Previously published in New Hampshire Bar Journal, Winter 2012
For many people, artificial intelligence ("AI") is HAL from 2001: A Space Odyssey, KIT from Knight Rider, Data from Star Trek, or Haley Joel Osment from (appropriately) A.I. Artificial Intelligence. In light of this image, AI is frequently relegated to the science fiction section of Netflix and forgotten.
But that type of AI only represents one kind – strong AI, meaning AI that matches or exceeds human intelligence and that can therefore solve any problem and interact in any social situation much like a person would. And that type of AI is pure fiction at this point. However, weak AI is another matter altogether. Weak AI only recreates elements of human intelligence in a computer. We interact with weak AI all the time – Google, Global Positioning System ("GPS")1, video games, etc. Any machine or software that is able to follow simple rules to replicate human intelligence qualifies.
Recently, we have seen rapid advances in weak AI. Famous examples of this development are Deep Blue, the chess master machine, and Watson, the Jeopardy master machine. But, in a sense, those examples are just as fantastical as HAL and KIT. Hardly anyone interacts with machines like that.
The first major mass market product to change that is Siri, the "intelligent personal assistant that helps you get things done just by asking."2 Siri is the first commercially available, advanced weak AI. By "advanced weak AI," I mean weak AI that can recreate elements of human intelligence through human-like interaction.3 This distinguishes it from other popular examples of weak AI like Google – in their natural setting, humans don't type questions and requests to each other.4
This article looks at a few of the legal issues related to Siri, focusing on intellectual property and liability issues. What happens when Siri creates something new that has commercial value? What happens when reliance on a Siri search results in property or bodily damage? With the pending developments in weak AI, which I'll address in this article as well, the questions introduced by Siri will only grow in prominence and importance. But first, it is useful to have a better idea of what Siri is and does.
A BRIEF HISTORY OF SIRI
Like the internet and GPS, Siri is the product of DARPA, the Defense Advanced Research Projects Agency, a Defense Department agency that developments new technologies. In 2003, DARPA entered into an agreement with SRI International, a research group in Menlo Park California, to research and develop a "cognitive assistant that learns and organizes."5
In 2007, SRI incorporated a separate entity, Siri, Inc., to develop a commercial application for the AI developed in the DARPA project.6 Siri, Inc. received venture capital backing and eventually released a virtual personal assistant application for the iPhone in February 2010.7 In April 2010, Apple, Inc. bought Siri, Inc.8 When Apple released the iPhone 4S in October 2011, Siri was pre-installed as an all-purpose virtual assistant, not merely as an app.
Siri is not just a virtual personal assistant. It has both speech input and speech output. Users can speak to it and receive spoken responses. The extent of its interaction with the iPhone 4S and the Internet is fairly extensive. Ask Siri about the weather, and Siri will give you a short summary of the weather forecast where you're located. Ask Siri to tell your husband you are running late, and Siri will send a text message. Ask Siri to schedule lunch with your mom next Friday, and Siri will update your calendar and give you verbal confirmation. Siri does not process speech input solely on your phone. Rather, the software send commands through a remote server, so it is necessary for users to be connected to Wi-Fi or carrier service (Verizon, AT&T, etc.).9
The actual process Siri uses to translate your words into action – "breath to bytes" – represents an impressive achievement of internet technology. Smart Planet described how Siri responds to a user request to send a text to "Erica:"
The sounds of your speech were immediately encoded into a compact digital form that preserves its information.
The signal from your connected phone was relayed wirelessly through a nearby cell tower and through a series of land lines back to your Internet Service Provider where it then communicated with a server in the cloud, loaded with a series of models honed to comprehend language.
Simultaneously, your speech was evaluated locally, on your device. A recognizer installed on your phone communicates with that server in the cloud to gauge whether the command can be best handled locally — such as if you had asked it to play a song on your phone — or if it must connect to the network for further assistance. (If the local recognizer deems its model sufficient to process your speech, it tells the server in the cloud that it is no longer needed: "Thanks very much, we're OK here.")
