Dichotomy of Agentic AI - on Telecom, in specific.

My Idea of Agentic AI, in Telecom Networks specifically, is about two lines of business or approaches - Inferencing based and Reasoning based.

1. Inferencing based

Inferencing based is for more real-time action, light weight, cost and resource efficient. It can provide silo inferences and holistic inferences with coordination within a system that could be utilized for policy generation, classifications etc.

This is kind of agentic AI uses for mostly technical systems, like telecom network optimizations. Their objective is real-time dynamic actions with short contextual understanding as they are applied in dynamic systems or environments where ask is efficiency and optimization as per the conditions and objectives.

Its primary characteristics from telecom network point of view are

 Worth on Edge

 Low memory

 Low computing

 Real time or Near Real time

 Highly contextual

In telco network, specifically on existing architecture like ORAN, SDWAN, etc. the inferencing Agentic AI could be masterpiece. I guess it is already in the helm of discussion and defining the new standards to use it in widespread manner.

Some of the RAN specific areas where it has most significant role in network optimization.

• Load balancing

• Policies and classification

• Interference management

• Beam management

• Channel Estimation

• MIMO optimization

• Code rate optimization

• Rank Setting

• Neighbor list

• HO optimization

What is Inferencing?

Inferencing for me is to sneak in future window, expect the thing or possible outcomes, it is of my interest as it involves probabilities and possibilities. Inferencing I searched in AI and what is there is very similar to reasoning in definition but it certainly differs at large in its approaches and objectives than reasoning.

Inferencing can be intuitive, contextual, prediction of possible outcomes or assessment of existing and current affairs. That’s where it needs to be real time or near real time.

Also inferencing provide a good classification based on past experiences, emerging trends and pattern recognitions.

Still there are various inferencing ways.

1. Rule-Based Inferencing: Applying predefined rules to data to derive conclusions. For example, if a medical diagnosis system has a rule that says "If a patient has a fever and cough, then they might have the flu," it uses this rule to infer the possible diagnosis.

2. Probabilistic Inferencing: Using probability and statistics to make predictions based on uncertain or incomplete information. This is common in systems like recommendation engines, which predict what a user might like based on past behavior.

3. Bayesian Inferencing: A specific type of probabilistic inferencing that updates the probability of a hypothesis as more evidence becomes available. It's widely used in machine learning and data analysis.

4. Neural Network Inferencing: Involves using trained neural networks to make predictions or classify data. For example, a neural network might infer the category of an image based on patterns it has learned during training.

2. Reasoning based

Reasoning based Agentic AI is about handing workflows, like chatbots response generations, customer managements, maintenance activities, operational activities etc. it could also be in silos or holistic with coordination within a system. It may need expensive resources and processing infrastructure and may be not real-time in nature, as their objective is to generate efficient response with long contextual or holistic understanding.

Its often showcase its deployment with following attributes

 On backend servers

 Need extensive data

 High memory and computing

 Near Real time or offline

 Holistic approach

There are wide uses of this kind of agentic AI in telecom, which include areas like-

• Predictive maintenance

• Anomaly detection

• CI/CD workflow

• Issues resolution

• Customers experience management

• Change management

• Network Planning

• Manual decisions

What is Reasoning?

If Search AI general answers are – it is an ability to process information to provide conclusion, results, decisions etc.

Reasoning is more about logical deduction, relation understandings, inferencing, comparison, match and mixing, etc etc.

If AXZ is CAT and BWY is HEN then CYZ is ANT. Machines are adapt of this already. but AI bring it a enormous scale.

There are various kind of reasoning as defined below, and I find using AI.

1. Deductive Reasoning: Drawing specific conclusions from general principles or premises. For example, if all humans are mortal and Socrates is a human, then Socrates is mortal.

2. Inductive Reasoning: Making generalizations based on specific observations. For instance, if you observe that the sun has risen in the east every day, you might conclude that the sun always rises in the east.

3. Abductive Reasoning: Inferring the most likely explanation from incomplete information. This is often used in diagnostic systems, where the AI suggests the most probable cause of a problem based on symptoms.

4. Analogical Reasoning: Solving problems by finding similarities between different situations. For example, using the solution to a past problem to solve a new, similar problem.

5. Commonsense Reasoning Mimics human common sense to interpret everyday situations. Example: If it’s raining, the ground will probably be wet.

6. Non-monotonic Reasoning Allows revising conclusions when new information comes in. Unlike standard logic, conclusions can change. Example: Birds can fly → Sparrow is a bird → Sparrow can fly But then: Sparrow is a bird → Sparrow can’t fly high → Sparrow can fly but not very high.

Reasoning is used in many areas of AI, like:

 Expert systems

 Natural language understanding

 Robotics

 Planning and decision-making

 Want me to break it down further with real-world AI examples?