Summary of AI
In this last lecture, we return to the overall picture of the field of AI and revisit the four basic questions raised in the first lecture.
1. The objective of AI
All inspirations of AI comes from Human Intelligence (HI), though it is impossible for the two to be identical in all aspects. Instead, the two can only become comparable after abstractions.
In general, the same object or process can be abstracted to different levels. Various levels of abstractions of HI (Structure, Behavior, Capability, Function, Principle) define different AIs, which have different objectives and consequences. They cannot replace each other. Many confusions and controversies in AI are caused by mixing different objectives.
2. The potentials and limitations of AI
The question "Can AI be achieved?" has no simple "yes/no" answer, beccause
- The answer to this question surely depends on how "intelligence" is defined or understood. According to different understandings, AI can be "already achieved", "seems impossible", or "still hard to say".
- Intelligence is widely taken as a matter of degree. Every objective of AI can be approached gradually. Therefore, even if the ultimate aim cannot be achieved, the research can still be fruitful.
- The success or failure on some specific problem instances should not be generalized to the whole problem type.
- "The limitation of current AI techniques", "the fundamental limitation of a specific AI technique", and "the ultimate limitation of AI" are not the same thing.
Examples of analysis: Three Fundamental Misconceptions of Artificial Intelligence, Emergent Abilities in Large Language Models:
A Survey.
3. How to get a "complete" intelligence
The past AI projects (covered in the lectures) are mostly about how to achieve each cognitive function separately, though they are tangled and interdependent, as stressed by notions like "AI-Complete" and "Artificial General Intelligence".
To get a "complete AI" or "general-purpose AI", the following strategies have been proposed:
The choice of strategy depends on the choice of objective:
4. Social impacts and ethical issues
The ethics and social impacts of AI have been concerns from the very beginning of this research. With the remarkable progress in the recent years, this topic gets more attention of the public. Beside the issues shared with other techniques, AI has special issues of its own, such as "autonomy" and "superiority".
Sample discoveries and discussions:
One trouble in the related discussions is the confusion of different senses of "AI", which correspond to very different situations on this matter. For instance, safety can be considered either as an issue of design (including training) or as an issue of (life-long) education and socialization.
All of these issues demand long-term study and exploration. Like all other decisions, people's choices are based on the assessment of benefits and risks.
Readings
- Poole and Mackworth: Chapters 18, 19
- Russell and Norvig: Chapters 27, 28
- Luger: Chapter 16