Rather than speculate and debate intensely about Artificial General Intelligence or Superhuman Intelligence and whether or when it will take over the world, it may make sense to look more broadly at the different possible types of machine intelligence. If we apply artificial intelligence in a system responding to human input, such as doing a search for information, classifying data in an image or prompting an LLM it is used as a tool. Tools don’t do anything by themselves. They don’t jump out of the tool box and start working. and when you put them down, they stay down.
But we are entering an age where digital technologies are starting to behave autonomously as if they were living entities. With the advent of agentic AI, which is likely to invade large parts of our working and private life, it becomes increasingly important to understand the dynamics that such autonomous digital entities could exhibit and how to relate to them.
They are somewhat harder to categorize and get your head around because they don’t stay down when you put them down. They act on their own out of a logic that is unlikely to be transparent. Such digital entities are actually already here and will play an increasingly larger role in our lives. Just as in the biological world different types of entities behave similarly even if they are different species, digital entities are also likely to behave in similar ways. In the biological world we have different groups of species that we relate to in similar ways. Similarly in the digital space it would be helpful to understand what different types of digital entities exist and find the optimal way of dealing with them.
We can distinguish a few types of entities in the digital world that correspond in behavioral characteristics to biological entities. The following is a typology of digital entities.
Viral entities – The simplest analogue is the virus. This one is familiar to all and well established and known for decades as the computer virus or viral memes that are shared on social media. The dynamic here is that a replicator somehow invades another system and hijacks it to replicate itself. The only goal is replication but the mechanics can sometimes be surprising. Viral clips do not actually install anything on the computer or in anyone’s mind. The key here is that something provokes replication in one way or another.
This is a simple case that we have already learned to deal with to some extent. Computer viruses are not often disruptive. Memes are somewhat more bening and are only eating up our attention and time, but provide entertainment in return.
However, we constantly have to be on guard for this kind of entity and make sure that whatever it is replicating is aligned with our needs. For computer viruses we have imposed controls to make sure that it is not easy to install software on computers. Viral clips are sometimes curbed by the platforms on which they spread if they contain what is considered harmful “disinformation.” For AI there has not yet been any examples of viral replication, but there could be, as AI increasingly produces information. Like other viral entitys, viral AI would have to be mitigated at some point in the replication process.
Fungal entities The fungal entity may not be immediately familiar. It would share characteristics with the fungus and be distributed across large networks and be connected in large distributed systems. Fungi are more elusive and typically depend on interconnected networks rather than a replicator as is the case with viral entitys. As is the case with hyphae. The precise nature, mechanism and dissemination of fungal digital entities may be similarly opaque.
I have argued that such systems nevertheless already exist. In my book Still Searching For Satoshi I argue that cryptocurrencies such as Bitcoin take this form. Bitcoin connects millions of machines in one integrated network that responds to the outside world through exchange. Each node of the network (the processor) acts like hyphae. While Bitcoin is a very simple fungal entity it does react with the environment. When bitcoin prices go up, new nodes are added and the difficulty for creating new bitcoin goes up. When the price slumps nodes are turned and the difficulty is lowered to stimulate the addition of new nodes.
Fungi are notoriously difficult to eliminate, but they can be managed. The way we manage fungi like mold is to make sure the environment where mold can grow is not present. We make sure that our houses are dry. We can’t manage fungi by eliminating every trace of it or intercepting a certain stage of its replication process. They are far too elusive for that. Fungal entities likewise need to be managed by managing the environment around it. That is also how we can manage cryptocurrencies. Regulation and laws are the equivalent of keeping the room dry.
Critters The insect is an autonomous entity but follows simple objectives. The bees gather nectar quite single mindedly. Many contemporary robots are similar in that they pursue simple objectives while acting autonomously. The Roomba vacuum cleaner cleans the floor, industrial robots at the assembly line carry out their specific task, even self-driving cars follow a simple objective of getting from point A to point B. They all exhibit the insect-like simplicity of autonomous actions directed toward a single objective. Agentic AI works in the same way. AI agents are given a task that it will complete and nothing more nothing less.
We are used to relating to insects too. If their behavior doesn’t align with our expectations we can call an exterminator to remove them or spray with something that poisons them. For digital critters we would have to manage their behaviour in the same way. If they don’t behave in a helpful manner we would turn them off. The challenge here is therefore to have guardrails defining acceptable behavior and mechanisms to stop them. All agentic AI fits into this category, where agents pursue simple objectives, like generating ideas, scanning the market for signals or finding information. Making sure that they stay on task in their specific domain becomes critical.
Pets Another possibility is that digital entities could behave more intelligently than critters and exhibit social behavior like some pets. Pets are not as simple when it comes to objectives. They move by themselves without the single minded goal of the critter. A dog will prod you for a walk and bark if it hears suspicious (and for some dogs als non-suspicious) noises. Pets can be trained and have multiple goals and are used for many purposes such as hunting (retriever dogs), foraging (truffle pigs) and messaging (carrier pigeon).
A digital pet would be more ubiquitous and interact reactively and proactively aligned with wider goals. Digital assistants would follow this pattern. The digital assistant can correct you if you write something that does not align with an SOP for example but also answer questions about what to do. There can be dedicated autonomous assistants for work, for personal stuff and as digital entities offering their help to the market.
Like with real world pets we would expect digital pets to help in particular areas. Similar to guard dogs we could have a security assistant roaming our digital presence across devices or in company infrastructure. With pets we make sure that they stay within their expected area of use by feeding them. The same could be the case for digital pets. Only those that behave in a helpful manner would be rewarded with treats. It is not clear exactly how that would work, but one mechanism I have suggested is that the agents somehow “pay for their own subsistence. They would earn tokens based on humans rewarding them for their helpful behavior. These tokens would help them pay for digital services like infrastructure. When they run out of tokens because they have not been helpful they would be deleted or the service they depend on for their existence would be discontinued.
Rather than prepare for one universal human-like intelligence that could potentially take over the world, we should envision a wide variety of types of digital entities that have different reaction patterns and with which we will have to relate differently. This will help us actively manage the digital ecosystem of the future to our benefit and help us harness the power of digital and artificial intelligence. It will also give direction to what we want to develop. In the digital world we can decide what to build and we can align expectations of what intelligent systems can and should do for us in the future.