What Is Artificial Superintelligence (ASI)?
Artificial Superintelligence (ASI) is a theoretical concept referring to a complex computer system whose cognitive abilities are far more superior to even those of a human with genius-level intelligence. First theorized by the Swedish philosopher Nick Bostrom in his book “Superintelligence: Paths, Dangers, Strategies”, it is seen as the peak of technological evolution, even if it may never be achievable.
Key Technologies and Methods for Developing Artificial Superintelligence
It should be noted that current types of AI can be characterized as Narrow Artificial Intelligence, since they can focus only on one task of interest: chess playing, picture generation, face recognition, and so on.
Así, on the contrary, can work on multiple different tasks at the same time. It exceeds human and Narrow AI in terms of:
- Speed. According to Bostrom, it can “accomplish an entire millennium of intellectual work in one working day.“
- Collectiveness. It can incorporate smaller AIs to handle numerous tasks.
- Quality. ASI’s Intelligence quality is comparable or superior to that of a human, especially in terms of understanding the world and solving problems.
Therefore, it is assumed that ASI needs to “absorb” multiple technologies before it can achieve its unprecedented status quo:
- Large Language Models (LLMs)
LLMs are key to gathering knowledge about the world in the forms of datasets and also interacting with humans.
- Multimodal AI
Multimodal systems are used in real life, including antispoofing solutions — this makes them more robust to spoofing attacks and malfunctioning. In order to analyze multiple sources — audio, literature, images — an ASI should also be multimodal or multisensory to process various data.
- Artificial Neural Networks (ANNs)
ANNs are modeled after the human brain, specifically neuron clusters, that work in unison to make decisions. An ASI also requires a highly advanced ANN that can perfectly emulate human thinking processes.
- Neuromorphic Computing
This is not just complex software, but also hardware infrastructure that should be able to mimic cognitive prowess of the human brain and nervous system. Neuromorphic computing allows for designing bio-inspired hardware solutions such as Dynap-SE, which is based on how the human hippocampus operates.
- Evolutionary Computation
This is an approach similar to natural selection: it scopes through a series of potential solutions, and then iteratively improves them to pick up the most suitable one.
- AI-Generated Programming (AIGP)
AIGP will pave the way for ASI to write its own code without human intervention. This way, it can auto-adjust its performance and add/remove functionality items if necessary. Currently, LLMs like ChatGPT-4 can compose computer code.
- AI-Generated Inventions
ASI should be capable of inventing essentially new solutions. An example could be the Large Computational Engineering Model Noyron that architected a rocket engine autonomously.
- Brain Emulation
In the field of AI, there’s a dilemma known as Moravec’s paradox. This theory states that high-level reasoning can be adopted by a computer relatively easily, yet it struggles immensely to emulate the perceptive abilities of a one-year-old child. Therefore, it is crucial to emulate the human brain and teach ASI how to acquire knowledge “humanly”.
Advantages and Disadvantages of Artificial Superintelligence
It is suggested that ASI’s introduction can cause both positive and negative consequences on the world scale:
- ASI’s possible advantages
It is proposed that ASI will be able to:
- Process gargantuan amounts of data and make more efficient decisions than humans ever could.
- Invent items, ideas, and solutions that are incomparable in their creativity.
- Execute mundane, yet crucial and even dangerous tasks: bomb diffusion, climate forecasting, luggage checkups, etc.
It is also argued that ASI can be humanity’s “last invention”.
- ASI’s possible disadvantages
Among ASI’s downsides are mentioned mass unemployment, weaponization, malicious usage, as well loss of control over the Superintelligence.
Possible Steps towards Artificial Superintelligence
As proposed by OpenAI, there are 5 categories that rate AI’s prowess: Narrow for utilizing certain tasks, Problem-solving for approaching basic problems, Autonomous that lets an AI act independently, Innovative that demonstrates a high level of creativity, and General that can replace entire organizations.
ASI would be the 6th class, which seemingly amalgamates all of the previous categories. Once they are achieved, they potentially can be integrated to create Superintelligence.
Perspectives of Superintelligence
ASI still remains a concept too vague and there’s no clear method to achieve it in the foreseeable future. Besides, its existence would raise a whole cavalcade of ethical issues, potentially jeopardizing human existence itself. To eclipse human cognition, it first would’ve had to learn to think like a human, which also includes such irrational components as emotions. However, a statement by the OpenAI dated from July 2023, suggests that ASI may arrive just within the next decade.
Antispoofing
