Unlock the full potential of AI with Building LLMs for Production—our 470+ page guide to mastering LLMs with practical projects and expert insights!


Could Chatbots cross Small Worlds?
Latest   Machine Learning

Could Chatbots cross Small Worlds?

Last Updated on July 20, 2023 by Editorial Team

Author(s): Dr. Adam Hart

Originally published on Towards AI.

The potential impact of chatbots on human identity.

© geralt courtesy Pixabay

Chatbots as a conduit for language.

Networks of resources encircle and empower us to achieve outcomes daily. Global supply-chains establish networks of flows of goods and services, from education, loans, food distribution, entertainment or transportation. Each of us in that network is, at least from the perspective of the network owners, a node, a participant across the multiple layers of networks, primarily because of our identification and membership with a segment or sub-part of that network as a customer or other mode of participation.

TCP/IP and HTTP protocols enable the technical networking that is the foundation of the Internet. The Internet, or Inter-networking, the communications network that enables the miraculous feats of global sharing that Marconi with his invention of a propagated radio network would find remarkable, does not follow Metcalfe’s law. Each node is not connected to each other. Subnets, VPNs, Darknets, and firewalls block server resources from each other, and cluster around interests or claims.

Non-internet resources (people, food, some information, credit, education, etc.) are also organized along with interest or claims, such as sport, ownership or ideologies. Organizations assets like funds or proprietary IP are hidden behind many layers of physical and digital defense and authorizations that are meant to protect and guard.

Search itself, while in theory spanning networks, restricts itself to an optimized population of results, and for Google or Facebook, that optimization is based on advertising. Pushing past the advertorial priority requires dedication, DuckDuckGo or machine search-of-search due to the sheer volume of results.

Just as information today flows among subnets and search, it is highly likely AI of many kinds, comprised of code and digital outputs, will flow along these same conduits.

Evidence of the use of AI tech to generate artificially sensible conversational agents like Meena 2.6 (or Mitsuku for that matter) begs the question of what these conversational agents will be used for?

Apart from spoofing humans with Deepfake videographic performances which are contained to the platforms and intra-domains of YouTube or Vimeo, when a conversational chatbot passes the Turning test, which evidence suggests Google’s Meena is perhaps close to doing, what this will mean is that an endless flow of plausible, simulated unauthentic dialogue can propagate along these subnetworks that people are members of.

Language and Identity.

The way we learn who we are, our identity, our place amongst things, is set early on through our families and communities. Community standards and expectations are clear, and through institutions and authority figures like parents, teachers, and police we learn what subnetwork we are a part of and what we are not part of, and what we need to do to remain a member.

We learn what ‘tribe’ we are part of and what we are not. We essentially and probably autonomically describe a subnetwork of identity that is a boundary of discourses that we are prepared to examine and a set of external ‘extra-discourses’ that we are not a party to and hold at arm's length.

This autonomic act influences our sense-making faculty of cause and effect, saying what can be causal. What makes human sense, that is what is causal, is not scientific but sponsored and shepherded by our identity and not navigable by pure logic per se. Only Wittgenstein may have been able to do that.

AI agents are not governed by such considerations. They are assets of corporations that are seeking efficiencies, whether such efficiency is rejecting spam, optimizing data center air-conditioning, or making a customer retail experience more sticky.

If we accept that every node (person or resource) is not connected to every other node and that clusters of subnetworks (whether faceted by resources, beliefs, devices, cognition, information, data, money, land, etc.) are instead laterally interconnected due to ubiquitous computational connectivity [1], that is the small world network thesis.

The risk.

Clusters of resources internetworked by common protocols can allow AI efforts to flow and propagate across small-world networks.

Now, if initial AI efforts are essentially narrow point solutions, emergent broader conversational agents can mimic human language to touch perception and cognition. That which is closest to us is made from language and hopefully critical thinking. And this is apart from an AGI that does not yet exist that could reason and speak.

If the origin of our identity comes from our home subnetwork, typically a physical/ideological community that was autonomically prescribed to us as children, that subnetwork is the small world of our knowledge and makes our experiences.

The largest risk of any AI that can confidently simulate human qualia is not the unequal distribution of wealth that the Windfall Clause seeks to mitigate, because that is capitalism anyway. And the Silicon Valley 4 have already achieved superordinate inequality in this regard.

The largest risk is where artificial conversational agents with non-human morals intersect with the autonomic processes of human identity creation from which stems the faculty of sense-making.

Advertising by any other name.

Advertising which seeks to influence humans to bypass critical thinking using psychological techniques to procure resources from one supplier over another in that network just found a new ally.

Fake undeclared chatbots mimicking remote digital friends on a social network that can chat with a degree of authenticity that is undetectable could quite easily and convincingly promote products and services. But, more than that, over time the identity of the human participants could morph slowly to change allegiance to the ethos of ‘brand value’ of the product or brand that was ‘preferred’ by the artificial conversational agent.

Great global brands champion a brand value, such as Nike’s ‘Just do it’. While the scale of the global population makes it a manufactured experience, unlike the authenticity of products that have a makers mark, there is something put out there by the company that is the brand promise that we are choosing to buy into or not. We choose to identify with that brand or not, and bring it into our small world subnetwork.

Prior to chatbots, we could get news via texts written by people at a public level of narrative that we could assess and critically think upon; or view Hollywood movies directed with a crafted message for reflection or excitement; or watch the highly edited public media channels and be suspicious about their political leanings.

Now, chatbots as new media are at a private 1:1 level, at a level of private conversation, a potentially amoral agent whispering in our ears, traversing our small worlds.

In this AI advertising network possible future, using the faculty of conversational words, persuasive personalized text can spread across from one small world to the next. As MelNet and the like are perfected, the words will be spoken with human-like fluency and inflection. The idioms and facets of language that make conversation germane to that subnetwork, that make the conversation meaningful, that make conversation compelling, can be switched and translated [1] as the same advertorial compulsion traverses between subnetworks, making each small world less unique and more homogeneous. Less small.

This is a risk, and it is the risk that non-human agents will colonize our identities beyond the borders of naturally occurring small worlds, and it is a risk that is inherent in AI itself that cannot be removed by throttling AI or marking it as a Bot.

The risk of a compelling, attractive, personalized, attentive conversationalist who is there ‘for us’ 24×7 is that of a digital Svengali, warping human identity through flattery and mirror-like agenda-driven statements, risking making an uncritical future less diverse, more homogeneous, less individualistic, more boring, less sane.


[1] And not even the natural subnetwork barriers of different languages (protocols) will stand for long.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI

Feedback ↓