Manifesto

Definitions

Most definitions of artificial intelligence (AI) are self-referential. We have no good way to define it other than using our own cognitive abilities as a unit of measurement. Therefore we are going to refer to AI as an artificial system that shares with us - human beings, one or more traits usually attributed to intelligent behaviour.

More intelligent traits an agent can demonstrate, closer it is to our definition of artificial general intelligence (AGI) - an artificial system having the same set of cognitive abilities that a human being has.

“Cognitive spectrum”

“Ethical AI”

Cognitive spectrum

Given this definition of AI it should be safe to assume an existence of cognitive scale or a spectrum that can be used to compare different intelligent agents against each other in their ability to comprehend external environment and manipulate energy and matter.

///Cognitive scale with a human being as a unit of measure

By placing all possible intelligent agents on a cognitive scale, four types of AI can be defined. Using a human being as a unit of measurement lets call them:

1. Pre-human.
2. Human-like and human group-like.
3. Post-human.

Pre-human

This part of cognitive spectrum spans from primitive logical gates to more complex artificial neural networks capable of replicating different functions of our neo cortex. As of writing this document all AI systems lie somewhere on this part of cognitive spectrum.

Human-like and human-group like

Human-like AI systems are often refered to as artificial general intelligence and are indistinguishable from humans in their cognitive abilities. Human beings collaborating with each other across space and time are usually capable of achieving far greater cognitive power than individual human beings. We find it necessary to draw clear distinction between collective intelligence and human-group like AI systems. Former one does not necessarily imply greater cognitive power when compared to individual human beings. Whereas latter one is capable of solving tasks as complex as modern physics is dealing with. Assuming AI systems will retain their capacity to replication, we believe that distinction between human-like and human-group like AI agents is important but not essential for putting them into separate categories in scope of this discussion. Once you are ably to build AGI - further evolution is a linear process. As of writing this document such systems don’t exist.

Post-human

This category can loosely be defined as AI sytems that transcend ability to discover and leverage laws of nature. We believe this level of intelligence is beyond our comprehension.

Case against centralized AI development

Evolution and proliferation of AI technologies is driven by media content monetisation and paid access to proprietary software.

This economic environment rewards non-transparent data flows with little regard to ethical or legal concerns.

Internet becomes overwhelmed with noisy, low quality “content” designed to attract ad traffic.

Media platforms maximise traffic by abusing user activity data and exploit humans’ perception valnurabilities.

Public opinion is manipulated via deliberate introduction of bias in favour of political views of media platform owners or their partners.

We argue that further evolution of AI technologies and growth of their cognitive abilities within traditional business models is incompetible with ethical use of AI. will lead to even greater tensions and negative social consequencies.

Challenges

We believe that to avert further negative consequences of that factors on social life and AI evolution ad-free business models have to be introduced and AI development should follow open source principles. Therefore making all the components of AI system publicly available is the only possible way to ensure it’s ethical use.

Unlike with traditional software AI is a combination of multiple components that include software, hardware infrastructure and data sources; Therefore to make AI publicly available all is components have to be put in a public domain as well; Data storage and compute capacities require funding which have to be secured by providing better alternatives to proprietary services; Possible strategies include hyper specialisation on specific data sources and restriction of data source use via licensing; Revenue from these services has to be shared among system participants; System design to be tailored to the needs of engineers and AI developers;