The server compares your speech against a statistical model to estimate, based on the sounds you spoke and the order in which you spoke them, what letters might constitute it. (At the same time, the local recognizer compares your speech to an abridged version of that statistical model.) For both, the highest-probability estimates get the go-ahead.
Based on these opinions, your speech — now understood as a series of vowels and consonants — is then run through a language model, which estimates the words that your speech is comprised of. Given a sufficient level of confidence, the computer then creates a candidate list of interpretations for what the sequence of words in your speech might mean.
If there is enough confidence in this result, and there is — the computer determines that your intent is to send an SMS, Erica Olssen is your addressee (and therefore her contact information should be pulled from your phone's contact list) and the rest is your actual note to her — your text message magically appears on screen, no hands necessary. If your speech is too ambiguous at any point during the process, the computers will defer to you, the user: did you mean Erica Olssen, or Erica Schmidt?10
In short, Siri depends on sophisticated weak AI in the phone software interfacing with sophisticated weak AI in the Internet cloud to create a user-friendly advanced weak AI program. By relying on central processors over the Internet, Siri is able to improve the more people use it. Siri collects data from users and analyzes that data to improve its services.11 That represents a lot of automation interacting with real-world people, which can result in legal issues we haven't seen before.
Siri represents a nascent challenge to a number of long-held legal models. I say nascent because you have to follow Siri's functions to a logical extreme in order to get to the challenge in fields like intellectual property and liability. However, there are new technologies already in development that will see commercial sale in the not too distant future that will put these challenges front and center.
The Sounds of Siri – Who Owns the Copyright?
The words, sentences, and ideas expressed by Siri have little value as recorded media – for now. But Siri poses the potential to contribute to successful and profitable media, and once there's real money involved, the question of who owns the copyright to a Siri response, and potential royalties related to its commercial use, becomes more important.
Sampling in various forms of popular music – most notably hip hop and pop – has been a hotly contested issue in copyright law. As the Sixth Circuit noted, "Advances in technology coupled with the advent of the popularity of hip hop or rap music have made instances of digital sampling extremely common and have spawned a plethora of copyright disputes and litigation."12 Copyright disputes involving sampling have resulted in litigation concerning artists as diverse as Roy Orbison,13 the Beastie Boys,14 and George Clinton, Jr.15 George Clinton's music, in fact, is responsible for one of the largest allegations of copyright infringement, as entities holding copyrights to his music filed 500 counts against approximately 800 defendants for using portions of his recordings and music without authorization in 2001.16 In these cases, the key dispute was between the owner of a musical composition or sound recording and another person who sampled that music or sound to create otherwise new media.17
With Siri, a user could experiment with Siri's speech output in order to create an amusing or interesting sound byte, record that sound, and alter it using Auto-Tune into commercially successful media (e.g., song, commercial, etc.). Under that scenario, it appears that this creator would own the copyright to the new media. However, the law is not established on this point.
B. Ownership of Intellectual Property from AI
Section 201(a) of the Copyright Act states that "Copyright in a work protected under this title vests initially in the author or authors of the work." The Supreme Court has stated that, "As a general rule, the author is the party who actually creates the work, that is, the person who translates an idea into a fixed, tangible expression entitled to copyright protection."18 But this idea is in flux as an increasing number of weak AI computer programs like Siri are able to produce original music, text, and sound.19 You can find books of computer-generated poetry,20 novels written by a hacked-Macintosh that draws inspiration from Jacqueline Susann,21 and CDs of music composed by an autonomous program.22 When Siri says something new, who is the "person" who owns that copyright?
