Google is one of the most popular and innovative companies in the world. But the question remains. How does Google work, and how does its algorithm function? Well, that is a secret. Specifically, a trade secret, protected under Intellectual Property Law. Therefore, ambitious competitors willing to learn how Google functions, and what the secret algorithm that makes this search engine the best in the world is, might have to go to great lengths to find out.
One could argue that the following dilemmas could arise:
how can a company effectively protect itself from having trade secrets exposed?
should algorithms be protected as trade secrets in the first place?
‘Trade secrets are intellectual property (IP) rights on confidential information that may be sold or licensed.’ [1] As a result, such information is necessary to be ‘commercially valuable because it is secret, be known only to a limited group of persons, and be subject to reasonable steps taken by the rightful holder of the information to keep it secret, including the use of confidentiality agreements for business partners and employees’. [2]
If one applies the above definition to Google’s algorithm, it translates to the following:
the algorithm to be considered and protected as a trade secret must have a commercial value, and
it must be known only to a few people, where those people have taken measures to keep it a secret, for example, by signing confidentiality agreements.
As mentioned above, one of the most efficient ways to protect one’s company from trade secret leaks by insiders (e.g., either current or former employees) is the formation of strong confidentiality agreements. But sometimes, employees break their agreements or decide to take risks as they end up sharing trade secrets with competitors in order to profit. This is evident from a recent case, where an ex-Google technologist took trade secrets from the company when he left, only to use them in his new position.
The story goes as follows. When Mr Levandowski decided to leave Google, he approached various talented employees of the company to found Otto, ‘a self-driving truck start-up’.[3] In 2016, Uber acquired Otto for $600 million. [4] In 2017, a lawsuit was filed against Uber for trade secret infringement. Namely, Mr Levandowski was accused of unlawful exploitation of trade secrets regarding Google’s autonomous vehicle research. He later used that information ‘to bolster Uber’s self-driving program.’[5] Reportedly, Mr Levandowski was accused of ‘33 counts of theft and attempted theft of trade secrets from Google.’[6] According to the Justice Department, ‘before resigning from his job at the search giant, Mr Levandowski had downloaded thousands of files related to the company’s development of self-driving cars.’[7]
The outcome of the story? Mr Levandowski was sentenced to prison for 18 months. Most people would probably agree that Mr Levandowski’s actions were wrong and that he was rightfully punished. No one should share unlawfully trade secrets and not face any consequences. Trade secrets are immensely valuable and should be properly protected. Adequate steps should be taken at all times by companies and inventors to prevent these situations from occurring.
What one ought to further examine, though, is whether algorithms, in particular, should be protected as trade secrets. Time and time again, algorithms have been accused of leading to biases, discrimination and inaccuracy. It has been argued that their regulation is of paramount importance. Regulating them could prevent negative outcomes such as discrimination from occurring or at least mitigating them to some extent. But how can algorithms be regulated, in the first place, since they are protected as trade secrets? As regulators do not know exactly how they operate and function, it seems that regulating them can be a very difficult, maybe a close to impossible endeavour. Some have even argued that ‘the IP sector should accept that the current use of trade secrecy law to protect companies’ AI work is a ‘bad’ solution because it creates a dangerous lack of transparency and accountability.’[8]
Others, in favour of trade secrets and their usefulness, state that in our ‘increasingly complex, highly competitive, hyperconnected world, some things that might ordinarily be protected by traditional IPRs [...] are best-kept secret (Nirwani)’.[9] This is to be expected since, for example, due to the ‘algorithms’ general anonymity and commercial value, they are eligible for trade secret protection.’[10] The issue lies on the fact that trade secret laws can block ‘access to the algorithm and to explanations behind automated decisions, dimming the transparency of the entire system.’[11] Additionally, trade secrets are described as ‘unlimited in nature lasting for as long as the information is secret’; this equates to the fact that ‘the opacity of algorithms might continue to exist in perpetuity ’.[12] Thus, the lack of transparency hinders the possibilities of achieving the proper regulation of algorithms and their outputs.
One can see that algorithms are especially useful to businesses and society at large, so the balance that needs to be achieved in protecting algorithms as trade secrets and preventing their negative outcomes from arising is a delicate one. This issue needs to be addressed in depth by regulators and lawmakers, and new solutions must be found so innovation can be fostered, while the public is protected from any potential harm caused by the unregulated use of algorithms within society. The potential negative outcomes that may surface from the unregulated use of algorithms due to their opacity should not be undermined.
Endnotes
[1] WIPO, ‘Trade Secrets’ .
[2] ibid.
[3] Kate Conger, ‘Star Technologist Who Crossed Google Sentenced to 18 Months in Prison’ (New York Times, 4 August 2020).
[4] ibid.
[5] ibid.
[6] ibid.
[7] ibid.
[8] Katarina Foss Solbrekk,‘Three routes to protecting AI systems and
their algorithms under IP law: The good, the bad and the ugly' (2001) 16 No. 3 Journal of Intellectual Property Law & Practice 258.
[9] ibid.
[10] ibid.
[11] ibid.
[12] ibid.