Is AI Selfish? | by Greg Satell

I read the second edition of Richard Dawkins' "The Selfish Gene" in the late 1980s, and since then have regretted not studying biology. Here Greg Satell starts by explaining it: "genes act as if they are selfish, working to replicate themselves in the most efficient way, regardless of what that entails for the organism that carries them" before leading onto memes (a concept also invented by Dawkins): "elemental bits of culture that compete to be replicated in the marketplace of ideas."

But I was unfamiliar with "Susan Blackmore's ...temes, elemental bits of technology, like lines of code... competing to replicate", which is the necessary stepping stone to applying Dawkins' original insight to AI, which "must be selfish as well, competing to get itself replicated through us... [so] What is the context we are creating for this competition and how will the rules affect our own fate?"

Key elements of the environment AIs will evolve in

Some key things to consider:

• "inclusive fitness, the ability of a gene to get replicated regardless of the effect on the body that contains it"
• how social media optimises for enragement: "ideas that spread aren’t necessarily the most useful or intelligent, but often the ones that invoke the most outrage, that gets our brains producing dopamine. Evolution has conditioned our bodies to recognize dopamine as a reward, so we keep going back for ideas that produce it."

This is therefore "the learning environment for our algorithms. The ones that are able to trigger dopamine rushes may be best adapted". Ouch!

Other aspects to the social/psychological environment which AIs will evolve in:

• as a species we put enormous resources into systems which promote collective action, like religion and sports: "the clapping, singing and chanting [which happen in both] are as important as the prayers, rules and myths"
• within any such ceremony, "there is almost always a very clear hierarchy of roles". According to Will Storr in The Status Game, "we act out our roles in search of status, which we pursue by playing three “games,” that of prestige, dominance and virtue.... signaling to others what we desire our roles to be so they know how they can best collaborate with us... [hence] it’s so important for people to signal identity" - cue the domination of virtue signalling in social media.
• Back to dopamine: "When people ... recognize the status we crave, our brains release dopamine". An easy way of achieving this "is conflict, which forces us to pick a team and express our identity"

Shaping that environment

The good news: we can shape the environment AIs will evolve in.

The bad news: In "Data, Strategy, Culture & Power, data expert Nicole Radziwill introduces “Radziwill’s Law,” which states:

Data cannot be decoupled from power. Organizations create and use data, analytics, and AI in ways that embed and reflect the power structures and power differentials between the people that develop and use them."

Moreover, unlike other challenges where humanity has managed to cooperate to build and protect the common good, "AI is different... constantly not only learning from our behavior [but] also generating cultural content that helps to shape our identities and how we pursue status. We are at once players and referees; teachers and learners; influencers and influenced... we are an algorithm’s way of creating more algorithms."