Q 711. What are the different types of AI based on its capabilities? Provide some use cases for each level. What are the challenges or limitations in advancing from one level to the next and then to the next?

 

Note for website visitors -

• This platform hosts two weekly questions, one on Tuesday and the other on Friday. 
• All previous questions can be found here: https://www.benchmarksixsigma.com/forum/lean-six-sigma-business-excellence-questions/. 
• To participate in the current question, please visit the forum homepage at https://www.benchmarksixsigma.com/forum/.
• The question will be open until Tuesday or Friday at 5 PM Indian Standard Time, depending on the launch day. 
• Responses will not be visible until they are reviewed, and only non-plagiarised answers with less than 5-10% plagiarism will be approved. 
• If you are unsure about plagiarism, please check your answer using a plagiarism checker tool such as https://smallseotools.com/plagiarism-checker/ before submitting. 
• All correct answers shall be published, and the top-rated answer will be displayed first. The author will receive an honorable mention in our Business Excellence dictionary at https://www.benchmarksixsigma.com/forum/business-excellence-dictionary-glossary/ along with the related term.

• Some people seem to be using AI platforms to find forum answers. This is a risky approach as AI responses are error prone as our questions are application-oriented (they are never straightforward). Have a look at this funny example - https://www.benchmarksixsigma.com/forum/topic/39458-using-ai-to-respond-to-forum-questions/

• We also use an AI content detector at https://quillbot.com/ai-content-detector. Only answers with less than 45-50% AI-generated content will be approved.

Artificial intelligence or AI refers to the ability of the machines (especially computers) to perform human like intelligent tasks. It may be language understanding, learning, problem solving, reasoning, and may also be a physical action (robotic machines). It works by processing large amount of data to make predictions and take decisions.

 

Capability wise, AI can be classified as follows:

1.     Narrow AI

2.     General AI

3.     Super AI

 

Let us see one by one about each level, along with some use cases, its limitations and challenges in advancing from one level to the another:

 

Narrow AI:

This type of AI also called as Artificial Narrow Intelligence is designed to perform a specific task or set of related tasks. It has limited constraints as it doesn’t have generalised intelligence like humans. Meaning, unlike humans who can adapt and learn to handle wide range of situations and solve various types of problems, these AI systems are typically trained for one domain or function. 

Some theoretical use cases of Narrow AI are:

Voice assistants: Siri, Alexa and Google Assistant use NLP (Natural Language Processing) to understand the request / prompt and perform simple tasks like answering questions, setting reminders, making calls, etc.,

Image recognition: Used in facial recognition software like Siri, android face unlock, etc.,

Autonomous Vehicles: AI algorithms are used to detect objects, follow traffic rules and make driving decisions within specific environments

More importantly the AI assistant tools like ChatGPT, GEMINAI and Copilot are also examples of Narrow AI

 

General AI:

This type of AI also called as Artificial General Intelligence has the capability to understand, learn and apply artificial intelligence across a wide variety of tasks similar to human cognitive abilities. It could also transfer knowledge from one domain to the another and think, reason and plan autonomously

Some theoretical use cases of General AI are:

Scientific Research: It will be able to conduct research across multiple scientific disciplines, generating hypothesis and conduct experiments without human intervention

Healthcare and Diagnostics: It could also simultaneously diagnose multiple diseases by analysing symptoms, patient history and medical research

Robotics: An AGI powered robot could adapt to any new environment and can perform simple to complex tasks starting from doing house hold chores to assisting in dangerous environments like nuclear plant related works. This technology can also be used in space exploration where it is risky and complicated for humans to perform such tasks

 

Super AI:

This type of AI also called as Artificial Super Intelligence is an advanced stage of AI where it surpasses human intelligence in every field. Super AI would not only be able to perform tasks but will also have an understanding of emotions, reasoning and social contexts far superior to human brains

Some theoretical use cases of Super AI are:

Defence and Security: Super AI can be used in defence & security purposes like Global peacekeeping, cybersecurity, etc., It could predict and mitigate conflicts before they happen, using advanced simulations and geopolitical analysis and also it could autonomously defend against cyber-attacks and can enhance the security of organizations and nations.

Solving Global Problems: Super AI could tackle complex global issues like climate change, poverty and disease by offering solutions beyond human capability

Human Augmentation: Super AI could also be used in enhancing human abilities by improving areas such as        health, memory and creativity through brain-machine interfaces

 

Challenges and Limitations:

The capabilities of Narrow AI are already in active use today across industries like Customer service, entertainment, healthcare, finance, etc., The General AI is a long term goal and is still theoretical. Whereas, Super AI is still theoretical and considered as an aspirational but distant possibility.

Limitations of Narrow AI to advance into General AI is that it lacks the ability to generalise across different tasks and building such system requires significant progress in learning algorithms, memory and cognitive understanding

Similarly, challenges of advancing from General AI to Super AI includes incorporation of Human cognition complexity, requires enormous details on computational resources, vast datasets and training environments and also more of human ethical concerns such as decision accountability, impact on human employment, etc.,

 

Advancements in AI research are continually moving us towards AGI but it will likely take many more years or even decades to achieve.

Levels and Use Cases of AI Capabilities

 

There have been various ways through which the capabilities of AI can be classified, but the general view on levels commonly referred to includes:

 

Limited or Weak AI:
These include those systems designed to carry out specific or particular limited tasks; they can never perform beyond the limits set for their functioning. Examples include Siri, Amazon's Alexa, IBM Watson, and OpenAI's ChatGPT.

