The Critical Importance of Decentralized, Explainable AI for Better Social Networks

The scary collective-psychology implications of current centralized social networks — and the promise of decentralized alternatives.

Ben Goertzel
13 min readJul 28, 2020


Every day one sees more and more people waking up to the stupidity and malevolence of current online social networks. One even sees increasing talk about the need for better online social networks — ones that foster positive human experience and development, reasonable thinking and imaginative creation.

What I haven’t yet seen, however, is much real understanding of why our current online social networks are the way they are, and what might be done to fix them.

What we need to create a better social network, in my view, is relatively straightforward to state:

1. Code should be open source.

2. Ownership and control should be decentralized… so there is no company and no government tasked and burdened with calling all the shots. Embedding blockchain technologies deep in the technical design of the network is the right way to achieve this.

3. Major decisions should be made democratically. If some group systematically doesn’t like the outcomes votes are getting, they can fork the code and make a new version, or copy the code and create a new network using the same software.

4. Recommendations of connections and content should be made using AI that has the ability to explain the reasons behind its judgments — and these explanations need to be shown routinely to users in a way they can understand.

Due to SingularityNET and a variety of other recent developments, we already have the basic technologies needed to support this sort of approach — as far as tech is concerned, we could have decentralized, democratic, transparent and ethical alternatives to Facebook Twitter, LinkedIn, TikTok and so forth right now. We even have some existing websites — such as — embodying significant parts of the above.

What we still don’t have — yet! — is the alignment of resources to get alternatives fulfilling all four of the above points fully built-out … and then pushed toward wide adoption.

In this essay, I’m going to dig a bit into the collective-psychology implications of current centralized social networks and future potential decentralized alternatives. Given the key role that social networks currently play in channelling, guiding and crystallizing collective human thought and emotion, these are not just issues regarding the infrastructure of a particular technology sector — they may actually be critical to the future path of humanity.

Decentralized Networks are Fundamentally More Appropriate than Governments or Corporations for Modulating Information Flow

The current mode of organizing social networks presents companies like Facebook with seemingly unsolvable problems. Should Facebook take down racist posts? What about posts with indirect racist implications? Posts explicitly aimed at affecting political elections using false information? What about questionable information? What makes information “too questionable”?

In relatively free regions like e.g. the US or Western Europe, government’s role in modulating public speech is restricted to extreme cases of directly dangerous speech (directly advocating violence, explicit forms of “hate speech” etc.). I believe this is for the best, as the alternative of giving government stronger control over speech has glaringly obvious failure modes — i.e. it can be used by fascist-minded government leaders to squelch the public dialogue needed to organize opposition to their rule.

However, less extreme, less directly dangerous instances of public speaking can also have a dramatic and systematically destructive impact on individual minds and the public mind — and in the current situation in the “free world”, these are being modulated by corporations in a very ad hoc way. And these corporations, run by digital marketing and tech geek types, are wholly unsuited to the subtle problem of modulating online information, knowledge and relationship flow for humanity. These organizations are optimized for delivering software and delivering ads and making money — not for guiding the public mind toward the broader good.

On the positive side, there are all sorts of brilliant ideas and wonderful creations being put out there on the Internet every minute of every day — but most of these have a hard time finding the audience they deserve because most of the currently widely operational information dissemination mechanisms are oriented toward corporate profit maximization rather than general human (or transhuman!) benefit maximization.

So given that it’s best to keep governments out of the modulation of non-extreme, not-directly-dangerous public speech — and given also that corporations running social media networks are not well suited for the job — then what can we do?

What we can do, of course, is roll out information and relationship coordination systems that allow the community of individuals engaged in social networks to carry out intelligent information and relationship management themselves — via their explicit judgments and decisions, and their implicit actions.

SingularityNET’s weighted liquid rank reputation system is one tool designed to help with this — it provides a framework for leveraging data regarding user behaviours to estimate the reputation or value that a given user assigned to another person or an information source or piece of information. Using tools like this, each person guides what information they see, based on AI algorithms calibrated for their own benefit rather than the benefit of corporations seeking to profit from them.

You can learn more this here weighted liquid rank reputation systems for marketplaces, also on the following Discourse forums.

But how can people know these AI algorithms are acting for their own benefit? Transparent and explainable AI is another key ingredient here. SingularityNET reputation system can leverage either opaque or transparent AI for making predictions of how much a given user will appreciate a given other person, source or datum. Embedding neural-symbolic AI into the reputation system is one strategy that can enable a reasonable degree of explainability, as symbolic patterns can generally be simplified and rearranged into explainable form via automated transformations.

