Human rights in the age of

Publishing the code and the data allows third-party experts to identify potential issues. 
Opponents of transparency in AI systems question the utility of publishing the code and training data for such 
complex systems that may rely on millions of data points and have models that change over time. 
Although auditing AI systems may present a new set of technical challenges, it does not render transparency 
meaningless. AI developers vet training data and AI outputs to identify sources of bias and test for equitable 
outcomes. Transparency would ensure access to necessary training data and AI outputs for independent 
experts to identify sources of bias and test for equitable outcomes, including identifying any problems 
developers missed or masked. This process necessarily increases accountability and fosters user trust.
127
Full transparency is vital in high-risk cases and need not damage companies. 
Opponents of transparency also argue that transparency requirements reduce incentives to invest in new AI 
systems because it would allow replication.
While history has shown that open source projects are often not only successful but facilitate innovation, 
there may be an option in limited circumstances where private-sector actors determine that route is 
untenable. In such cases, private-sector actors could facilitate access to relevant code by trusted, identified 
third parties for audits and testing. Recently, Facebook has given select researchers access to data to study 
election interference on the platform.
128
Given increased public scrutiny of the role of algorithms in our lives, 
it is conceivable that other companies will follow suit. However, information on data sets used and outputs 
should still be published, as well as any other information that could facilitate understanding and measure bias.
However, states should always be required to provide full transparency for government use of AI systems. 
This is particularly important in areas such as in law enforcement and the justice system. Although 
companies supplying the software may have reasons for not publishing the code and the training data in 
these cases, fundamental human rights cannot be sacrificed for the sake of corporate interests.