Agentic AI, our new minions (Web 4.0, as we know it to be our future internet)

     As you read this text, there are big forces(ChatGPT, Google, Anthropic, Perplexity to name a few...) in play that are fundamentally altering the way information/content is presented for consumption over the web. MCP and A2A are two fundamental protocols, with A2A playing a major role in this shift.

    Consider the following use case in various times; 

• A restaurant display its menu and capabilities for a human to order items from that Menu, using catchy and vivid images of the food. The interface as it exists today is/was meant for human interaction. 
• Along came search engines that had to write custom crawlers to handle the various formats in which restaurants were making that information available. The onus was on the search engine to put the effort to understand the formats that were used by restaurants and make them available as answers for humans, so search engines could drive more human traffic to their search bars.
• As more and more started to rely on search engine output to look for menus, rather than typing the address of the restaurant, the power slowly shifted away from restaurants to search engines. The restaurants had to design their website to not only meet the requirement for human consumption but also meet the need of web crawlers, so they could parse the content properly and expose them as search results. The final aim was to ensure that those results would show up as top results for human search queries
• Thus the sites began to be Search Engine Optimized web pages.
• I have not done much this research, but it is reasonable to posit that; many of those restaurants would have sacrificed their web design to support SEO web pages, as search engines started to provide options to order items directly from their own interface and taking a slice of that order pie. This is akin to restaurants only supporting take outs instead of dine-in, post covid(say... cloud kitchen), since it made more sense economically, as well as human behavior to dine in was met with reluctance.
• With the advent of transformers and its subsequent use in developing LLMs, this has exploded the need for further automation with the introduction of AI Agents or Agentic AI.
• The idea is to offload tasks that can be automated
• Like booking your trip to Las Vegas. Given some bounding parameters, the task can be offloaded to an AI Agent, that scours the airlines, hotels and restaurants, reserves seats for you with your credit card and generates a itinerary that you can follow it to a T. (replace your travel agent or even yourself browsing daily for interesting deals)
• As more and more humans start using AI Agents, it is reasonable to posit; a similar shift to develop websites that are Agentic AI aware first, using A2A protocols would become the norm, de-prioritizing regular human users.

    This new priority, IMO is going to change how information is published and it is going to make the use of AI agents a necessity, rather than a helpful tool.

    OpenAI is leading the adoption because of their first to market advantage and Google has its grip on all its legacy users, with others competing for leftovers. I predict that the Phone Apps and Web browser wars are going to heat up with this new push and you can already see this with Perplexity trying to buy Chrome from Google(as part of anti-trust deal).

    Now, comes the interesting part that I foresee; Cloud as we know it, was/is meant to provide an elastic compute, helping customers scale up/down based on demand and not having to pay cost for infrastructure up front. It is like a food buffet but you pay only for the weight of the food that you are going to eat.

    AI Agents require a lot of resources and they need to run offline when your desktop/laptop are powered off or sleeping. How do you achieve that? Offload those executions to the cloud. You interact with the tool via Phone App/Web browser and start an AI Agent in the cloud per task and don't worry about scheduling/running/connectivity of that task. Cloud takes care of running it for you in isolation with all the required compute/network needed.

    With this, I see a second generation or 2.5 explosion in the use of cloud to support web 4.0. This would be only limited by how far we can advance using AI models.

Chatbots(Generative AI) and my thoughts

     Long before chatbots, we had search bars. You would type in keywords and expect to be thrown links across your way. And the expectation is; one of those links would provide an answer to your worldly concerns. With page rank, search history and GPS location, it became possible for search engines to provide quality links to quench your thirst. With all this information, search engines had to do all the backend heavy lifting to correlate all the above and guess with high accuracy about what exactly are you searching for. There is always a plausible deniability, since this is all conjecture up to this point.

    Now comes Chatbots powered by Generative AI(GenAI). A conversational assistant that is to replace an adept human on the other end of the pipe. Don't get me wrong. Chatbots have been there since search engine, but our interaction with these bots never met our expectation. (i,e) typing in a our thoughts and expecting a coherent answer, not a bunch of coagulated gibberish. The information to generate a good answer has always been there and it has been growing at a pace outrunning disk capacities to index and store them. With Large Language Models(LLMs) to predict text and Transformer architecture to provide better context, we have reached a pivotal moment to generate text spewed by Chatbots that is almost human.

    With ChatGPT, Bard and many others, it is prophesied to become the replacement for search bar. (i,e) you type in your ask and you get answers to read, instead of links to peruse. Eg: If you type in "plan my dinner", GenAI would give you a list of things to consider for your dinner like recipe, time, number of people to cook for and stuff like that... instead of just giving you links to other blog posts.