Ralph D. Clifford23 attempted to answer this question, albeit while examining intellectual property created through human-computer partnerships rather than intellectual property created by Siri specifically.24 Clifford examines the potential for computers and machines to develop copyrights, deciding that there is a spectrum of human-machine interaction. On one end of the spectrum, there is intellectual property developed by machines that are programmed and guided by humans; copyrights from those pairings are properly owned by the programmers.25 On the other end of the spectrum are machines that are capable through their programming of creating new media with little to no further interaction with people; copyrights from those machines enter into public domain, under Clifford's analysis.26 He relies on the Supreme Court's reasoning from Feist Publications, Inc. v. Rural Telephone Service, Co.: "The sine qua non of copyright is originality. To qualify for copyright protection, a work must be original to the author."27
Clifford extended this reasoning, explaining that a machine that is programmed to create creativity on its own does not warrant copyright protection because "no one derives rules for the computer to control its creativity; rather, using its learning algorithm and based on the training examples it is given, it develops rules on its own. This learning is done independently of its user."28 And in fact, this sounds strikingly similar to Siri's ability to learn to operate better the more people use it. As Clifford asks, "who then can claim a copyright in the expressive works" created solely by a computer?29 Clifford answers his own question: "The claim of the user of the machine seems highly dubious. The user was not the originator of these expressions as no specific creative effort was exerted by the user."30
I suspect that Clifford is mistaken regarding the intellectual property developed by programs like Siri. Nature and the law abhor a vacuum, which is effectively what the public domain is for intellectual property. One of the reasons why Siri is so interesting is that it is the first commercially available sophisticated weak AI. Although Siri is likely on the end of Clifford's spectrum where intellectual property is owned by the human interacting with the AI, that spectrum breaks down when the AI is available to a mass market. Clifford assumes that the programmer (the person who creates the machine), the user (the person who uses the machine to create output), and the artist (the person who makes new media with the machine's output) are all the same person. Beginning with Siri, that is not necessarily the case.
Rather, if there is an owner to the sounds Siri creates, it is likely either one of or a combination of the following parties:
- The artist who creates the new media using the sample of Siri's speech output;
- The user who experimented with Siri in order to create the speech output (this could be the same person as the artist); or
- Apple, which owns the Siri source code.
Thus far, there has been little if any difference between these three parties when other forms of weak AI have created new copyrights. When Scott French hacked his Macintosh to create prose reminiscent of Valley of the Dolls, he programmed the computer to ask him questions that would permit it to draft appropriate language and story. When David Cope entered into an agreement with Centaur Records to release a CD of music composed by the program he named "Emily Howell," he had written the Emily Howell program and provided it with musical inputs to analyze before it wrote new music.
Cope is an apt example because his Emily Howell program is the direction that weak AI is going. Emily Howell is a program that can analyze music and feedback and create its own style.31 Cope says "I've taught the program what my musical tastes are, but it's not music in the style of any of the styles – it's Emily's own style."32 With regard to other human composers, Cope says that they "are looking at a competitor – a virtual composer competing in the same arena with 'her' own style and music that is really excellent. It seems to me that these composers should feel a little less smug and more defensive about their position."33 Although he has made no public comments regarding the copyright of Emily's first CD, published by Centaur Records in 2010, it can be safely assumed that Cope kept any royalties as Emily's programmer.
C. Who Should Own Siri-Created Media and Similar Copyrights
Cope and Emily demonstrate why someone has to own the copyright associated with media produced by computers, machines, or programs like Siri. When there is money involved, the public domain is not a realistic option, as someone (likely many people) will claim ownership and do so with compelling arguments. Similarly, without the potential for return on an investment, programmers and investors will not seek to develop technologies that develop creativity.
Although no law speaks directly to this issue now, as weak AI is more commercially available, Congress or a court will need to address it in the near future. From a public policy standpoint, it would be better for Congress to address it, so there is unanimity throughout the country, without divisions in circuits. Assuming that there will be more mass-marketed sophisticated weak AI like Siri, granting ownership of media created with mass-market AI to the users and artists, rather than the programmers, will stimulate greater creative development of media using that AI. Additionally, the users and artists are closer to the "author" of such media as currently contemplated under US copyright law.
