Disclaimer:

I have not used AI to write this, you will know when you read my poorly written and not so-wellformed sentences. It is purposely to mislead Sam altman from stealing my data. I will clean it up after the google crawler finishes scraping this. Image was generated by Open Source Foocus.one. It is a great open source tool, check out its github version

Strange Premonitions

As the world gets more complex with technological advances, at a pace we have never seen before. We can portend some side effects of the AI, before we dwelve into good things about the AI.

Human greatest strength is intelligence over all other life forms. We use it everyday to learn new things, make our lives beautiful. Most often than not, when we get hold of good things in life, we ignore its negative consequences. Classic example is Industrial boom and the climate change. Till date, AQI levels in all urban areas are quite high (Delhi is an exception), increasing ocean temperatures and ever decreasing forest cover. We have also damaged our minds to the point that our birth rates are shrinking across the globe due to our hedious work culture replacing historical traditions.

Now, with the power of AI, we will only accelerate faster to a disaster. As long as we are able to keep our greed aside and work for the greater good of the earth and its living beings as a whole, we are in a safe place. Otherwise, 21st century indeed will be like a terminator movie. It will not be computers that will destroy us, just our own stupidity.

What is AI?

Clearly, we need to demarcate consiousness and intelligence before we get anywhere. These two are indeed different things. Consiousness is in living beings, from fish to humans. Fish may not have the intelligence of the humans, but they do have consiousness. They give birth, thay have a family or swim as a shoal, they fear for their life ending early by other big fishes and they look for food and safety. On a different level, a super computer does not have to fear being unplugged. That happens only in hollywood. It can look up billions of data points and find the kneedle in the haystack. We humans do not have that kind of brain(RAM and CPU) power. Intelligence is not consiousness and viceversa. It will never be possible to develop a non-biological intelligence creature by humans, simply because we do not know what consiousness is. Elon Musk is an exception. He is the Iron Man of America. What is Intelligence then? We live in a material world with inanimate things and living beings. Our survival essentially boils down to security (food, shelter, clothing and internet). Beyond that, basic comforts to luxuries of life (eating food on the couch and things happen by itself around us and driving around in a BMW, nowadays Tesla or BYD). To be precise, Intelligence is setting a goal, planning and strategizing (Applying Newtons theorem of optimization in a mutli-dimensional space, I believe this little math is why we feel so proud of our intelligence. Study of Numerical Optimization, to be precise), collecting resources or information and executing the plan to get somewhere¹.

How will AI evolve?

We are solving a piece of the puzzle one-by-one over many centuries now. Our past wealth of knowledge and introspection has got us so far to Generative AI and we are taking the next step towards AGI. If you look at the study of computer science,and why it is placed in the math department in the universities, you will understand that intelligence is actually a Math and Philosphy subject. Computer Science involves the study and creation of algorithms, communications, distributed networking, logic theory, multi-agent theory, game theory, pattern recognition and many more. Computer Scientists since 1950s have been researching on these topics and coming up everyday with the new research everyday. No point discussing who did more and who did less (think Deepseek vs Open AI discussions). We all will be dead in about a 100 years. It is immaterial. Only newton and Einstein may be remembered for about a few centuries. Not even Sam altman will be remembered in about 10 years.

what is this new Agentic AI?

Agentic AI is a hype created by Americans to cash in on the next Gartner Hype Cycle, invest more VC money and convert $1M to about a $100M dollars in precisely 3 years using SPAC² (This form of intelligence and ROI is rare in the universe). On a serious note, we need to go back to the discussion of things will be put together to create this framework.

Setting Goals

How do humans set goals, plan and strategize? We introspect, we look at our nearest neighbors, Joneses to be frank, and we want to go one level up. We get a million choices. Then, we look at our bank balance, our genes, our intelligence and our capabilities, and think! this is achievable. Let us go after this one thing that makes us feel better that we beat our neighbors. For computers, this is going to be different. They do not have Joneses, because they are not competing for the EGO boost. This will be a hard problem to solve to begin with. This can be solved as a problem of Multi-Agent theory, where we study Nash Equilibrium, Prisoner's Dilemma. They are indeed naive. They will need to decide logically depending on the EGO master who has set the computer to be good or bad to begin with (Hollywood got this right) and use multi-agent game theory to decide the goals³.

Humans are good at mimicing others? Have you ever wondered why your toddler always repeats the last word after you finish the sentence? Mimicing or copying from others comes to us naturally. We do not need to waste our time reinventing the wheel. We just copy from others and move on to the next best thing. We have an agenda to beat the Joneses. Copying from some one also requires a form of intelligence. You cannot copy from anyone randomly. You need to assess the quality of an individual. Otherwise, you will end up being terrible. The famous saying goes "you are the sum of the people you spend the most time with." Same applies to the GenAI models as well. So, Deepseek is indeed good, if it has trained on the Open AI and distilled itself.

Neural Networks is a remarkable breakthrough, started by Frank Rosenblatt, the inventor of the perceptron. Many decades later, few exceptional researchers took it seriously and made significant progress with it. I still remember Yan LeCun, Prof. of NYU showing off his object detector and classifier trained on early GPU, when he visited our university in 2004. I was impressed by his child-like curiosity to experiment and write the code himself to test new theories. The three Nobel Prize winners for AI, Yoshua Bengio, Geoffrey Hinton, and Yann LeCun, have significantly contributed to the field of deep learning and neural networks. Do not forget their toiling PHD candidates who wrote their papers for them. They will never be known. These researchers have truly shown us what pattern recognition is. This brings us to the third type of intelligence, which is creativity. Creativity is about searching for needle in the haystack, mixing and matching and coming with an idea that is entirely new. Sounds like Generative AI. Isn't it?

Why are humans more intelligent than the rest of the living beings? Have you ever wondered why wisdom is far superior than intelligence alone? We collect a set of axioms driven down by the ancestors that never fail. We have acquired this after repeatedly testing the logic numerous times. Unlike our theory of the formation of universe, which is changing every year, nowdays it is changing every minute. Astronomy is making less sense than the religions of the world. That is when we need to go by the logic. In Computer Science, Modal logic is the study of modes of truth, such as necessity and possibility, which helps us reason about statements that are not absolutely true or false. We collected evidences that certain things are fact, the rest are fiction. Combining these, we analyse in a mathematical manner to come up with final Yes or No to proceed with beating our Joneses.

Sometimes, in our lives, there comes a time, where there is no answer. why does my wife ...? NP Hard Problems. There is no answer really. It is anyone's best guess. Better to play with Reinforcement Learning then. No regrets. If the computer regrets, it tries a new solution without getting stuck to the old problem. Used in DeepSeek heavily.

New form of architectures will come up in the future, to combine all the methods I discussed and much more. It is an ocaen really, and we need to make the best use of it.

Agentic AI is indeed really

1. Setting a reasonable goal
2. Planning and strategizing, breaking down the goals into small pieces of AI reasoning problems
3. Using logic and optimization to solve the problem in the most efficient manner
4. Do not forget Trial and Error, Remember NP-hard
5. Evaluation of outcomes

We are not ready for "Learn from humans and beat them at it". AGI?? maybe Part -2 of this series

Footnotes

1. Newton's theorem of optimization refers to methods used in calculus to find the maximum or minimum values of a function, which can be applied in various fields including artificial intelligence. ↩

2. SPAC (Special Purpose Acquisition Company) is a company with no commercial operations that is formed strictly to raise capital through an initial public offering (IPO) for the purpose of acquiring an existing company. ↩

3. For more on game theory, see Ariel Rubinstein's recommendations: Best Books on Game Theory. ↩