Author: Shabbir Chunawalla

The very first wave of AI was predictive AI. It enabled businesses to forecast trends and make businesses data driven. The second wave of AI consists of generative AI which generates content (text, images, voice, code) and facilitate conversations with humans. Time has come to recognize the third wave of AI — agentic AI. Here AI systems become autonomous to execute tasks. They can also interact with other AI agents.

Agentic AI is different. Here the AI agents go beyond prediction and generation of content. They are action oriented. They can interact with other AI agents. They make decisions (within defined parameters). They can execute complex tasks. Agentic AI can automate the entire tasks. They can perform on behalf of us. It is a significant leap.

AI agents can staff the workplace. They can play the assigned roles. They augment human capabilities. These agents can be trained. There could be co-ordinators of AI workflow. There are team managers. There could be automation and there could be a hybrid model of AI agents and humans.

AI agents are useful in customer service. AI agents can act as returns manager for e-commerce firms. AI agents can make a deal with a car rental company. AI agents can prepare a medical history pf a patient. AI agents can settle insurance claims. In finance, they can resolve complaints.

These agents must work as a team. They can attend meetings. They can contribute by giving valuable inputs. AI agents and humans could have a blurred boundary.

There could be errors of judgement. Agents can err. It is disastrous. Thus. guardrails must be set up. Humans must have oversight over agents — essentially must exercise control over critical points.

The relationship between AI agents and humans is evolving. In future, every executive would have his own agents. Organisations can have specialized agents. There should proper delegation to AI agents.

Work environment has to be reimagined. Workflow must be broken up to be divided between humans and agents. We have to collaborate with AI agents in future.

To sum up, predictive AI focused on historical data to forecast future outcomes. Generative AI created new content, e.g. ChatGPT for text, DALL-E for images and AlphaCode for programming. Agentic AI focuses on autonomous decision making and environmental understanding. It sets and pursues goals. It adapts to dynamic environments. It acts with minimum human intervention. The examples are personal digital assistants, robots managing supply chains, or software trading autonomously in financial markets.

Agentic AI is a significant leap since it shifts from assisting humans to acting independently. Of course, there are issues of safety, accountability and societal impact. After all, it is a matter of balancing innovation with regulation.

Agentic AI system has a workflow — a process where it autonomously plans, decides and executes actions to achieve certain goals in a given environment. Traditional workflows are pre-defined and static. Agentic AI workflows are dynamic, adaptive and aware of the context. The process requires decision making and action without human intervention.

Microsoft’s Agentic Cookbook focuses on implementing generative AI agents using tools like AutoGen. It is a practical guide for developers looking to integrate Agentic AI to workflows (GitHub).

Sam Altman, OpenAI CEO, expects to reach AGI within next five years, with the current hardware. Anthropic CEO, Dario Amodei predicts AGI will be achieved in 2026 or 2027. His prediction is based on extrapolated curves of the progression of AI models. The limitations the corporates face while developing advanced models are the lack of high-quality data for training and adequate funds.

There is a lurking fear of the termination of humanity with the advancement of AI. Still, they pursue the AI bubble. There is no possibility of AI reaching its end point — it is an endless cycle.

Advanced models are capital intensive and are difficult to maintain. There is a mad rush for supremacy. Corporates launch models in rapid succession. With most models having similar capabilities, corporates aim to reach the AGI. Still, it remains elusive. It requires vast resources, touching trillions of dollars to get infrastructure of semiconductor chips and extra data centers.

As we have previously observed, Altman finds AGI, ‘a few thousands day away.’ He has now a more defined time frame of 5 years.

Altman thinks AGI may not have a profound impact on mankind. It may whoosh by. He admits that AI revolution could lead to radical social changes and quick scientific progression.

Bitcoin has surged amazingly high with its price ruling within a touching distance $80,000 per Bitcoin. Escalation of this nature is based on the expectation that Trump could create a Bitcoin Strategic Reserve. All other cryptos too have gained — Ethereum (30%), Dogecoin (50%) , Cardano (70%). China has been quietly buying Bitcoins.

