless there is a good reason not to do so. And so our gold standard becomes tainted once these bugs are fixed. Which is of course no reason not to use the new figures for marketing purposes. (‘Press statements aren’t delivered under oath.’ - Jim Hacker, PM)
Performance is not just an e-discovery issue. It is raised in many other legal technology contexts as well. For example, the Swedish trade mark law start-up Markify prides itself with the ‘99% accuracy’ of its system, for example in this recent Arctic Startup profile. The actual study on which this claim is based is also available. The results are based on querying a set of 1000 actual cases of successful US trade mark oppositions and the question was whether the different services would return the correct mark (that of the opponent) when queried for the mark that is being opposed. Here are the results, and for
|Thomson Reuters Compumark||45.5%||0.21%||0.4%|
If recall is all you are after, improving on that 99% is really easy. Simply by returning the entire database for each query you can reach 100% just like that. At the same time precision naturally drops down to epsilon but so what. Of course this is not quite fair (there was no proper indication of the placing of the correct answer on the list of results), but still, just returning the desired answer is definitely not enough, at least when it is returned in a needle-in-a-haystack mode, where, even if the result is there, it is increasingly likely to be missed by the person reading the results the longer the list is. For what it’s worth, I tried to search for ‘äpyli’ (that’s Helsinki slang for ‘apple’) on Markify's system and quit after 10 pages of results at which time the trade mark of a well-known Cupertino-based fruit company had not yet shown up, and the results being shown at that point were already much further away. I suppose ‘the other [sic!] high quality paid trademark search services that can run $500 a word’ can still breathe easy.
Saturday 5 May 2012
As someone who has followed and worked in language technology for about two decades now, I see AI & law as being now in the same state as language technology was in the early 1990s. I have presented some lessons learned on how to approach real-world problems at the detail level in my robo-judge paper, so I won't go into them here. Instead, my proposal here looks into language technology as a field that has succesfully reinvented itself a couple of times already. Early on, it was only known as natural language processing (NLP) as a subfield of AI and a form of basic research rarely with any concrete application in mind. (A notable exception to this is machine translation, which also happens to be older than the term AI itself. More on that in a separate post as well as an article written jointly by me and Anniina Real Soon Now.) Then came computational linguistics, which was centered on using computational models and techniques as a tool for linguistic research. (This is where I think AI & law is now.) Of these, in particular corpus linguistics has become mainstream in virtually all subfields of linguistics, but other computational methods are now widely used outside computational linguistics proper as well. Through the 1990s computational linguistics also started to find its way into commercial applications in domains such as language checking, information retrieval, text-to-speech and vice versa, dialogue systems, and machine translation. As these real-world applications started to generate increasingly important research questions in their own right, language technology was born.
"Legal technology" as a term is not my invention. For example, in the US there has been a bicoastal biannual conference called LegalTech® since 2006. As far as I know, most of the technologies presented there are are not all that interesting from an AI & law perspective, with topics such as case management and billing platforms, synchronizing your BlackBerry with your Outlook and stuff like that, and whatever new cruft Westlaw and LexisNexis have come up with each year.
More to the point are for example the LawTechCamps arranged by Daniel Martin Katz (of Computational Legal Studies) and others in June 2011 and next week in Toronto and in the end of June in London. There is also a growing number of start-up companies in the field at least in the US, as listed just the other day on the eLawyering blog. Most of the start-ups listed seem to be working on applications having to do with contracts (possibly a sign of flock mentality from the venture capital side?). Contracts are also the target of the only out legal tech start-up I know of here in Finland, Sopima. With a large number of companies on the same domain fighting over the same market from somewhat different perspectives, it is clear that only some of these companies will be able to succeed (at least as far as the US companies are concerned, Europe is a different kettle of fish because of the different legal culture(s) and the prevalence of non-English languages). The best products have to address a real-world problem and solve it well and efficiently. Usability is another key success factor, and it still seems to be generally neglected in legal IT. Just because a certain design is a possible way to do something does not mean it is the best way (indeed it rarely is, though at least it usually is not quite this bad; required reading: Donald Norman's The Design of Everyday Things, MIT Press 1989). In particular, just replicating ancient practices from the age of pen and paper (and secretaries) and possibly adding some bells and whistles is a true and tried pattern, unfortunately. And the result is an application that takes a week-long course just to get started with it. All the same, the technically best solution does not necessarily win the game. In the end, it all boils down to the viability of the business model and the ability to make it into a reality. (Here's a convenient rule of thumb: marketing costs money, selling makes money. Close early, close often.)
