The Ethical AI – How to use AI to check AI
Let’s start with the low-hanging fruit before getting into the business case. Machine learning and AI became necessary because of big data. Without them, most social media platforms would have collapsed under their own volume.
Consider the scale involved – Instagram sees roughly 95 million photos uploaded per day, Twitter around 500 million tweets, Snapchat some 3 billion snaps, Facebook 6 billion likes and comments, WhatsApp manages 65 billion messages. No legacy technology, and no amount of added compute, processes that at human speed.
Here’s the CxO-level version of the problem – the same forces that created the need for AI at that scale; that is, unstructured, multilingual, culturally-loaded content, also created the conditions for its misuse. Just as hackers grew alongside the networked world, deep-fakes, hate-mongers, trolls, and bad actors are growing alongside social media, and this cannot be solved by legislation, by breaking up platforms, or by throwing more human moderators at it.
We will need to use AI against AI.
Every major platform is either building its own ethical guardrails or claiming to. That fragmented approach won’t work. It produces a patchwork of narrow, platform-specific systems that look more like applied AI than the general-purpose ethical layer the industry actually needs. Building something durable requires the industry to convene as a consortium, not compete on safety claims individually.
My working hypothesis is that every computing unit, laptop, server, phone, has a profile tied to an organization, an individual, or both, much like we profile people using demographic data. Profile the computing unit by what it does, and an ethical AI layer can continuously learn, adapt, and counter malicious behavior on its own terms. Call this the Behavioral Profiling Framework for ethical AI governance. This isn’t a side project for the legal or trust-and-safety team to handle in Q3 and is much bigger than just one player to address. But the same concept can be applied to an enterprise. For a deeper discussion on building this into your own governance model, reach out. I’d welcome the conversation.