One nation state is acquiring Bitcoins, and is now among the top five holders of the bitcoins. The top five countries holding Bitcoins are the US (2 lac Bitcoins), China (2 lac Bitcoins), the UK (61000 Bitcoins), Ukraine (50000 Bitcoins), Bhutan (13000 Bitcoin) and El Salvador (5000 Bitcoins). Most of these holdings are through seizures through donations.

El Salvador, we have already observed, has adopted Bitcoin, as legal tender. Bhutan, the Himalayan kingdom, is into Bitcoin mining. Qatar is buying several billion dollars’ worth of Bitcoins.

Trump would like the USA to be the ‘crypto capital of the world’. US senator Cynthia advocates a Bitcoin Reserve to hedge the national debt. It would like to buy 1 million Bitcoins over five years. Pennsylvania would like to ensure Bitcoin as payment method and would like to bring taxes on Bitcoins. They would like to pass a crypto bill.

Bitcoin rallied past $84000 for the first time as Trump is considered digital-asset-friendly and the prospect of a Congress with pro-crypto lawmakers.

In a recent memoir, Le tells the story of ImageNet (The World I Saw). It was a large database. Prior to ImageNet, people did not believe in data. Li pursued the project for more than a couple of years. All this happened at Princeton (2008). She took a job at Stanford in 2009. She thus took the project of ImageNet to California.

In 2012, a team from the University of Toronto trained a neural network on the ImageNet dataset. There was amazing image recognition. The model was called AlexNet after its author Alex Krizhevsky. This set into motion a deep learning boom that has been continuing till today.

AlexNet was assisted by CUDA platform of Nvidia which converted GPUs into non-graphics applications. Fei-Fei Li pursued the unorthodox ideas. The second visionary was Geoffrey Hinton, a computer scientist at University of Toronto. He promoted neural networks for decades despite skepticism. The third visionary was Jensen Huang of Nvidia.

Hinton teamed up with two of his former colleagues at USSD Rumelhart and Williams to describe backpropagation for efficiently training neural networks in a landmark 1986 paper. Of course, backpropagation was not discovered by them, but their paper popularized it. Hinton moved to the University of Toronto in 1987.

One French computer scientist Yann LeCun was attracted by Hinton here to do post-doc work with Hinton before moving to Bell Labs in 1988.

Backpropagation facilitated handwriting recognition, and by mid-1990s LeCun’s technology was adopted by the commercial banks to process cheques.

Neural networks were not suitable for larger and complex images.

GPUs have many execution units which are tiny CPUs. All are packed on a tiny chip. There is parallel processing which results into better image quality and higher frame rates. GPUs were introduced in 1999. In mid-2000’s Nvidia CEO Jensen Huang suspected that a GPU can be useful beyond gaming, say for weather forecasting or oil exploration. In 2006, the CUDA platform was announced by Nvidia. Here programmers write kernels — short programmes to run on a single execution unit.

Kernels allow the computing task to be split up into bite-sized chunks which could be processed in parallel.

In 2006, CUDA was thought to be a useless thing. In 2008, Nvidia’s stock declined by 70 per cent. In 2009, CUDA was downloaded and reached a peak. Again there was a decline for three years. Huang did not conceive of neural networks or AI when CUDA was thought of. It was Hinton’s backpropagation concept that split up the task into bite-sized chunks. Hinton was quick to recognize the potential of CUDA. Human speech recognition was made possible using CUDA platform in 2009. Still, Hinton was never given a free chip by Nvidia. They obtained the Nvidia chips GTX580 for AlexNet project (Hinton, Alex and Sutskever).

At Princeton, Li wanted to build a comprehensive image dataset. She was suggested to use WordNet where 1.4 lac words were organized. She called her new dataset ImageNet, since she used WordNet as a starting point to choose her categories. Verbs, adjectives and intangible nouns were eliminated. There were then 22000 countable objects ranging from ‘ambulance’ to ‘zucchini’.

She adopted Google’s image search to find candidate images, and used a human being to verify them. The images were chosen and labelled. However, it was a humongous task that would take years.

