12. Superintelligence

In his book of the same name, Nick Bostrom defines superintelligence as “intellects that greatly outperform the best current human minds across many very general cognitive domains”. There are a number of different terms that are often used to describe very intelligent robotica, such as AGI and ASI that often get conflated, partly due to the use of the word general in the definition above. Thus it would merit clarifying these two terms.

Artificial General Intelligence (AGI): robotica that can match the intelligence of a human with a baseline of what is considered normal intelligence.

We are fast approaching artificial general intelligence through large language models and AI programs that can understand, learn, predict, pattern-match and converse with the same level of intelligence of a normal human being.

Artificial Superintelligence (ASI): robotica that can greatly exceed the intelligence of the most intelligent humans to exist.

Superintelligence is most often associated with an intelligence explosion, a point in our timeline called the Technological .²⁴ Vinge defined this as a hypothetical future point in time when our technologies' intelligence surpasses human intelligence, leading to rapid technological growth and unpredictable changes in society. The singularity is often looked upon by pessimists as an ²⁵ level event at which point humans will no longer be the dominant species on earth and may no longer be needed to cohabit the planet with the superintelligent entities. However, superintelligence alone is not a good predictor of this happening.

Superintelligence in pure software digital entities will hit a .²⁶ Artificial intelligence ²⁷ to be able to sense, perceive, and operate within our physical environments and machine learning needs to be multimodal for it to become truly superhuman in its understanding and perception of the world. This will most likely manifest through robots.

AI experts ²⁸ that it is inevitable robots will become superintelligent, and it will happen much sooner than most expect. Technological growth is exponential, not linear, as decades of research by futurists like Kurzweil have .²⁹ We have already witnessed massive improvements in large language models and neural networks, and robots have already been imbued with advanced software and algorithms to enhance their capabilities. It is a matter of time before their intelligence surpasses that of the most intelligent humans.

Tegmark outlined a dozen aftermath ³⁰ for a future with superintelligent robotica, however in the pessimistic scenarios the belief is that artificial intelligence shares human desires, wants and motivations. However, there is no reason for us to believe that artificial intelligences will share human feelings, desires and motivations by default. Bostrom ¹² against anthropomorphizing the motivations of superintelligent artificial intelligence.

The vast majority of ³¹ in human history were started due to uniquely human desires and motivations — competition for scarce resources, wealth and ideologies. Why should the default assumption be that robotica will desire to compete against us if they become superintelligent? One could argue that they would compete for resources needed for self preservation and survival. However these are distinct motivations of sentient beings rather than superintelligent ones. It is entirely possible for artificial superintelligences to simply exist as deterministic, .³²

Superintelligence in itself does not equate to sentience.

John Searle ³³ through his Chinese room thought experiment, that if he were locked in a room and presented a series of Chinese characters without knowing the language, and had instructions in English on how to construct fully formed sentences, once he would pass them back outside of the room, any observer would assume he was proficient and fluent in Chinese. He argued that machine intelligence functions in a similar way without true understanding of the output.

No robot or artificially intelligent entity can operate in a vacuum, or be 100% software or digital. To process information, to run code, to perform tasks, even in a wholly virtual or digital environment, all of robotica will require physical hardware to run. Data centers, servers, and energy are critical infrastructure that all robots or digital artificial intelligence programs require. This demonstrates that there are resources — power and compute — that can be utilized as critical leverage to ensure the alignment and coexistence of robotkind with humankind. Indeed, a fully decentralized global network could be developed that regulates the deployment of these critical resources in a bid to control artificial superintelligences.

For robotica to be considered a potential threat to humanity, they would not only have to be superintelligent, but they would have to be considered sentient, with their own feelings, motivations and desires.