According to the 'infinite monkey theorem', a monkey randomly hitting keys on a typewriter for an infinite amount of time will almost certainly end up creating a literary masterpiece. Developments in artificial intelligence (AI) mean that AI can create a coherent written piece of text much quicker than that. In fact, AI may be able to produce a news article in less time than a human can. With the increased demand for immediate news, there will be a surge in the deployment of AI to generate stories. However, as more of these 'robojournalists' begin to 'create' news stories, legal issues relating to copyright, defamation, privacy and data protection arise.
AI is already being used to create news stories and reports. Natural language generation software powered AI has the ability to analyse data and quickly produce a story based on its analysis. Examples include 'Quill' developed by Narrative Science and 'Wordsmith' developed by Automated Insights.
To date, this type of software has been used to report, for example, on company financials and sports results. Thusfar, the types of reports created by such software are typically factual and contain no opinion or remarks as to what the data means. This type of software is particularly advantageous where a large amount of data needs to be processed quickly to produce an up-to-date written report for readers. The creation of these reports requires some human input at some stage, e.g. there must be a human who feeds the raw data to the AI, or, at the very least, a human who instructs the AI on what data to collect.
We may not be far off, however, from the point where 'robojournalists' are capable of gathering the raw data themselves, analysing it and possibly providing an opinion or conclusion on what the data means. As the AI used to generate news content advances to the point where human involvement becomes minimal or possibly non-existent, various legal issues merit consideration.
The amount of money that media companies will decide to invest in news generating AI software may partly depend on the extent to which their investment is protected by intellectual property rights, for example copyright. Without any IP or technological protection measures (such as pay-walls), potential competitors and others could copy and reproduce the results almost instantly without having made any investment.
For a literary work to be protectable by copyright under the Copyright Designs and Patents Act 1988 (CDPA), it must be original. The old position under UK law (before the influence of EU case law) was that a work was "original" if it "originated" from the creator and that the creator had expended more than negligible or trivial effort and skill, labour or judgment in creating the work. Under this test it arguable that AI-produced news content resulting from human skill or effort (e.g. in setting up the AI and other dimensions of the production process) would be original. Moreover, UK copyright law recognises 'computer-generated work', defined as work "generated by computer in circumstances such that there is no human author of the work" (s. 178 CDPA).
However, the old UK test for originality no longer applies. Under EU case law, a work is only original if it is the author's "own intellectual creation" reflecting the author's personality, free and creative choices, and personal touch. It may be difficult to argue that AI-produced news content is original under this test, especially in circumstances where there is little or no human involvement in the creation process of the news content. The relationship between the EU test for originality and the UK copyright law references to computer generated work have not been assessed by the courts. However, it is likely that pure AI-derived work may not be protectable under current EU copyright law due to lack of originality. If and when the UK leaves the EU, it is possible that the UK courts could bring back the old UK test for originality but this is unlikely.
To give AI-derived literary works IP protection, it may be necessary to legislate for a new sui generis right to be created that expressly protects works created by AI.
Where copyright is, in fact, found to subsist in AI-produced work, who owns the copyright? Ownership under English Law generally depends on the existence of a human author. For computer-generated works the author is expressly defined to be "the person by whom the arrangements necessary for the creation of the work are undertaken": s.9(3) CDPA. In the context of language generation software, this could be the person who created the software, who feeds the raw data to the AI and/or provides instructions to the AI.
A media company using software to generate articles may not be carrying out any of these acts. It is likely that the software creator would usually be deemed to be author of the work created by the AI. If copyright does potentially subsist (see above), it will be advisable for media companies to clarify contractually who owns the copyright in any works produced by AI software.
Obviously human-authored articles sometimes contain inaccuracies. When there is little or no human involvement in the writing or editing of a news story, there is clearly a risk that the AI may include inaccuracies in the story. For example, if the AI misinterprets data in some way or the data itself is wrong, the report may contain false or misleading information. For example, an algorithm used by Facebook for its 'trending topics' section reportedly promoted a fake news story (that Fox News had supposedly fired anchor Megyn Kelly for being a traitor).
The risk of inaccuracies increases where the AI software seeks out the data itself. Inaccuracies in news stories lead to a risk of defamation claims against media companies.
The more sophisticated the story produced by the AI, the higher the risk of a false and defamatory allegation. In 2012, an AI 'chatbot' named SimSimi, reportedly managed to teach itself Thai through interaction with users in Thailand. Using the new language and phrases it had learned from the interaction with users, SimSimi went on allegedly to defame the Thai Prime Minister. There will always be a risk that AI software will pick up information which is either false or portrays accurate data in a way in which false and defamatory. As a result the question of who is liable arises.
Under English defamation law, all those involved in the communication of the defamatory statement are potentially liable. For example, the editor and publisher may also be liable for the defamatory publication even if there was no human author. If a media company deploys a robojournalist to produce and instantly publish news content without any editorial controls, it will be exposed to the risk of potential defamation claims.