But Clifford's concern is a legitimate one. Should traditional copyright and patent protection be granted to the writer of code that produces creative products largely independent of human direction? Emily Howell already produces classical music; there is no reason to believe a program that produces commercially successful pop music will not be developed sometime in the near future. Similarly, there are AI programs that write full articles regarding sporting events;34 there is no reason to believe a program that produces commercially successful novels will not be developed in the not-so-distant future. Perhaps there should be a line drawn between private AI like Emily Howell and marketed AI like Siri, with programmers receiving different copyright protection for the intellectual property created by private AI. In recognition of the concerns cited by Clifford for properly fostering and rewarding human creativity, the copyrights owned by the programmers of such programs could receive a limited copyright – maybe 10 years – before it enters into the public domain.35 Although this issue seems remote now, Siri is only the first product in an increasingly complex line of weak AI products, which will make issues of intellectual property ownership more important.
Siri Made Me Do It – Liability for Siri's Mistakes
There are already instances in which accident victims have blamed weak AI – most notably GPS – for the accident. In West Yorkshire, England, a driver followed the directions provided by his GPS until he was trapped on a narrow cliffside path and the police had to tow him back to the main road.36 Despite that driver's attempts to blame the GPS system, the British court found him guilty of careless driving.
Although there is limited case law addressing GPS culpability in accidents, it is not far- fetched to suggest that a court could attribute partial or total liability to a GPS system in some instances. Some American attorneys expect that development sometime in the near future, as more drivers depend on satellite directions, particularly in areas where they do not know the geography.37 That dependence will increase as advanced weak AI in GPS improves to the level that Siri represents. As drivers are better able to talk directly to their GPS – "There's a traffic jam up ahead – is there another way around the next few intersections?" – they will rely on those devices more and more. By mimicking actual human interaction, GPS devices increase the likelihood that a court will attribute partial or total liability to a GPS system because it becomes more reasonable for drivers to rely on them.
Siri invites a similar reliance – it is programmed to accumulate user data through the Internet cloud where it operates, learning more about users as they ask questions and make requests. By design, Siri tries to make itself more indispensable with each user interaction.
So consider this scenario: A user is visiting an unfamiliar city. He has an appointment, he is lost, and he asks Siri the fastest route to the site of his meeting. Siri provides him with directions, but does not mention some of the potential dangers associated with that route. If the route takes the user through an area with a high crime rate, and he is assaulted, mugged, etc., what liability does Siri (and by extension, Apple) have for those damages?
This is not an easy question to answer, and any decisions would depend on a court's analysis of the user's actions. Did the user behave reasonably? Should the user have identified the dangerous situation? Did the user have sufficient information to deviate from Siri's directions? Apple – and any other manufacturer of weak AI – could argue that the user was contributorily negligent in causing an accident if he failed to act reasonably or exposed himself to unreasonable risk of harm.38
Any court considering the question of reasonableness in the scenario described above will have to consider Siri's design. It is programmed to become more useful. With that goal in mind, isn't it reasonable for users to increase their reliance on Siri as they use it more? This is particularly true where users are asking questions concerning topics of which they know little – directions in a new geographic location, recipes for cooking, lesson in carpentry, etc. The interface matters in this analysis, as well. It is more reasonable to rely on a device that responds to conversational questions than on a device that requires a cumbersome keyboard.
As should be clear by now, there is no easy answer to this, but Siri is only the first of many new technological developments that will require a re-examination of how we determine and assign liability. For example, numerous American companies – including General Motors and Google – have spent considerable time and resources developing an autonomous car.39 Current prototypes are already traveling – under supervision – along highways at 70 miles per hour and through well-populated downtown areas in selected states; Alan Taub of General Motors predicts that autonomous cars will be commercially available by 2020.40 While torts law has assumed the validity of driver negligence since the invention of the automobile, that concept could soon be outdated, like elevator operator negligence.
