Updated: Apr 30
This is the first blog dedicated to the ideas my team and I developed at the Indian Society of Artificial Intelligence and Law in the last 2 years.
Whenever we read “AI”, usually, this makes us think about sci-fi movies and robotics, and then the term artificial intelligence comes in the fray. Interestingly, we all know that artificial intelligence is not 1 technology but a family of technologies. What does this even mean?
Now, let me show you a chart made by our team at ISAIL to understand, how to classify AI in its legal and industrial sense:
You can view this chart, and find how scientific, legal and industrial classifications differ.
Classification Of Artificial Intelligence
For example, maybe, some legal aspects of facial recognition technology and unmanned aerial vehicles could be similar. I am not claiming there is a perfect alignment out there. Yet, these technologies in view of their purpose are completely different. So, even if they are like AI, in that pop culture sense, they still are different “species” of artificial intelligence. This is exactly how the idea came into being when my team at ISAIL and I were brainstorming on this issue. This can happen even with general technology systems, such as robots and analytics services. They might be considered like artificial intelligence, but their operational and manifesting value and purpose are completely different. Work robots in a factory which make things easier are not delivering real-time analytics insight and vice versa. Obviously you can mix or merge the features and have 1 system that might try to deliver both of these things. Still, that is very subjective and depends on each and every case, to decide.
The ISAIL Classifications
When the 2020 Handbook on AI and International Law came into being, I had proposed a 2-tier model on classifying artificial intelligence. I will now discuss the basis of the classification in brief and then explain how this classification can help us refine our legal understanding of not just AI technologies, but also other kinds of disruptive technologies.
As the chart explains, there are two kinds of classifications:
Concept, Entity and Industry (CEI)
Subject, Object and Third Party (SOTP)
The first classification has been depicted above, as how they work. As the word Concept is, many argue that AI is an abstract concept, and so it is important to keep ground open for subtle and important positions of policy intervention to redefine the legal understanding of technologies like artificial intelligence, since, like all disruptive technologies, AI is an ever-shaping technology.
The word Entity has a special value in legal literature around the world. In legal terms, an entity may be implied to be a company, a natural person, an NGO or any other corpus, whose rights, duties, liabilities and responsibilities are possible to be defined, not just in principle, but also in the spirit of implementation. Generally, polities across the world provide two kinds of statuses - LEGAL and JURISTIC. Legal means that a law has been framed or a set of regulations have been adopted, which have given a clear-cut picture of how that thing will be reckoned in the legal system, and how will the State address that thing. Juristic means that the status is based on some un-codified intervention, maybe via a court order/judgment or the manner in which that thing has been interpreted by the administrative component of governments (bureaucrats). Here, any technology within the family of AI technologies, can be recognised legally, or given some ad hoc (specific) status. Sometimes, giving status not even equivalent to that of a human, a company or any other entity in that generic sense, is also considered giving some juristic status. Considering that AI technologies require proper auditing and policy interventions, the juristic status works sensibly well here.
The Industry thing is quite simple. AI technologies differ for every industry’s needs, and as those needs are catered, the technology class becomes valuable for that industry. For example, algorithmic trading services cannot be used by a content creator to make AI-based animations. Similarly, facial recognition software, cannot do the work of voice recognition software literally. That itself is based on how classifications are made. Similarly, what facial recognition software can do for Instagram Reels or Tiktok, is not what the latter would do that exact way. Hence, based on a case-to-case use, this is a first principle understanding that industrial needs differentiate the classes of technologies within the artificial intelligence family.
The second classification is also simple. That kind of classification is context-oriented.
As per this excerpt, assume that X is a human being, Y is the “AI” system.
Subject means that X, the human is being subjected to the environment of the AI system, in which it is sharing its data. Object means that the roles are reversed as Y, the AI system itself is subjected to the human environment, since the circumstances in which Y is being used, must be taken into fair account. Third Party is a case where Y acquires special features, which show that it has a sense of explainability and foresight as an AI system. Generally this is not a perfect scenario, but as an ideal case, we have kept it to analyse more developments in the world of technologies.
This is a short graphic explainer of how AI can be classified. Now, there is a principle, which we use to classify. Feel free to read the 2020 Handbook on AI and International Law to know about that principle, in its Chapter 1.