A Failure of Global Resource Allocation

Creating analogues of modern social networks architected according to the above principles would not be incredibly hard given available technologies. There would be a lot of engineering required, but mankind is now capable of doing a lot of engineering.

The essential reason fair and democratic and non-stupid large-scale social networks built along these lines do not exist, is that nobody with adequate financial resources has wanted to put in the money to build one.

One may ask why such a thing hasn’t just been created by volunteer OSS developers or entrepreneurial devs. One core issue is that such a social network framework would only be valuable once it had a lot of users. Because of this, the reality or perception is that getting such a project off the ground at this stage — with huge entrenched centralized competitors — would require a massive marketing budget.

I would love to see SingularityNET and OpenCog used as foundational tools for the implementation and operation of this sort of social network, and would welcome partnerships with other groups interested to collaborate on this sort of work.

But what I want to explore a little more here in this article are some of the subtler reasons why transparency, explainability and decentralization are so important in social network construction.

I’m going to get a little theoretical here — but before long I’ll get back to the practical.

Addiction as a Form of Hebbian Learning

How and why did our social networks become this way?

Of course, the profit motive underlying the companies that build and operate these social networks — Facebook, Twitter and the rest — is part of the problem. The goal of the social network becomes maximum addictiveness because this is the most straightforward way to maximize profit in this particular business niche … and human psychology being what it is, the way to achieve maximum addictiveness is to appeal to a mix of emotions in a rather perverse way.

Much as junk food perverts our taste and smell system to create unhealthy addictive food experiences of a sort that didn’t exist in the Paleolithic environment we evolved for, similarly, current social networks pervert our social, cognitive and emotional systems to create unhealthy addictive informational and relational experiences of a sort that didn’t exist before Internet tech came along.

But the drive to profit from maximum addictiveness is only one of the more proximal aspects of the problem. The destructive perversity we see in modern social networks is a self-organizing phenomenon, which is fed by the drive on the part of social network designers to create addictive experiences but has additional special properties.

The nature of addiction is that: Not only is the addictive thing enjoyable, but its absence is painful … and the amount needed to get enjoyment keeps going up. So if you’re addicted to something, the more you use it, the more urgently you need to use it more… etc. in a feedback loop.

One may rephrase this as: The more an addictive thing is done, the less psychological energy it takes to do it again … and the more psychological energy it takes to not do it again.

In a way addiction is a lot like long-term potentiation (LTP), a basic principle of learning in the brain — the more a synaptic connection between two neurons is used, the easier it gets for the charge to spread along that connection. I.e., the more that charge flows along with a connection, the less energy has to be spent to get a given amount of charge to the other end of that connection.

LTP in the brain — also known as “Hebbian learning” after 1940s neural learning theorist Donald Hebb — leads to the formation of complex cell assemblies representing knowledge, ideas, behaviours and so forth. Similarly, addictive dynamics in social networks — addiction to messaging in particular groups, or to particular people, in particular ways — leads to the formation of complex social and informational assemblies, which then take on lives of their own.

The designers of modern social networks almost surely did not intend to foster assembly formation via Hebbian-type learning powered by addictive reactions to particular social interactions. But they lucked into something quite interesting.

The assemblies formed in social networks are groups of people who reinforce each others’ ideas and reject others’ ideas, and sets of ideas that reinforce each other and repel opposing ideas. So-called “filter bubbles” are one aspect of this — but filtering is only part of the story. Assemblies of inter-reinforcing ideas are also cauldrons of creativity in which new ideas are formed — and assemblies of inter-reinforcing people are similar creative cauldrons for the creation of new selves and identities for participants.

Nudging Formation of Better Human and Conceptual Assemblies via Promoting Mindfulness and Self-Awareness

The core property of a better social network is that it better fosters the formation of better assemblies of people and ideas.

So what is a better assembly? One simplistic but useful way of looking at it is: A better assembly of people or ideas is one that provides the people involved with a higher degree of satisfaction over time.

But how is it possible to get people to choose assemblies that will provide them with more satisfaction, collectively, over time — , instead of those that will provide maximal immediate gratification?

I believe the key to this is self-awareness — mindfulness, to use a currently popular phrase.

And this is where the lack of transparency in modern social networks becomes incredibly acute.

One problem with the lack of transparency in the operation of modern social networks is that it covers up the way the companies owning the networks are exploiting the data obtained from you, the user, to sell things to you and your friends.