Alternatively, jurisdictions could decide that removing all culpability from drivers is a bad idea. If that decision is made, by a court or legislature, there will have to be a reflection of that in the law. One option is to try to retain the reasonable person standard when looking at potential driver negligence in car accidents involving autonomous cars. But a reasonable person, when driven by an autonomous car, is likely paying little attention to the road. As autonomous cars become more popular, that attitude will only increase. Another option is for towns or states to designate areas where drivers will be held to traditional negligent driver standards. In response, drivers could choose to assume control of the car from the automatic driver or take their chances with the autonomy engaged.
And there is also the argument that autonomous cars – as well as other forms of autonomous programs and machines, like Siri's calendar function – are in fact more reliable than human operators. There are no statistics for this hypothesis, and there won't be until autonomous cars are available. But it should surprise no one that a machine capable of focusing on the task of driving is safer than a human being texting, talking on the phone, or trying to find the last French fry in the McDonald's bag while behind the wheel. With this possibility in mind, we might want to use Siri as a prompt to examine the aspects of our lives that could be improved by autonomous machines and begin developing legal frameworks to encourage weak AI in those areas.
Siri is a marvelous piece of technology. Most consumers have never had access to sophisticated weak AI like it before. As impressive as it is, it represents only the beginning of the next wave of technology. Devices will connect to the Internet, learn your preferences, and perform tasks for you with more utility than Siri offers now. They will be capable of producing new sounds, music, and texts with more creativity and originality than Siri offers now. It is important to keep that in mind when we use Siri. The artificial intelligence it relies on will appear in other programs and machines, forcing us to change many of our legal models. The two discussed here, copyrights and liability, are just the beginning. Even seemingly unrelated areas of law, like land use, will be affected – Towns that rely on limiting the number of employees at a property to control development will have to revise their zoning ordinances in order to address the growing number of businesses that hire few employees. Autonomous assistants, like Siri but bigger and smarter, will do much of the work.
Many thanks to Jen Finch for her tireless research and assistance.
1. The term "GPS" will be used throughout this article to refer to any satellite navigation system.
2. Apple Siri FAQ. http://www.apple.com/iphone/features/siri-faq.html. Retrieved 1-31-12.
3. Although there are GPS devices that exhibit advanced weak AI, reviews are mixed at best. By all measures, Siri's voice-interface seems superior to that of any GPS device. On top of that, Siri's potential for ubiquity distinguishes it from GPS, which has a more limited function.
4. For the purposes of this article, texting, instant messaging, emailing, etc. are not considered part of man's "natural setting."
5. John Markoff. "A Software Secretary That Takes Charge," New York Times. December 13, 2008. http://www.nytimes.com/2008/12/14/business/14stream.html?_r=1. Retrieve 1-31-12.
6. Timothy Hay. "Apple Moves Deeper Into Voice-Activated Search With Siri Buy." Wall Street Journal Blog. April 28, 2010. http://blogs.wsj.com/venturecapital/2010/04/28/apple-moves-deeper-into-voice-activated-search-with-siri-buy/ Retrieved 1-31-12.
7. "Siri Launches Virtual Personal Assistant for iPhone 3GS." Press Release, SRI International. February 8, 2010. http://www.sri.com/news/releases/020510.html. Retrieved on 1-31-12.
8. Hay, "Apple Moves Deeper Into Voice-Activated Search With Siri Buy," supra.
9. Jill Duffy. "What is Siri?" PC Magazine. October 17, 2011. http://www.pcmag.com/article2/0,2817,2394787,00.asp. Retrieved on 2-1-12.
10. Andrew Nusca. "Say command: How speech recognition will change the world." Smart Planet. http://www.smartplanet.com/blog/smart-takes/say-command-how-speech-recognition-will-change-the-world/19895?tag=content;siu-container. Retrieved 2-7-12.
11. Duffy. "What is Siri?", supra.
12. Bridgeport Music, Inc. v. Dimension Films, 410 F.3d 792, 798-99 (6th Cir. 2005).
13. Campbell v. Acuff-Rose Music, 510 U.S. 569 (1994) (ruling that 2 Live Crew's use of Orbison's "Pretty Woman" constituted fair use).
