Remember when Mark Zuckerberg, back in 2020, put on his best tie and said there should be more regulation of harmful online content… but added (with a calculated blankness) that it wasn’t Facebook’s job to decide what this content should be.

The world’s largest publisher didn’t want to carry the responsibility of what hosting content meant.

To be fair, an attempt to regulate AI has been considerably swifter than how we handled the social media platforms. We let them loot us for years.

In that spirit, after a scant three years, the EU’s charge against the machines (in the form of the AI Act) is complete. It was voted in by the European Parliament last week.

The law will come to life in May and people living in the EU will start seeing changes by the end of the year.

Regulators will need to enforce the law properly and companies will have up to three years to get themselves sorted.

So, what is going to change?

Created with ChatGPT

Some AI stuff is going to get banned

If an AI product puts a person’s health, education or security at “high risk”, then it is going to have to go. They will be outlawed by the end of the year.

This is fair enough, but there are also some grey areas: the Act will ban AI systems that deal with “subliminal, manipulative or deceptive techniques to distort behaviour and impair informed decision-making,” or exploit vulnerable people. That is a large part of the internet at this point.

I’m not for these types of products, but it feels like it is wide open to interpretation.

It will ban real-time facial recognition software in public places. And creating a database of faces by simply scraping the web will be outlawed.

This is great news, but it comes with a huge asterisk: the police will still be able to use biometric data and facial recognition software in public places as long as they can claim that the data is being used to fight serious crimes like terrorism or kidnappings.

Read more in Daily Maverick: AI shaping up to become the greatest geopolitical weapon in history

Digital rights organisation Access Now (for whom we made a podcast years ago) called the new law a “failure for human rights” because it did not ban facial recognition across the board.

And while companies and schools are not allowed to use software that predicts what a person is feeling, they can use it if they can claim they are doing it for the health or safety of another person.

AI is going to be more obvious

Red Dwarf (the UK sitcom set on a spaceship that I watched repetitively as a kid) had the astute idea that if humans create robots, then they will aim for them to be identical to humans, but this will be too disturbing for consumers and so the next series, though more advanced, will look less sophisticated and more robotic.

The AI Act will force tech companies to label deepfakes and tell people when they are chatting to a bot. Also, if you are generating AI media, you will have to create it in a detectable way. This is good news for those fighting disinformation.

However, it is still difficult to detect if content is AI-generated, no matter what law you put in place.

EU citizens will be able to complain (if they want to)

An office is going to be set up and the folks there will receive and process various grievances. And they are even accepting CVs.

The paperwork is going to be a nightmare

Even if what you are doing doesn’t fit into this “high risk” category and you are creating a general AI language model, you will still need to create and keep all the technical docs.

These will have to show how you built the model and how you are keeping to copyright law, and you must illustrate (in a public document) what training data went into the model.

This is pretty huge and beneficial.

And if you are OpenAI or Google, you will need to go through a whole host of checks to make sure your system is secure (or face big fines and potentially be banned from the EU completely).

The answer is to be free

If you keep your model open-source and share every detail about how the model was built, you will not be harassed and you’ll mostly be exempt from the punitive side of the AI Act.

And what about the regulation (and opportunities) of AI in Africa?

The African Union — made up of 55 member nations — is trying to figure out how to implement its own regulation. However, the projected benefit of AI adoption on Africa’s economy is intense.

Nigeria, Ghana, Kenya and South Africa alone could see up to $136-billion injected into their economies by 2030, thanks to AI. But this is only if businesses in those countries begin using more AI tools.

So, Africa needs to regulate away the harm, but make sure that they harness the benefits so their economies don’t get left behind.

In the news

Apple is reportedly in talks with Google to insert its Gemini AI model into the iPhone. If everyone plays nice, it would give AI capabilities to billions of users when the new iPhone operating system launches later this year.

Apple also reportedly met with OpenAI about using its models, but Sam & Co are now in talks with Samsung. DM

Develop Al is an innovative company that reports on AI, builds AI-focused projects and provides training on how to use AI responsibly.

Subscribe to Develop AI’s newsletter here.