But another, less often discussed problem is: Opaque systems do not provide people with the explicit visibility into their own perceptions, actions and judgments that they need to have — if they want to overcome their more egocentric and shortsighted aspects and place more weight on their longer-term benefit and on broader benefit to the world.

Transparent systems help people who are questing to understand themselves, what they like and dislike, what they are reacting to, what they are attracted to versus repelled from. They are valuable tools for people to use to improve their consciousness and experience and become wiser and happier and more useful to the world.

Opaque systems provide much less aid to self-understanding and self-awareness and push the user to instead behave more mechanistically, following their inclinations without much reflection.

Right now we have social networks that hide from people the logic via which they make recommendations and guide behaviours — and also help people hide from themselves, the psychological reasons for their own behaviours and judgments. Because the networks are predicated on people falling prey to addictive behaviours, which is going to be less likely if people are open with themselves about what they’re doing on the networks and why.

What Kinds of Transparency Do We Need?

What we need is, first of all, easy to visualize and explore and edit ways for a social network user to see why certain recommendations (of connections or content) are being made to them.

We also need tools that a user can deploy to understand the judgments they are making themselves. What kind of people am I choosing to follow on Twitter? What kinds of images, when shown to me, am I most likely to look at longest? Which other people tend to read the same documents as I do regarding medicine? Or regarding physics? Or regarding Chinese politics?

There is incredible positive potential to be realized via applying AI and information visualization to allow people to study their own behaviours online. This could be a powerful tool for helping people to achieve greater self-awareness and move toward more positive consciousness — as manifested in their online interactions and more generally. However, this is clearly not how the social network tech industry is oriented.

We also need the AI that makes recommendations to people, to better understand the states of consciousness of the humans it’s involved with, so that it can learn patterns regarding what recommendations have positive consciousness impact. This is perfectly achievable using modern AI tools — it is very feasible to build classifiers that identify user consciousness states based on online activities.

However, if all this advanced analytics is to be deployed to study the behaviours and consciousness states of every social network user — this only highlights the extreme importance of data sovereignty and privacy and strong encryption. There is tremendous dystopic potential in centralized AI-powered social networks that more fully understand user behaviour and user consciousness. In some ways, the current social networks’ focus on advertising may have been beneficial for humanity, in that it has caused the AI analytics to focus on recognizing patterns useful for selling people stuff, instead of deeper and even more nefarious forms of AI analysis.

Explainable AI can be an incredibly powerful tool for either good or bad in social networks — and if these networks are implemented in a decentralized way, the possibility of a robustly good outcome seems much higher. A centralized implementation can be great with the right dictator at the centre — but relying on a great dictator is not sustainable … and also not as effective as one might hope in the presence of powerful nonlinear-dynamical self-organizing phenomena. For instance, the CEO of Twitter may well be genuinely good-hearted and social-benefit-minded, yet the perverse dynamics of the Twitter network remains — because a social network takes on a life of its own, modulated to a degree but far from controlled in every detail by its creators or managers.

Decentralization is not a panacea either, but it opens the door to more robustly beneficial ways to do things — and SingularityNET’s reputation system, deployed in conjunction with explainable AI, is an example of how decentralization can be used to enable robust democratic governance in a diverse decentralized network.

What Happens Next?

There are fairly clear technology paths to creating social networks that do everything today’s corporate centralized social networks do — and more — and that is controlled in a democratic and participatory and privacy-respecting way. The broad utilization of such networks would have an extremely positive impact on our society — indeed one can argue that society’s collective decision-making, political and otherwise, is likely to be profoundly sub-optimal and defective until decentralized and democratic social-network systems become the default.

The creation of these alternative social networks is going to take significant resources, and the same establishments and institutions that benefit from the current order of things, are unsurprisingly not chomping at the bit to allocate resources to the creation of more broadly beneficial alternatives. However, with rising public awareness of the unsatisfactory nature of the current state of things, the tide may be changing. The more advanced that supporting tools from the decentralized-AI ecosystem (of which SingularityNET platform is a leading example) become, the less onerous the task of disrupting and dethroning the corporate social network overlords becomes.

Join Us

We thank each and everyone one of our community members for their continuous support. SingularityNET plans to reinforce and expand its collaborations to shape the coming AI Singularity into a positive one, for all. To read more about our other partners, click here.

SingularityNET has a passionate and talented community which you can connect with by visiting our Community Forum. For any additional information, please refer to our roadmaps and subscribe to our newsletter to stay informed about all of our developments.