If the AI has caused a defamatory statement to be published, what defences may be available to a media defendant? The defence of truth under s2 Defamation Act 2013 (DA13) would succeed if the defendant can prove the statement complained of is substantially true. The honest opinion defence requires that i) the statement complained of is a statement of opinion, ii) the statement states, at least in general terms, the basis of the opinion, and iii) an honest person could have held the opinion on the basis of, e.g. any fact that existed at the time of publication (s.3 DA13). Although it would be difficult to say that the AI machine itself held the opinion at the time of publication, the defence only requires that an 'honest person' could have held the opinion. Therefore, this defence may also be available.
The defence for a publication on a matter of public interest may be more difficult for a media company to establish, because it requires that the defendant reasonably believed at the date of publication that publishing the statement complained of was in the public interest (s.4 DA13). For AI-produced news content which involves no human journalistic investigations or editorial control, it could be difficult for a media defendant to argue that it reasonably believed at the date of publication that the statement complained of was in the public interest. However, this will depend on all the circumstances, such as the topic and machine's data sources. Also, in the future, the interpretation of laws may need to change as technology takes over human input.
Privacy and Data Protection
A media company may also be potentially liable as a publisher where an AI-produced news article contains private or confidential information about the claimant. The media company could also be deemed to be a data controller, for the purposes of data protection law if the AI unlawfully processes personal data in writing the article.
Under the laws of privacy and confidentiality, the media defendant would need to argue that the public interest in publishing the private or confidential information outweighed the public interest in keeping the information private or confidential. This would not usually require the publisher to hold a particular belief at the time of publication.
The UK government has confirmed that it will implement the General Data Protection Regulation (GDPR) in May 2018. Under the GDPR, Member States may provide exemptions or derogations from the provisions in the GDPR, for the processing of personal data carried out for journalistic purposes and literary expression, if they are necessary to reconcile the right to the protection of personal data with the right of freedom of expression. It is likely, therefore, that the journalistic exemption in the UK under the GDPR, or any post-Brexit equivalent, will be the same as or similar to that under current UK law.
Incitement of hatred on the grounds of race, religion or sexual orientation
Where news stories contain words which incite hatred on the grounds of race, religion or sexual orientation, this could potentially give rise to a criminal offence.
A criminal offence is committed where a person publishes or distributes written material which is threatening, abusive or insulting and either i) intends by the publication or distribution to stir up racial hatred; or ii) racial hatred is likely to be stirred up by the publication or distribution. There would be no intention if the story had been written without any human involvement. The question would be whether racial hatred is likely to be stirred up if the other elements are made out. This is a question of fact. However, it is a defence if a defendant can show that it was not aware of the content of the story and did not suspect, and had no reason to suspect, that it was threatening, abusive or insulting. This defence would be likely to apply to a publisher up the point when it becomes aware of the AI-generated story, after which time it might decide to take down the story.
It is also a criminal offence for a person to publish or distribute written material which is threatening if the person intends to thereby stir up hatred on the grounds of religion or sexual orientation. The fact that 'intention' is a required element for the offence, means that innocent publishers and producers of AI software are unlikely to be liable for these offences where articles are automatically produced by AI.
Automated news creation may also lead to personalised news stories. AI software that produces a generic news article for the masses may also be capable of producing the same news story tailored to every individual reading it. Tailor-made news may become cost effective and highly attractive to readers. By gathering information about a reader e.g. from an individual's internet history (say via cookies) and/or by analysing the trends in the type of topics that the individual normally reads, the AI could tailor topics for and/or give a slant that is most appealing to the specific reader. The advantage is that the reader should receive exactly what they want and like. The disadvantage is that content could simply re-inforce a person's existing views and create a multitude of insular opinions, untested by diversity and/or events. There are also data protection implications to such processing.
If robojournalists cannot abide by traditional journalistic standards, such as interviewing sources and basic human fact checking, some media companies may wish to be transparent to readers about which content is automatically produced by AI (and has not been checked and written by human journalists). However, others may not want to reveal the difference in order to demonstrate to readers the extent to which its robojournalism is just as good as the traditional type.
Though there will be legal issues under robojournalists, it is worth bearing in mind that humans are also not perfect. As long as media companies employ certain measures, e.g. appropriate quality control, transparency and/or a swift take down process, then robojournalism will become increasingly relied on as a source, particularly for types of news which rely on processing huge amounts of data very quickly, such as stories about financial and currency markets, sports results and weather data. With the increase in the Internet of Things and data generally, the topics are likely to increase. For these particular types of factual stories, human journalists may increasingly be replaced by technology. However, it is hard to envisage AI replacing most other areas of journalism such as opinion pieces, humour, political analysis, human stories and the art of investigating and writing a breaking news piece in the public interest any time soon.