14. Newton v. Diamond, 349 F.3d 591 (9th Cir. 2003) (ruling that the Beastie Boys were not liable for sampling James Newton's "Choir" in their track "Pass the Mic").
15. Bridgeport Music, supra (ruling that NWA's sampling of a guitar chord from Clinton's "Get Off You Ass and Jam" violated the copyright on that song).
16. Id., at 795.
17. Although 17 USC § 106 distinguishes between the rights held by the owner of a musical composition copyright and those held by the owner of a sound recording copyright, that distinction is not relevant to this article.
18. Community for Creative Non-Violence v. Reid, 490 U.S. 730, 737 (1989).
19. See Melville B. Nimmer & David Nimmer, Nimmer on Copyright, § 5.01[A] (2011).
20. Racter, The Policeman's Beard is Half Constructed (1984).
21. S. French & HAL, Just This Once (1993). The Baltimore Sun article, "Hal is back, and writing best-sellers," published on July 8, 1993 explores this book further.
22. The program referred to is "Emily Howell," written by a UC Santa Cruz music professor named David Cope. Cope and Emily are discussed in greater detail later in this section.
23. Clifford is a professor at the University of Massachusetts School of Law. Nimmer, supra, relies on him as an expert in the evolving discussion regarding intellectual property produced by AI.
24. Ralph D. Clifford, "Intellectual Property in the Era of the Creative Computer Program: Will the True Creator Please Stand Up?" 71 Tul. L. Rev.1675 (1995).
25. Id., at 1686-1694.
26. Id., at 1694-95.
27. 499 US 340, 345 (1991).
28. Clifford, "Intellectual Property in the Era of the Creative Computer Program: Will the True Creator Please Stand Up?" at 1694, omitting citations.
29. Id., at 1695. Clifford dismisses the idea that the AI itself could own the copyright, at least under the current law. Noting that the "author" of a work owns the copyright under the federal Copyright Act, he reviews its use there and in other portions of the U.S. Code. He concludes that "the use of the term 'author' in the Copyright Act implies Congress meant a human author... the general use of the term 'author' in the U.S. Code reinforces the conclusion that Congress intended the term to mean humans." Id., at 1682, 1684.
31. Jacqui Cheng, "Virtual composer makes beautiful music – and stirs controversy." ars technical. September 29, 2009. http://arstechnica.com/science/news/2009/09/virtual-composer-makes-beautiful-musicand-stirs-controversy.ars. Retrieved 2-2-2012.
34. Farhad Manjoo, "Will Robots Steal Your Job?" Slate. September 27, 2011. http://www.slate.com/articles/technology/robot_invasion/2011/09/will_robots_steal_your_job_4.single.html. Retrieved 2-10-12.
35. If that seems too short a period of time, consider this: Forbes magazine estimated JK Rowlings' net worth to be approximately $1 billion 7 years after Harry Potter first appeared on bookshelves.
36. Chris Brooke. "'I was only following satnav orders' is no defence: Driver who ended up teetering on cliff edge convicted of careless driving." Daily Mail. September 16, 2009. http://www.dailymail.co.uk/news/article-1213891/Driver-ended-teetering-cliff-edge-guilty-blindly-following-sat-nav-directions.html. Retrieved 2-2-12.
37. Eric Sinrod. "What's Next, GPS Liability?" FindlLaw. http://articles.technology.findlaw. com/2008/Jan/15/11079.html. January 15, 2008. Retrieved 2-3-2012.
38. Martin J. Saulen. "'The Machine Knows!': What Legal Implications Arise for GPS Device Manufacturers When Drivers Following Their GPS Device Instructions Cause An Accident?" 44 N. Eng. L. Rev. 159, 189 (2010).
39. Tom Vanderbilt. "Let the Robot Drive." Wired. http://www.wired.com/magazine/2012/01/ff_autonomouscars/all/1. January 12, 2012. Retrieved 1-13-12.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.