TrademarkNow (and its research background)
1 like481 views
The document discusses legal AI and the use of fuzzy logic in trademark law, referencing the research and development of systems to determine likelihood of confusion in trademark cases. It outlines the evolution of cognitive processes in legal reasoning, emphasizing the interaction between intuitive and analytical methods. TrademarkNow, founded in 2012, utilizes insights from this work to develop AI-driven trademark search and monitoring tools.
![Inescapable open texture:
No boozing in a park (but “civilized”
drinking is okay)
Section 4
Intake of intoxicating substances
The intake of intoxicating substances is prohibited in public
places in built-up areas [...].
The provisions of paragraph 1 do not concern [...] the intake
of alcoholic beverages in a park or in a comparable public
place in a manner such that the intake or the presence
associated with it does not obstruct unreasonably encumber
other persons’ right to use the place for its intended
purpose.
(Finland: Public Order Act (612/2003))](https://image.slidesharecdn.com/codex2015-150604220045-lva1-app6892/85/TrademarkNow-and-its-research-background-14-320.jpg)
![Mosong: the domain
Article 8
Relative grounds for refusal
1. Upon opposition by the proprietor of an earlier trade mark, the
trade mark applied for shall not be registered:
(a) if it is identical with the earlier trade mark and the goods or
services for which registration is applied for are identical with the
goods or services for which the earlier trade mark is protected;
(b) if because of its identity with or similarity to the earlier trade
mark and the identity or similarity of the goods or services
covered by the trade marks there exists a likelihood of confusion
on the part of the public in the territory in which the earlier trade
mark is protected; the likelihood of confusion includes the
likelihood of association with the earlier trade mark.
[...]
(CTM Regulation (40/94/EC))](https://image.slidesharecdn.com/codex2015-150604220045-lva1-app6892/85/TrademarkNow-and-its-research-background-15-320.jpg)
TrademarkNow (and its research background)
- 1. TrademarkNow (and its research background) CodeX at Stanford University 2015-06-04 Anna Ronkainen @ronkaine Chief Scientist and Co-Founder, TrademarkNow [email protected]
- 2. The real innovator’s dilemma 1. do research 2. ... 3. profit!
- 3. ‘Preliminary try-outs of decision machines built according to various formal specifications can be made in relation to selected administrative or judicial tribunals. The Supreme Court might be chosen for the purpose.’ (Harold Lasswell 1955)
- 4. ‘Can we “feed” into the computer that the judge’s ulcer is getting worse, that he had fought earlier in the morning with his wife, that the coffee was cold, that the defence counsel is an apparent moron, that the temporarily assigned associate judge is unfamiliar with the law and besides smokes obnoxious cigars, that the tailor’s bill was outrageous etc. etc.?’ (Kaarle Makkonen 1968, translation ar)
- 5. ”As we know, there are known knowns. There are things we know we know. We also know there are known unknowns, that is to say, we know there are some things we do not know. But there are also unknown unknowns, the ones we don’t know we don’t know.” – Donald Rumsfeld (2002)
- 6. (Un)known (un)knowns known unknowns known knowns unknown unknowns ??
- 7. (Un)known (un)knowns known unknowns known knowns unknown unknowns unknown knowns
- 8. (Un)known (un)knowns conscious ignorance conscious knowledge unconscious ignorance unconscious knowledge
- 9. Dual-process cognition System 1 • evolutionarily old • unconscious, preconscious • shared with animals • implicit knowledge • automatic • fast • parallel • high capacity • intuitive • contextualized • pragmatic • associative • independent of general intelligence System 2 • evolutionarily recent • conscious • distinctively human • explicit knowledge • controlled • slow • sequential • low capacity • reflective • abstract • logical • rule-based • linked to general intelligence (Frankish & Evans 2009)
- 10. Systems 1 and 2 in legal reasoning: interaction System 1: making the decision System 2: validation and justification (Ronkainen 2011)
- 11. What’s that got to do with legal AI? - MOSONG, my 1st (and so far only) system prototype - built for studying the use of fuzzy logic in modelling various issues in legal theory - specifically, the use of Type-2 fuzzy logic for modelling vagueness and uncertainty - trademarks initially just a random example domain - but the knowledge acquired through this research also proved useful for TrademarkNow...