She then learnt about Amazon Mechanical Turk which cut the time to complete ImageNet to two years. ImageNet was ready for publication in 2009. It was presented in computer vision conference. Still, it did not get the type of recognition Li expected. She made a smaller dataset with 1000 categories and 1.4 million images. She arranged a competition in 2010, and 2011. Still ImageNet was too much for algorithms to handle. A third competition was held in 2012. Geoffrey Hinton’s team submitted a model based on deep neural networks. It gave amazing accuracy. The winners were to be announced at the European Conference of Computer Vision. Li was not inclined to attend since she had a baby to nurse. Then she witnessed how well AlexNet has worked on her dataset. She reluctantly attended. Even Yann LeCun too was in the audience. It was a turning point in the history of computer vision. It endorsed Hinton’s faith in neural networks. AlexNet was CNN that LeCun has developed 20 years back for recognizing. There were few differences between AlexNet and LeCun’s image recognition networks of 1990s. AlexNet was far larger. It had 8 layers and 60 million parameters. LeCun could not have trained a model of this magnitude in the 90s since there were no GPUs. Even collecting images would have been tough in the absence of supercomputers, Google and Amazon Mechanical Turk.

Li provided the training data that large neural network needs to reach their full potential.

Hinton and his students formed a company, which was later purchased by Google where Hinton worked while retaining his academic post at Toronto.

AlexNet made Nvidia chips the industry standard for training neural networks.

Three elements of modern AI converged for the first time — neural networks, big data and GPU computing.

Nvidia, a very successful company making AI chips, is known for entering new markets with astounding success. Huang, Nvidia’s CEO, transformed its gaming graphics cards company into one leading the AI revolution.

Huang in 2024 is talking about AI’s next wave. He believes robots will bring AI that conforms to the laws of physics and interprets the world around them. He visualizes a world with robotic factories and many products around us will be robotic. He also sees a world with humanoid robots in the coming years.

Nvidia is the market leader in chips that can train and run generative AI systems. He would like to focus on three other areas for potential growth — autonomous vehicles, quantum computing and robots. Nvidia cannot be for years to come just one product company. It is too vulnerable if there are changes in technology. There is a heavy demand for AI chips, but this cannot last for ever.

Nvidia can think of chips for robots, the software and the hardware. It can integrate these functions seamlessly and build the whole robots. There could be legal issues, say anti-trust laws. The company lacks the supply chain and manufacturing expertise for building robotic hardware. Jumping into robotics manufacturing will erode its profits from AI chips. Nvidia has to focus on developing new markets for its chips, rather than jumping into building the whole robots. As a person Huang may not like to enter the industry in a fragmented way. However, it is advisable for it to build the brains and tools for robots. That precisely defined market may not be as profitable as AI’s gold rush today. However, it puts Nvidia in an enviable position — it plugged into robots too.

Since the advent of computers, ‘learning to code’ is a valuable exercise. Yossi Matias, Head of Research, Google calls it a basic skill which is as relevant today as it was when the computers arrived. The catchphrase ‘learn to code’ became popular in the 2010s. In 2020s, ten years later, it still holds in this of age of AI.

These days corporates lean on AI for some coding duties. There are tools such as GitHub Copilot. There has been a reduction of 70 per cent in coding time. It could affect the career of software engineers, and they suffer from anxiety. Coding is an essential part of computer science course.

Some engineers face challenges in placement and are denied the experience they could have got. AI facilitates coding at junior levels, and still AI has not taken over the whole coding process. According to Sunder Pichai, a quarter of all code is now generated by AI. However, the code is still reviewed and approved by engineers.

Matias still believes that coding skill should be ubiquitous. Basics are more important now than ever before. There are opportunities to build now on your basic skills. Even those who are not software engineers must understand how technology works. Technology has now become mainstream especially after the arrival of ChatGPT in 2022.

AI will transform many areas — biology, chemistry, medicine. Google has integrated AI tools to flood forecasting models. Image classification tools help in radiology and healthcare.

AI alerts ordinary people to see a medical practioner. Education too will change on account of AI. AI can create quizzes to make education effective.

Since AI affects many aspects of life, it is important to master the basics such as fundamentals of coding.

