Over the next two years, Europe and the UK are preparing laws that will limit the annoying content that social media companies have allowed to go viral. There has been a lot of skepticism about their ability to look under the hood of companies like Facebook. Regulators, after all, lack the technical expertise, manpower, and salaries Big Tech boasts of. And there’s another technical snag: The artificial intelligence systems used by tech companies are notoriously difficult to crack.
But opponents should keep an open mind. New techniques are being developed which will facilitate the exploration of these systems. The so-called black box problem of AI is not as inscrutable as many think.
AI powers most of the actions we see on Facebook or YouTube and, in particular, the recommendation systems that line up the posts that go into your News Feed or the videos you should watch next, all to make you scroll. Millions of data are used to train AI software, allowing it to make predictions more or less similar to those of humans. The hardest part for engineers is understanding how AI makes a decision in the first place. Hence the concept of the black box.
Consider the following two images:
You can probably tell within milliseconds which animal is the fox and which is the dog. But can you explain how you know? Most people would struggle to articulate what it is about the nose, ears, or head shape that tells them which is which. But they know for sure which picture shows the fox.
A similar paradox affects machine learning models. It will often give the correct answer, but its designers often cannot explain how. That doesn’t make them completely impenetrable. A small but growing industry is emerging that monitors the operation of these systems. Their most popular task: improving the performance of an AI model. The companies that use them also want to ensure that their AI does not make biased decisions when, for example, they review applications or grant loans.
Here is an example of how one of these startups works. A financial firm recently used Israeli startup Aporia to test whether a campaign to attract students was working. Aporia, which employs both software and human auditors, found that the company’s AI system was actually making mistakes, granting loans to some young people it shouldn’t have, or withholding loans unnecessarily. loans to others. When Aporia took a closer look, she discovered why: students made up less than 1% of the data the company’s AI had been trained on.
In many ways, the AI black box’s reputation for inscrutability has been exaggerated, according to Aporia chief executive Liran Hosan. With the right technology, you can even – potentially – decipher the ultra-complicated language models that underpin social media companies, in part because in computing, even language can be represented by numerical code. Finding out how an algorithm might be spreading hate speech, or failing to deal with it, is certainly more difficult than spotting errors in digital data that represents loans, but it is possible. And European regulators will try.
According to a European Commission spokesperson, the upcoming Digital Services Act will require online platforms to undergo audits once a year to assess the “risk” of their algorithms to citizens. This can sometimes require companies to provide unprecedented access to information that many consider trade secrets: code, training data, and process logs. (The commission said its auditors would be bound by confidentiality rules.)
But suppose European watchdogs can’t delve into the code of Facebook or YouTube. Suppose they can’t probe the algorithms that decide which videos or posts to recommend. There would still be a lot to do.
Manoel Ribeiro, a Ph.D. student at the École Polytechnique Fédérale de Lausanne, Switzerland, published a study in 2019 in which he and his co-authors tracked how some YouTube visitors were being radicalized by social media content. ‘far right. He didn’t need to access YouTube’s code to do this. The researchers simply looked at comments on the site to see which channels users went to over time. It was like following digital footprints – painstaking work, but it ultimately revealed how a fraction of YouTube users were lured into white supremacist channels through influencers who acted like a gateway drug.
Ribeiro’s study is part of a larger body of research that has tracked the psychological side effects of Facebook or YouTube without needing to understand their algorithms. While offering relatively superficial insights into how social media platforms operate, they can nonetheless help regulators impose broader obligations on platforms. These can range from hiring compliance officers to make sure a company follows the rules, or providing precise, random samples to auditors about the types of content people are being pushed towards.
It’s a radically different perspective from the secrecy in which Big Tech has been able to operate so far. And that will involve both new technologies and new policies. For regulators, this could well be a winning combination.
More from this writer and others on Bloomberg Opinion:
Zuckerberg’s biggest bet may not pay off: Parmy Olson
China’s cyber-isolationism has serious security implications: Tara Lachapelle
• No, Musk is not responsible for Twitter’s slowdown: Martin Peers
This column does not necessarily reflect the opinion of the Editorial Board or of Bloomberg LP and its owners.
Parmy Olson is a Bloomberg Opinion columnist covering technology. A former journalist for the Wall Street Journal and Forbes, she is the author of “We Are Anonymous”.
More stories like this are available at bloomberg.com/opinion