Strong Artificial General Intelligence (AGI):
A concept that is, as yet purely theoretical. AGI shall be capable of successfully engaging in any intellectual activity of which a human is capable. Apply much previous knowledge, many skills, and experience to decide what actions to perform in order to achieve goals in an assortment of complex environments without any human guidance.

Artificial Superintelligence (Super AI):
Yet another theoretical that is yet to be achieved. Super AI would outperform human intelligence in that their ability to think, reason, learn, and make decisions would be better than those of any human mind.

Reactive Technologies:
These AI systems have only a limited number of inputs to which they would react and cannot learn or remember. Examples are vending machines or traffic lights.

Limited Memory AI:
These are systems that learn from data how to do better in the future and then store bits of information for later use. Image recognition systems and natural language processing systems are examples.

Theory of Mind AI:
At this level, a system understands and will respond to human emotions and needs; it will finally converse and interact like a human. Virtual assistants and chatbots come into this category.

Self-Aware AI:
This is purely theoretical at this point in time. A self-aware AI would have its very own consciousness, therefore be able to fathom its existence and, using this, to form goals and motives.

Human-Level Intelligence AGI:
The work that this AGI could perform would be any intellectual task that a human carries out and apply knowledge to a wide range of activities. But this level of AI is still purely theoretical.

Superintellectual Intelligence ASI:
It is conceptual in form, where AI performs intellectual tasks much faster and better compared to human beings. Again, no such system exists as of now.

 

Tools/ Architectures/Systems of AI Implementation:

Knowledge Graphs:
These have been used for knowledge representation in the form of interconnected graphs of ideas and relations. Various examples include Google Knowledge Graph and Microsoft Concept Graph. They are scalable and adaptable, though handling ambiguity, especially over massive amounts of data, remains very challenging.

Cognitive Architectures:
These are human cognition or thinking and decision-making process models. Examples include LIDA, ACT-R, and SOAR. These can realize genuinely human-like thinking but are complex and require extensive and higher-order knowledge in cognitive science.

Machine Learning:
Deep learning methods, reinforcement learning, and transfer learning are some of the other machine learning techniques that factor into how AI learns from data, furthering its ability to make adaptations to new challenges. While it is possible to apply machine learning broadly in most situations, often this requires a great deal of big data, and it may not always generalize very well to new situations.

Hybrid Approaches:
Combinations of several methods, such as knowledge graphs with machine learning methods, which extend AI's common-sense reasoning. While effective, these hybrid strategies are complex in implementation owing to the expertise needed in many diverse fields.

Multi-Modal Learning:
It also enables the training of learning from various sources of data, including computer vision, natural language processing, and even audio processing. Of course, it can provide greater insight, but all at once it is complicated to work upon; hence, expertise from different domains is associated with it.

Cognitive Computing:
While the development of human-like reasoning in systems such as IBM Watson or Microsoft Azure Cognitive Services has much better insight into the world around, it is hard to develop and apply in practice.

Explainable AI:
These systems can explain the justification and reasoning of their decisions using various techniques such as saliency maps and feature importance. Explainable AI works to increase transparency and trust but, in turn, involves high degrees of AI and machine learning knowledge.

Human-AI Collaboration:
Such AI systems work with humans to solve complex problems. Examples include platforms such as AutoML from Google, which melds human intuition with AI insights. However, developing such systems is somewhat complicated and demands an understanding of both artificial intelligence and human-computer interaction.

 

Challenges in Developing General or Strong AI:

Making Artificial General Intelligence or even stronger variants of AI is highly uphill for several reasons. These are explained below.

Understanding Human Intelligence:
Human intelligence itself is not well understood, which can perhaps be one point as to why it's so difficult to mimic these actions within machines.

Complexity of the Human Brain:
The brain is an immensely complex system, and it has proved very challenging to model using the rich knowledge acquired in neuroscience, cognitive psychology, and computer science.

Lack of Data:
An AGI requires truly large-scale diverse and high-quality data, which does not exist today.

Algorithmic Limitations: The current AI systems struggle both to learn and generalize. New algorithms are needed toward the kind of adaptability needed for AGI.

Transparency and Explainability:
Most AI systems are black boxes, where their decision-making process is not transparent. AGI should be much more transparent if people were to trust it.

Safety and Control:
The AGI systems must be controllable to avoid accidents due to their unintended outcomes, but how to make AGI controllable is not very well understood.

Value Alignment:
The AGI systems should be aligned with human goals and values, but that too is a problem to be resolved by the researchers.

Cybersecurity:
AGI should be secure against cyber threats, but how it is guaranteed is yet another open question.

Ethics and Morality:
Ethical decision-making processes of AGI have reached no consensus as to how moral conduct is ensured in machines.

Social Impact:
The social impact of AGI should be in service to society, yet the implications for the future remain to be seen.

Technical Challenges:
One major technical challenge is in developing hardware and software necessary for AGI.

Funding and Investment:
Research and development of AI are very capital intensive, which is not taking place with respect to the development of AGI.

Overcoming such challenges calls for:

Heavy investments in research and development related to AI,
New algorithms, architectures, and techniques,
Ethical, moral, and social issues,
Safely assured, controlled, and cyber-safe,
AGI system alignment with human values and goals;

By solving these problems, we make another step towards the creation of General or Strong AI to help humanity.

Mohammad Riyadh Al Kamal has provided a very comprehensive, detailed and well drafted answer to this question and hence declared as the best answer.