By definition, artificial general intelligence is intelligence that can reason like humans. It is still a theoretical concept. It is a hypothetical concept that can perform any human task through methods that are not constrained to its training. Sam Altman calls it will ‘elevate humanity,’ Others call it God-like AI.

It is still debatable when we will attain it. Corporates are working on it. OpenAI’s o1 model is a step in this direction. Miles Brundage who recently left OpenAI believes some form of AGI will manifest in the next few years. Dario Amodie, CEO of Anthropic believes that some form of AGI could be seen by 2026. He prefers to call it powerful AI. Geoffrey Hinton puts a timeframe of five years to realize AGI. It could extend to 20 years. These are uncertain times. Demis Hassabis, CEO, DeepMind and a Nobel laureate feels it is a decade away. Andrew Ng is conservative about his time estimate. Richard Socher, formerly of Salesforce defines AGI in two ways — automation of jobs which are AI-powered and an intelligence that can learn like humans. It could be achieved as early as a decade or could take 200 years. Yen LeCun is not optimistic about AGI’s arrival soon. It may take years. And it will not be an event. It is going to be a gradual process.

It was in the year 1993 scientists considered teleportation is feasible. IBM has released a paper about teleporting a quantum state in Physical Review Letters — the paper talked about teleporting a quantum state, rather than just an object. In 1998, this was put into practice by the physicists from the California Institute of Technology and the University of Wales, UK by teleporting a photon — a particle that carries light. It was done through coaxial cabling used to connect satellite signals and broadband internet.

The ability to move something across physical space is amazing. Scientists are sure that a breakthrough in quantum computing could make this possible.

Though the experiments have relied on photons, in 2020, it was discovered that even electrons could be teleported. Electrons would be able to sustain their quantum state for longer periods of time.

This could lead to transportation of more complex bodies. The porting could be applicable to whole atoms, molecules and some human test object. Whole humans? Is it possible that the collectivity of the particles inside the human body can be reassembled at its destination? Can the body which has broken up atom by atom, cell by cell add up to one intact body after teleportation?

Quantum computing, as we have observed in previous write ups, is based on quantum entanglement. It is an area of quantum mechanics where matter and energy behave weirdly at the sub-atomic level. The state of physical properties between entangled particles are transferred from one particle to the other, regardless of distance. Such properties include position, momentum, spin or polarization.

Classical computers are based on two states — 1 or 0. Quantum computing runs on qubits or quantum bits, where the two states exist simultaneously. It is called (coherent) superposition.

A qubit can perform two computations at once. If these qubits are linked together using quantum entanglement, it will enhance computing power exponentially. What a supercomputer can take thousands of years, a quantum circuit could calculate in 200 seconds. This is a practical example of the leverage of entanglement. The other application is its use as a tool of teleportation.

In 2002, scientists teleported particles using quantum entanglements. In 2016, a particle was teleported in Canada for six kms. In 2017, Chinese scientists teleported photon from earth to a sabellite orbiting the earth above 186 miles. There was a critical milestone in teleportation in 2012. Photons were teleported between two land masses in Spain’s Canary Islands. It was teleportation through open air, and no other medium such as cable was used. This is akin to teleportation in science fiction.

Teleportation is based on a mysterious force that keeps the physical states of distant particles in sync.

If there are more than two particle (say particle A, B and C), then properties of particle A could be transferred to C, after A and B get entangled. This occurs though A and C are not in contact. In other words, A has teleported to C. Einstein called such effects ‘spooky action at a distance.’

Researchers think that entangled particles communicate through wave function.

When a quantum state is applied to an entangled particle, the original particle’s quantum states is destroyed.

This collapse of the original properties makes teleporting humans an ethical issue. This is the holy grail of teleportation.

It is to be noted we are not transporting matter, but the information that characterizes quantum state.

John Clauser, Nobel laureate, 2022 asks you a thought-provoking question. If teleporting atoms in the body are disassembled so that you are destroyed, would it be okay that a replica of yours starts moving around afterwards?

