A recent paper from researchers at the University of Cambridge demonstrates that all is not lost for ethical hackers, with the authors arguing that they can help to restore some of the faith in artificial intelligence that has been lost during the so-called “tech-lash.”
The authors suggest that the AI industry would benefit greatly from allowing a community of ethical hackers to stress-test new AI-based products for potential harm. The researchers found that while a growing number of organizations have created ethics policies pertaining to their use of AI, there is often a distinct gap between these policies identifying the need to do something and the organizations’ tangible tactics for change.
One of the recommendations made by the authors is for companies to set up “red team” hacking processes to allow ethical hackers to test systems to try and find flaws, with rewards paid out for any that are found. This would allow companies to better prove the integrity of their AI applications to the public and other stakeholders before releasing it.
They believe that if such measures are not deployed, then the already low levels of trust the public has in the tech industry, and in AI specifically, will fall even further. This is hugely important because AI-based technologies play an increasingly important role in our society, whether driving social media, powering autonomous transportation, or providing the brains behind virtual assistants.
Opening up the black box
Many AI systems today are essentially black boxes, with their inner workings often shrouded in mystery. What’s more, the intense nature of competition in the space has prompted many tech firms to bypass the kind of auditing or third-party analysis that can help them to identify any biases or other flaws in how the systems operate.
As such, the researchers believe that companies would be well served by offering up financial incentives to help raise the trust levels in their products and that such an approach could help them bypass the need for stricter regulation. That such an approach could be driven from within the industry is also a crucial step in showing the wider world that it is taking this problem seriously and not merely paying lip service to it.
“There are critical gaps in the processes required to create AI that has earned public trust. Some of these gaps have enabled questionable behavior that is now tarnishing the entire field,” the researchers say. “We are starting to see a public backlash against technology. This ‘tech-lash’ can be all-encompassing: either all AI is good or all AI is bad.”
Separating good from the bad
They argue that for the technology to be truly trusted, it will be vital for society to be able to distinguish between reliable implementations and wholly unreliable implementations. This then creates an incentive to be trustworthy because it’s possible to tell good from bad and, therefore, less pressure to cut corners.
While most developers have no real desire to create untrustworthy or flawed technology, it has not always been clear what concrete steps they can take to do that. The concept of “red-teaming” has been commonly deployed in the cybersecurity community for years to help identify weaknesses and vulnerabilities in digital systems so that those holes can be plugged. It’s an approach that the researchers believe could be just as valuable in identifying flaws in AI systems.
Suffice to say, few but the largest of organizations have the resources to develop red teams internally, and indeed doing so would raise its own ethical concerns. As such, the researchers suggest that a better approach would be for a third-party community could evolve to interrogate AI systems and report any findings they make to the developers. Such global resources would enable a much wider range of organizations to tap into red teaming capabilities.
The initiative would carry financial rewards for any ethical hacker or researcher that discovers flaws in the AI, especially if it has the potential to damage public trust or safety. It’s an approach that has already been tried by Twitter when they offered financial rewards for anyone who could identify biases in their AI-based image-cropping algorithm.
“Lives and livelihoods are ever more reliant on AI that is closed to scrutiny, and that is a recipe for a crisis of trust,” the researchers conclude. “It’s time for the industry to move beyond well-meaning, ethical principles and implement real-world mechanisms to address this.”