    IIUC, one of the reasons that Google did not release this type of functionality before(even though they authored the research paper on Transformers) is to not to eat their own lunch. The shear amount of processing power needed to generate this type of content is way more than just spitting out relevant links. So the profit per query goes down comparatively(This highlights the need for Antitrust and competition).

    Search terms by themselves do not mean much unless you can tie that to your search history providing a pattern, thereby helping the algorithm to provide better links that suit your need. Search engines would allow information collected from User 1 to be used for User 2 also, if the search terms and patterns are similar. Eg: If you type in "stream oscars for free" and which ever link that you clicked finally and stopped your search, would be first link to pop in someone else's answer, if they use similar search terms.

    With GenAI and Chatbots, you type in exactly what you want rather than search terms related to your thoughts. Now this gives companies like OpenAI and Google your exact thoughts compared to conjecture with search terms and history. IMO, this reveals far more information(intention) about the user, compared to earlier search bar. Chatbots used in therapy, dating apps are prone for mis-use by bad actors due to personal intimate details that we are prone to share in those conversations(hint: blackmail). More people tend to use Chatbots for above purposes, because of their prevalance and in many cases actually helpful. They are more prevalent now, since it is easy to have therapy Chatbots instances, rather than paying an actual human therapist(hint: High college tuition). If something is free to consume, then you the consumer are the product that is being sold.

    Plagiarism, hallucinations are few among many issues that plague GenAI. OpenAI and Google are working on these with attributions and grounding respectively, to mitigate these unintended consequences. Attribution only goes so far about fixing issues related to paid content. In a subscription economy, each time a content is used by a unique user, then it should result in revenue, compared to earlier one time purchase and use it however you want.

    In conclusion, LLM and Transformer architecture have opened up a million ways to help evolve our species to a bright future, but also opened a can of mutated worms that needs to be dealt with. But that is history repeating itself with the likes of Industrialization, Internet and Web 2.0. I consider Web 3.0 a small tribal evolution compared to Web 2.0, a revolution in the way we generate and share data(information).

AI and the invented conundrum

Recently I read this article The Metamorphosis that talks about AI (Artificial Intelligence), its positive & negative impact on our society(which is already in motion), as well as the invented social conundrums that goes along with it.

AI has been in use for the past few decades in various forms. Either it's voice recognition on a phone to guide you through a pre-fabricated menu or auto correction on your preferred word processor. It started with automating mundane, repetitive tasks to free up the human intelligence to tasks that require additional attention or were difficult to automate. Human touch was required to tasks, that were either prone to errors on automation or the return on investment (ROI) was not sufficient enough to garner attention. It seemed that the AI's requirement of costly processing power was not able to compete with cheap Human brain power.

In the beginning of this decade, few pieces of the AI puzzle started to fall in its place and that propelled  its adoption.

• Cheap CPU cycles
• Cheap Storage
• Cheap Data 

With the advent of Social media, the explosive power of search and moving all our daily activities to a digital platform; they became the source of fire hose of data, that could be channeled and collected on flexible cloud infrastructure with cheap CPU and Storage. This formed the underlying basis of big data and the analysis of this data for patterns and anomalies led to breakthrough in solving various problems. This includes and not limited to targeted advertisements, credit card fraud, predicting stock sentiments and so forth.

Humans as a species have always tried to evolve forward, either through discovery or by invention. But as apart of that process, there has been collateral damage in the wake of progress. When the Spaniards landed at Central America, they decimated the local population with their diseases, but the world was introduced to this new continent along with its abundance of natural resources and culture.

With AI, progress would be made in various fields(insurance, medicine) and AI systems would be allowed to make decisions, that would affect human lives. The decision could be as simple as; approving an application for health insurance considering the various aspects of the applicant, age, or propensity of the applicant to fall sick often. Or a decision to prescribe a life saving drug(limited resource), considering the propensity of the applicant to be a drug abuser or a negative impact to society. 

Humans already have similar decisions on things like organ transplant, but we are at peace with this decision, since we know it was an informed decision by a respectable quorum of doctors. Would we be fine with this,  if that decision was left to an AI. What if, there was a bug in the algorithm that skewed the decision making process. AI algorithms and the data fed to these algorithms are generated by humans and humans by nature are biased. Be it along lines of race, religion or sexual orientation, but implicit bias is an issue that would have to be dealt with. We deal with this, by explaining the process used to achieve the result. Would an AI be able to rationalize its decision making process in human terms?

The ethical veracity of these AI decisions needs to be dissected before we let these systems rule all aspects of our lives.

In conclusion, AI systems will open up a plethora of opportunities from driverless cars to mining meteors for minerals. It also opens up a pandora's box, along the lines of social inequality and racial injustice. As humans, we have to pay attention to details and correct the system as required in an unbiased manner to continue to evolve as an intelligent species.