Despite the speculations about the advent of AGI becoming an everyday affair, it seems to be real now when Sam Altman, CEO, OpenAI in conversation with Gary Tan said that AGI could be within reach as early as 2025. In fact, they are getting there faster than people expect.

They are on the right track and know what is to be done. It is difficult but still it is all very exciting. Altman’s optimism is based on the internal developments. Outside the company, the very mention of AGI is treated as a crazy and impossible idea. Still they keep pushing the envelope. OpenAI may lack the resources others have, but they have the focus. It is the median view of the researchers at OpenAI. AGI is closer than what most people think.

They had published a paper in which agents employ self-supervised learning with multiple rounds of emergent strategies. There is multi-agent competition. It adapts to complex environments. It leads to skills similar to human tasks. All this was five years ago, where OpenAI reached the third stage — agents. First was chatbots, second the reasoners, and third the agentic experiences (L1 to L2 to L3). The next stage L4 will be that of AI creators and L5 will be that of organizational AI.

OpenAI’s latest model o1 is a significant leap towards AGI. It will advance to Agents with human-level reasoning. (Level 3). It is for this reason that Altman declares that AGI is achievable with current hardware.

OpenAI will also launch its own GPU chip in 2026. Nvidia has already delivered its advanced Blackwell AI chips to OpenAI and Microsoft in October, 2024.

Recently, OpenAI introduced a crucial clause in its contract with Microsoft — as and when OpenAI achieves AGI, Microsoft loses access to OpenAI’s technologies. It is a leverage for a better deal. Under the terms of contract, it is OpenAI which will decide when AGI has arrived. It could be a loophole to get out of the contract. It is, however, certain that Microsoft will not stop OpenAI to realize its AGI dreams. In a developers’ conference, in Dec. 2023, both OpenAI and Microsoft declared that their partnership would help build AI together. The support of Azure cloud was crucial for OpenAI.

It should be noted that OpenAI and Microsoft are not the only ones in the race. Google has rolled out Gemini 1.5, Gemma 3, NotebookLM and Data Gemma. Gemini 2 has better reasoning. Meta plans to release Llama 4 by early 2025. It pushes the company towards Autonomous Machine Intelligence (AMI). There are others such as Anthropic, xAI. Just getting close to AGI is a big miracle.

Yann LeCun, the French American scientist and one of the godfathers of AI, advised youngsters in their twenties to concentrate on subjects which have a long shelf-life, e.g. physics, basics of computer science and applied mathematics. All these subjects contribute to help you understand the next generation technologies and AI systems.

His comments should be appreciated in the backdrop of computer coding such as Google’s Gemini Code Assist, Microsoft’s GitHub Copilot, Amazon’s Q Developer (formerly Code Whisperer) and AI software agents such as Devin (Cognition Labs).

A twenty-year old wonders whether it is worthless to learn computer science since AI systems are going to programme better than he can. LeCun is of the opinion that this is not true.

A mobile app course has a short shelf-life, say three years whereas quantum mechanics has a long shelf-life. While exercising your choice, opt for quantum mechanics.

LeCun now works as VP and AI Chief Scientist at Facebook.

To people in their thirties and forties, LeCun advises not to put all their eggs in one basket, thinking that it will be the next big thing. There is a change every 3-5 years.

Certain choices make you a prisoner of a hypothesis about the trend in technology. There could be a complete upheaval.

LeCun currently runs FAIR Lab — Fundamental AI Research. He works on next generation AI systems. They perhaps would not be based on LLMs. The new systems would understand the world. They have a persistent memory; they can plan and reason. It is a major challenge for the next five to ten years.

India should be a part of the research community. They then would be able to leverage AI systems.

Almost a decade back, they created a research lab in France. It influenced the local eco-system. Students who could have gone into finance diverted to PhD in AI. Some promising candidates have been hired at FAIR. Most of them are a part of the eco-system and have a set up startups.

India could replicate this model. Maybe some cities here, say Bangalore or Chennai can take the lead.

Unlike Hinton and Bengio, the other two godfathers of AI, LeCun is optimistic about the benefits of AI. He advocates open-source research. He was in Bangalore for Build with AI Summit.