- 12. Open texture ‘Whichever device, precedent or legislation, is chosen for the communication of standards of behaviour, these, however smoothly they work over the great mass of ordinary cases, will, at some point where their application is in question, prove indeterminate; they will have what has been termed an open texture.’ - (Hart 1961)
- 13. Standard example of open texture : No vehicles in a park ‘When we are bold enough to frame some general rule of conduct (e.g. a rule that no vehicle may be taken into the park), the language used in this context fixes necessary conditions which anything must satisfy if it is to be within its scope, and certain clear examples of what is certainly within its scope may be present to our minds.’ (Hart 1961) ... but that’s a bad example because vehicles are already categorized in excruciating detail so being more precise costs nothing
- 14. Inescapable open texture: No boozing in a park (but “civilized” drinking is okay) Section 4 Intake of intoxicating substances The intake of intoxicating substances is prohibited in public places in built-up areas [...]. The provisions of paragraph 1 do not concern [...] the intake of alcoholic beverages in a park or in a comparable public place in a manner such that the intake or the presence associated with it does not obstruct unreasonably encumber other persons’ right to use the place for its intended purpose. (Finland: Public Order Act (612/2003))
- 15. Mosong: the domain Article 8 Relative grounds for refusal 1. Upon opposition by the proprietor of an earlier trade mark, the trade mark applied for shall not be registered: (a) if it is identical with the earlier trade mark and the goods or services for which registration is applied for are identical with the goods or services for which the earlier trade mark is protected; (b) if because of its identity with or similarity to the earlier trade mark and the identity or similarity of the goods or services covered by the trade marks there exists a likelihood of confusion on the part of the public in the territory in which the earlier trade mark is protected; the likelihood of confusion includes the likelihood of association with the earlier trade mark. [...] (CTM Regulation (40/94/EC))
- 16. Mosong: the domain Tentative rule Article 8 Relative grounds for refusal 1. Upon opposition by the proprietor of an earlier trade mark, the trade mark applied for shall not be registered: (a) if it is identical with the earlier trade mark and the goods or services for which registration is applied for are identical with the goods or services for which the earlier trade mark is protected; (b) if because of its identity with or similarity to the earlier trade mark and the identity or similarity of the goods or services covered by the trade marks there exists a likelihood of confusion on the part of the public in the territory in which the earlier trade mark is protected; the likelihood of confusion includes the likelihood of association with the earlier trade mark. REFUSAL = MARKS-SIMILAR and GOODS-SIMILAR
- 17. ‘Training’ set: 119 cases
- 18. “Training set” 119 cases from 1997–2000, of which 107 from the Opposition Division (1st instance) and 12 from the Boards of Appeal (2nd instance)
- 19. Results for the training set 0 0.2 0.4 0.6 0.8 1
- 20. Validation set 30 most recent (2002) relevant cases: 20 from the Opposition Division and 10 from the Boards of Appeal Result*: all cases predicted correctly * when coded into the system by a domain expert
- 21. Results for the validation set 0 0.2 0.4 0.6 0.8 1
- 22. Non-expert validation • done by non-law students taking a course on • intellectual property law (n=75) • original validation set in two parts (15+15 cases) • at the beginning and the end of the course • completed non-interactively through a web form • correct answer: 54.6±6.5% • incorrect answer: 25.9±7.5% • no answer: 19.5±5.2% (± = σ)
- 23. Non-expert validation % ±stderr before after total group 1 (n=15) 41.3±1.7 65.8±2.8 53.5±1.7 group 2 (n=12) 46.1±2.0 65.0±3.0 55.6±1.9 group 3 (n=48) 43.3±1.3 65.9±1.3 54.7±0.9 total (n=75) 43.4±1.0 65.8±1.1 54.6±0.8
- 24. Initial conclusions from this work - it (sort of) works; using fuzzy logic makes sense in this context - poses more questions than it answers... - ...and that’s how I ended up trying to reverse-engineer human lawyers rather than just trying to build systems based on existing legal theory literature
- 25. Implications for legal AI - using rule-based methods has its advantages - human-readable - comparatively quick to develop - modifiable (esp. relevant wrt legislative changes) - but they can’t do the work alone - can’t make sense about situations which they weren’t specifically built to handle - real-world complexity needs (sometimes) statistical/machine-learning approaches
- 26. So, about that “...” ...
- 27. About TrademarkNow - founded in 2012, based in Helsinki, NYC and Kilkenny, now ~30 employees - products based on an AI model of likelihood of confusion for trademarks, based on my own basic research in computational legal theory (since 2002) - NameCheck: intelligent TM search - NameWatch: intelligent TM watch
- 28. A month ago, this happened...
- 29. How trademark searching is conventionally done - wildcards! - Nice classification - trademark registries - lots of back-and-forth between a lawyer and a paralegal (typically taking 2–7 days altogether): - Lawyer: create search strategy - Paralegal: carry out search - L: evaluate results, request more info on most significant ones - P: produce more info (repeat as needed) - L: give final risk assessment
- 30. Our version: From the query DAGNIAUX, yogurts, EU