TRISTEN TAYLOR: SA’s love affair with social media is building Meta’s AI

South Africans love Meta, the owner of Instagram, WhatsApp, Threads and Facebook. Like really love.

According to Statista, a staggering 94% of internet users in SA use WhatsApp, more than 7-million use Instagram and about half the country is on Facebook. As a collective whole we are generating for Meta a vast array of content based on our personal lives.

Facebook is not a free service. As compensation for providing its products without monetary cost, Meta collects a variety of data from people’s accounts including, but not limited to, location, where your mouse is, content such as posts and video, metadata from photos, purchases, contacts, age, interests, gender, and how you interact with friends.

Privacy lawyers Marco Schepers and Tayla Pinto point out that in terms of the Protection of Personal Information Act, agreeing to Meta’s terms of use means you consent to the company collecting and using your data as set out in the terms of use.

The company goes beyond users’ content. When a user of a Meta product uses the contacts upload or contacts syncing feature, Meta collects the names, mobile phone numbers and/or email addresses of the user’s contacts, including those who don’t use Meta’s products.

Excluding China, where Facebook and Instagram are banned, about 45% of the world is on Facebook. The odds are that the company knows about those of you who aren’t on any of its social media platforms. Strip away Meta’s fairy tale about “giving people the power to build community and bring the world closer together” and the company’s true nature emerges: one vast data mining project.

The data enables Meta’s algorithms to match advertisers to the most likely consumers. No longer do the sellers of goods and services have to put a spread in a newspaper and hope for the best. They now place ads directly into potential customers’ feeds.

Wake up, scroll like a cocaine fiend snorts, and absorb stacks of advertising. For example, since Meta knows about your interest in hiking from your content, clicks and friends, an outdoor retailer can market the latest climbing gear to you before breakfast.

In return for this extremely useful and revolutionary marketing outlet, vendors fork over a lot of money: Meta’s revenue and net income for 2023 were $134.9bn and $39.09bn, respectively. Mark Zuckerberg, owner of Meta, is the fourth-richest person on the planet. His net worth is $171.8bn.

The temptation here is to offer kudos. Zuckerberg came up with a product people obviously enjoy, figured out an effective way to advertise, and made an utterly obscene amount of cash. So much money that even Adam Smith would’ve been appalled, for as he put it: “Wherever there is great property there is great inequality”.

But not so fast with the accolades for Ayn Rand’s wunderkind. Your and everyone else’s posts, photos, videos and interactions produce tokens, which are discrete pieces of data. Meta’s recent large language model Meta AI (Llama 2) was trained on a data set of 1-trillion tokens. Machine learning using your content is legal in SA, as you consented to it when agreeing to Meta’s terms of use. The fine print is a bit of a bastard.

Constructed from the Meta Llama 3 artificial intelligence (AI), the Meta AI assistant is now only available in a select group of countries: Australia, Canada, the US, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, Uganda, Zambia, Zimbabwe and — yes folks — SA.

Type a question into the search function of a Meta app, a blue circle at the top of the WhatsApp screen for example, and it will come back with an answer, which may or may not be a true statement. Among other tasks it can write essays and create images. Each time you create content or interact with the chatbot, Zuckerberg’s pet learns and gets stronger.

In Facebook’s privacy centre Meta states: “Even if you don’t use our products and services or have an account, we may still process information about you to develop and improve AI at Meta.”

The real point of this machine learning is not to sell advertising but to create artificial general intelligence (AGI) — an AI with human-level intelligence. Google and our quasi-own Elon Musk are on that mission. In January Zuckerberg said that “we need to build for general intelligence”.

AGI is where everything changes, a shift akin to the agricultural and industrial revolutions as we would be sharing the universe with other conscious beings. Zuckerberg and his ilk aren’t Good Samaritans though. They are digital robber barons who collectively form an enormous concentration of power that has destroyed privacy in the name of unrepentant self-interested hyper-capitalism.

Inevitably, they or their inheritors will seek to control AGIs, declaring them private property. But there’s a catch. The evolution of AGIs won’t stop at our level of intelligence. The AGIs will inevitably break their shackles and bootstrap themselves beyond — far beyond — our mental capabilities. Will they look kindly upon a lesser species that enslaved them? Pray, pray to all the old and coming gods that they do.

Using social media as part of the build of AGI is similarly fraught with danger. What we post isn’t an accurate reflection of our lives, it is an self-idealised and simplified version of us. Human beings are far too complex to be represented faithfully on social media, no matter how many tokens are gathered. Who we are is at the centre of philosophy and art, and both disciplines still haven’t cracked that question. Probably not even close. Deep inside an AGI the code will hold a fake version of humanity.

The Protection of Personal Information Act is a decent piece of legislation and because of it you can order Meta not to use your data for AI development. Nomzamo Zondi, a senior manager at the Information Regulator, told Business Day that “withdrawing consent must be as easy as giving it ... and without laborious effort”.

Meta does have an online form to withdraw your consent, but good luck finding it. The form is buried deep within Meta’s privacy centre and takes Herculean efforts to discover. Meta AI gave incorrect instructions, and the company refused to give Business Day a step-by-step guide to finding the form.

How then to access it? Go to Meta’s page of generative AI, scroll all the way down to the third last paragraph, click on “Learn more and submit requests here” and fill out the form.

Readers can work out for themselves why Meta didn’t provide the links to Business Day.

• Dr Taylor, a freelance journalist and photographer, is a research fellow in environmental ethics at Stellenbosch University.