So how can the AI & law community contribute to the impending legal technology boom? One approach is to take an existing, reasonably well-developed Good Old-Fashioned AI & law technology, and to find a real-life legal problem which it could possibly solve. (I'm afraid I can't come up with an example.) The other approach is to take an existing problem (= market need) in the legal community, a problem of the kind that should be solvable by computing, and looking around all over the place in computer science in search of that solution. (Here e-discovery is a prime example, though it does not travel well, and performance-wise it is quite disappointing by the language technology metrics I'm used to but at least it is still equally reliable yet faster and cheaper than people doing the same job.) Since language is in a key role in law, language technology is one obvious place to look at but it should definitely not be the only one for any legal tech company. I'm sure the next 20 years will be a lot more interesting (and profitable) for the field than the past 20.
Tuesday 17 April 2012
That is, my thesis will consist of the obligatory introduction (which also includes a couple of things I could first publish as standalone articles but can't be bothered, namely looking at WEBSOM from a legal perspective and a
- MOSONG: A Fuzzy Logic Model of Trade Mark Similarity
- From Spelling Checkers to Robot Judges? Some Implications of Normativity in Language Technology and AI & Law
- Dual-Process Cognition and Legal Reasoning
- Grundnorm as Conceptual Bootstrapping: An Ontogenetic Perspective
- Zombies and Legal Personhood
- Burdens of Proof, Probabilities, and Alberto Contador
Friday 23 March 2012
I suppose there are basically two possible aspects on innovation. The first is taking something preexisting and familiar and making it just a little better. The other alternative is a radical break with tradition and doing things differently altogether. In a sense this echoes the distinction between normal science and scientific revolutions introduced into the philosophy of science by Thomas Kuhn. But in reality (rather than philosophy) these aspects are complementary rather than dichotomous.
To take a concrete example: the shift from, say, a 32 nanometre to a 22 nanometre process in semiconductor manufacturing is not immediately visible for the computer user and from this perspective it may seem like yet another meaningless number in the computer specifications orat best a small incremental change. From the manufacturing perspective, on the other hand, shedding those extra nanometres has required enormous technological advances. One additional constraint onthe design is also the amount of heat being generated into a smaller and smaller space that still has to be dissipated through a cross-section of comparable size, leading to the invention of 'darksilicon' (powering down parts of a chip that are unused at any given moment). And on the other hand, the cumulative effect of such changes in terms of processing power, storage capacity and so on (Moore's lawand all that) enable new approaches to all kinds of problems thatwould have been quite impossible a decade or two ago. After all, eventoday's smartphones are more powerful than supercomputers in most ofthe 1980s.
The ability to deal with immense amounts of data in real time isdefinitely one of the two biggest driving forces for artificialintelligence in the foreseeable future. Recently I looked into the history of machine translation, and one of the earliest systems actually took twice as long just to do what amounts to looking up each individual word in a dictionary and stringing them together into an approximation of a translation as what it would take for a human translator to produce a correct translation. A system like Google Translate, on the other hand, has ginormous collections of multilingual documents with aligned language elements and uses them together with some heavy statistical processing to do the same job and produce at least something understandable if not correct in justfractions of a second. Oh, and the other driving force? Bio-inspired AI, or seeing how Nature has solved a given problem and trying to reproduce that in an artificial design.
As for my own work, I try to keep these different perspectives in mind, and while I occasionally like to throw words like 'robot judge' around, it is more as an abstract target (and of course also asprovocation) rather than as something I am actually concretely interested in implementing. But it is certainly helpful in trying to keep in mind the whole range of issues potentially involved in working with legal AI, and not just the issues du jour the research community finds interesting at the moment. In my opinion, one part of the problem is also that mainstream legal theory does not study law and legal reasoning in particular as a form of cognitive activity and has managed to all but ignore all the scientific progress made in both linguistics and psychology over the past fifty years, and in much of my work I take theories from those disciplines and try to apply them to questions of legal theory in a very general sense (mostly because almost nobody else seems to be doing it). Still, the best way forward for me seems to be trying to model some very small corner of the legal system using some particular technique (and I think I'm stuck with fuzzy logic at least until I've finished my dissertation) to see whether it works and then trying to see whether there are some broader conclusions to be drawn based on it. And in the best case it might even do something useful (read: marketable) at the same time. In this respect, AI & law seems to be about twenty years behind language technology.
We don't particularly need robot judges, but for example judicial decision-support systems could help actual judges in making correct and consistent decisions more efficiently and reliably, thus perhaps enabling them to spend more time on cases where uniquely human capabilities are really required. And if at the same time technology also revolutionizes the way legal services are provided (as predicted by Richard Susskind in particular), maybe the parties do not even have to go to court in the first place.
Thursday 22 March 2012
"The combination of machines and ICT has brought exponential development into the engineering world. As a result we see the emergence of autonomous machines. The complexity of tasks as well as the complexity of environments where these machines can work is steadily increasing. We can build cars that can navigate autonomously through city traffic finding destinations without any human intervention. Therefore, it is fair to say that machines have, on a functional level, already reached cognitive abilities comparable to horses or dogs. But this is not the end of development. Soon there will be no type of manual labor in which machines will not outperform humans. This is technically already true today. Currently, machines are merely held back by economic and societal constraints. The weakest of these constraints is the cost of hardware. Moore’s Law guarantees that computing power that today can only be found in supercomputers will be available in pocket sized devices in little a more than a decade. Some other constraints are more difficult to overcome. The more powerful and more complex a machine is, the more damage it can potentially create. This is the reason there are no self-steering cars on the roads yet. We have suggested a path of best practices and ethics to improve machines and reduce intentionally malicious behavior. Nevertheless, even those best practices leave us with a residual risk. This residual risk is not necessarily small. It may indeed be so large that certain types of machines will not be able to enter the market because of liability concerns. This limitation will only be overcome by the creation of an ultimate machine. For human parents responsibility and liability for a child ends with it becoming an adult. Similarly a machine can become an ultimate machine by emancipating itself from its manufacturer/owner and indeed becoming a distinct legal or even social entity. Interestingly, this can be done by creating a legal construct around this ultimate machine that in itself has economical value.
Nevertheless, the big question remains: how will our societies hold up to this rapid change? For example, currently our entire tax and social system, indeed most of our culture, is centered on the concept of work as the means of creating one’s livelihood. For example, the European Union has set a goal of increasing the part of the population (between 15 and 64 years of age) in gainful employment to 70 per cent up to 2010. Yet when machines are able to perform manual labor cheaper and more efficiently than humans, what jobs will remain? Former US Secretary of Labor, Robert Reich, assumes that manual labor will eventually be replaced completely by machines. Nevertheless he argues that there will still be a high demand for a human work force. These new workers will have to be highly educated and trained “symbolic analysts” – lawyers, doctors, journalists, consultants, and the like – which create value beyond mere manufacturing. However currently only a fraction of the labor force is capable of performing these jobs. Even though goverments have stated their intention to increase investment in education it is questionable whether this goal can be achieved for everyone. And even if it were possible, the advancement in information technology is not restricted to manual labor. Machines have augmented the physical performance of man to the point were he becomes superfluos. The same augmentation is also taking place with our cognitive abilities. The famous quote of the computer being a “bicycle for the mind” becomes evident when we consider the vast amount of data a single person can analyze with the help of a personal computer. Therefore, the observation that machines in the long run are not destroying jobs but creating new ones is merely that; an observation and not a law. There might well be a threshold of automation that changes the rules of the game entirely. If that should happen this would be one aspect in which we have to change our culture radically. In any case, how well we are prepared for these new machines will determin the social acceptance and ultimately the cost of the transition. Since development is still gradual, there will be several years left to create new practises. There is likely not a simple nor a single answer. The convergence of disciplines and the accelerating speed of technological progress will require a holistic approach and result in ad-hoc solutions. Fortunately, we can start learning about the problem and its solutions already today. After all, the future is already here, just not equally distributed."
William Brace, Anniina Huttunen, Vesa Kantola, Jakke Kulovesi, Lorenz Lechner, Kari Silvennoinen and Jukka Manner,
in Bit Bang, Rays to the Future, Yrjö Neuvo and Sami Ylönen, Helsinki: Helsinki University Print, 2009, p. 236-263
Wednesday 21 March 2012
The Pirate Bay Planning "Low Orbit Server Drones"
And I thought that there is no way to combine file sharing and robot studies...
During the academic year 2008-2009 I participated in the Bit Bang post-graduate course.
“Bit Bang – Rays to the Future is a post-graduate cross-disciplinary course on the broad long-term impacts of information and communications technologies on lifestyles, society and businesses. It includes 22 students selected from three units making the upcoming Aalto University: Helsinki University of Technology (TKK), Helsinki School of Economics (HSE) and University of Art and Design Helsinki (UIAH). Bit Bang is a part of the MIDE (Multidisciplinary Institute of Digitalisation andEnergy) research program, which the Helsinki University of Technology has started as part of its 100 years celebration of university level education and research. Professor Yrjö Neuvo, MIDE program leader, Nokia’s former Chief Technology Officer, is the force behind this course.”
We wrote a joint publication based on the fall and spring group works. The book was published in co-operation with Sitra, the Finnish Innovation Fund:
During the fall term my group wrote about the processor and memory a book chapter “The Digital Evolution – From Impossible to Spectacular” and in spring we were given topic intelligent machines. Finally, our book chapter was titled “Augmenting Man”. And that's the way it all started. (To be continued...)
Monday 19 March 2012
Currently, I participate in the Graduate School Law in a Changing World:
"LCW graduate school covers all fields of legal studies, from various branches of positive law to general jurisprudential studies. Each doctoral student will get acquainted with the europeanisation and the globalisation of law... LCW provides the doctoral students with a systematic 4-year research training programme."
So far, we have had great fun together!
In addition, I am a member in a research project titled "New technologies in the content production and usage" funded by the Helsingin Sanomat Foundation:
Our principal investigator LL.D., docent Taina Pihlajarinne just published a book about linking (A Permission to Link) . It is available only in Finnish: