MxMIndia

Tag: Large Language Models

  • The resultant hype around ChatGPT has kickstarted the age of Artificial Intelligence, writes Ashoke Agarrwal

    ML, LLM, Graphs & Market Modelling

     

     

    By Ashoke Agarrwal

     

    Ashoke AgarrwalNine months after the launch of ChatGPT, the hype has died down and the real work of building upon the burgeoning availability of Large Language Models (LLMs), of which ChatGPt is but one example, has begun.

     

    The interest of businesses in the concept of Big Data rose exponentially in 2012. Many expected a paradigm shift in consumer marketing based on nifty analytical systems driven by Big Data. Much was expected in data-driven decision-making, personalisation and customisation, targeted advertising, sharper segmentation, predictive analytics and real-time insights. In 2015, in its report titled ‘Big Data, Analytics and the Future of Marketing and Sales’,  McKinsey laid out the expectations.

     

    In 2023, the future is different from what McKinsey expected. The paradigm certainly shifted for Google and Meta, who cornered the advertising market based on a humongous amount of real-time Big data and state-of-the-art analytical engines.

     

    Change also happened for brands that marketed and sold products and services to B2C and B2B markets. However, the difference had little to do with their use of Big Data and advanced analytical systems. It was in the emergence of the digital universe as a product, go-to-market and communication platform.

     

    Dig a little, and you will find that the thesis that brands under-exploited the opportunity that Big Data presented to them mainly because they used legacy databases and ERP systems from the likes of SAP, Oracle and Salesforces to collect and store their Big Data. As a result, while many created special teams to mine, warehouse and analyse Big Data, most crucial business, marketing and sales decisions continued to be based on traditional business analysis and market research.

     

    Meanwhile, Google and Meta (then Facebook) invested in an analytical system that eschewed the rigidity of the traditional IT-age relational databases – essentially tables with rigid rows and columns. In a seminal decision that presaged the age of AI, they decided to populate their databases in graphs – a network of nodes with many properties with multiple links to other nodes, with each link specifying a kind of relationship with a property of the originating node and a property of the linking node.

     

    Such a database structure allowed for:

    :: Fast and more varied analysis

    :: More immediate additions and reconfiguration of the database as conditions change

    :: Better situational analysis and insight discovery through “what-if” analysis that changed node and link configurations

     

    Based on their ever-expanding, ever-enriched graphs and the use of Machine Learning (ML) driven near-real-time analysis, Google and Meta created a mighty advertising service with the confidence to offer brands a pay-per-click service. As a result, brands willy-nilly outsourced the harnessing of the real opportunity of the digital and Big Data age to Google, Meta and other digital ad exchanges that sprang up.

     

    It is no surprise that Google and Facebook are among the most advanced AI players today, including in the field of LLMs with Google’s Bard and Facebook’s Llama. The essential process that powers LLMs is that they parse large storehouses of text into triples of subjects, objects and predicates and then model them into graphs with the nodes consisting of subjects and objects and the links signifying relationships in the form of predicates.

     

    As Google’s and Meta’s LLMs improved power, it turbocharged their graph databases, allowing them to automate the process of incorporating unstructured into their graphs. Perhaps left to themselves, Google and Meta would not have exposed their LLMs to the public as they have done so now but kept it themselves as a background technology powering customer-facing applications. Instead, OpenAI and ChatGPT forced their hand.

     

    The resultant hype around ChatGPT has kickstarted the age of AI, with the world at large now seriously scrambling to incorporate AI into businesses, Governments, schools, hospitals and wars. Publications like The Economist and others have started ranking Fortune 500 companies based on the potential competitive advantages/disadvantages that AI can deliver to them!

     

    How will businesses in general and marketers in particular respond to the new horizons of AI?

     

    The powers of MI, Graph Theory and Advanced Modelling will create a new platform that will change how businesses, if not entire societies, are run. Just as the arrival of digital media changed economies at the core, Graphs with Big Data inputs from structured and unstructured sources (processed through LLMs) will create dynamic market models that will change how businesses are structured, with business and marketing strategy being almost wholly automated with human inputs needed only at the highest meta-strategy level. The shift will be paradigmatic enough for the world to label the resultant business order as the Nth Industrial Revolution, a sub-set of the First AI Revolution that will redefine human society.

     

    The question is whether individual businesses will seize this opportunity, invest and build proprietary dynamic models that will run their companies, or will they, once again, as they did with the digital revolution, outsource it to the next Google or Meta?

     

    The stakes this time are higher. Not only will the businesses themselves be more beholden to those who own and run the models, but in the process, society will create AI-driven behemoths that will be the first step to the dysfunctional system that those who fear AI imagine.

     

    Therefore, it is incumbent on business leaders and thinkers to pay close attention to this evolving opportunity and invest all that is needed to harness it before it becomes an insurmountable challenge.

     

  • People whose business is to peek into the future approach Artificial Intelligence from two very different angles, writes Ashoke Agarrwal

    The Economic Promise & Cultural Peril of AI

     

     

    By Ashoke Agarrwal

     

    Ashoke AgarrwalArtificial Intelligence (AI0 is fast becoming the general-purpose technology that will determine humankind’s future.

    People whose business is to peek into the future approach it from two very different angles.

    Some hard-headed economist types see AI mainly as a disruptor of the world of business and economies.

    Others who study broader and deeper societal trends prognosticate the possible long-term effects of AI on human civilisation.

    Neither school sees AI developing into a threat on the lines of the Terminator-type robot shooting down people in the streets or a Skynet-type all-powerful entity trapping humans in a virtual matrix.

    The book “Power and Prediction: The Disruptive Economics of Artifical Intellgence.” by Ajay Agrawal, Joshua Gans and Avi Goldfarb, a 2022 follow-up to their 2018 book “The Prediction Machine: The Simple Economics of Artifical Inteliigence.” lays out the disruptive but possibly ultimately enhancing effect of AI on the world economy.

    The broader view of the impact of AI on human civilisation comes from Yuval Noah Harari, the historian-philosopher whose three books “Sapiens: A Brief History of Mankind.”, “Homo Deus. A Brief History of Tomorrow” and “21 Lessons for the 21st Century” introduced a deeply thought out yet lucid and vivid view of the factors that governed the evolution of human civilization.

    Harari has spoken at length about his views on AI at various forums. Recently he did a three-hour sit-down with Lex Friedman. Here is a YouTube link to the interview and a transcript. Harari’s views are grounded in his unique approach to the evolution of human civilization and startling in their clarity and scope. It also offers an almost sly but plausible take on the threat that AI poses to human society without going into Terminator and Skynet kind of fevered speculation.

    In their 2018 book “The Prediction Machines”, Agrawal et al. posited that AI at its core was a quantum leap in the science of prediction. Until the emergence of Deep Learning, prediction methods mainly used the science of statistics with tools like multivariate regression. With Deep Learning and its offshoots, predictions became progressively more accurate and cheaper. Agarwal et al. posited that technology finds more widespread use when it becomes more affordable. They offered the instance of electricity and computers. One of the vivid examples they offered about how better predictions could lead to changes in business models was of e-commerce players like Amazon shifting from a “shop-than-ship” model to a “ship-than-shop” model once they had the AI tools that predicted with reliable accuracy what their customers would buy next – that is they would ship the predicted product off to a consumer even before he had shopped for it on their site. In support of this insight, they cite that Amazon had filed for a patent for “anticipatory shipping”.

    In their 2022 book “Power and Prediction.” Agrawal et al. revise their view of the economic future of AI. They posit that the widespread adoption of AI will not happen with point solutions like replacing processes where traditional forecasting is currently the norm with AI-based forecasting. Instead, it will compel economies and businesses to go beyond and identify areas where AI-based prediction enables them to switch to decision-based procedures that optimize resources instead of rules-based processes that compromise efficiency in the face of uncertainty.

    Also, because AI-based predictions will have system-wide ramifications, the optimal adoption will happen when economies and businesses redesign entire systems to accommodate AI. Agarwal et al. identify two design approaches that can drive systemic changes: coordination and modularity. Their book details these approaches and illustrates them with examples from the health, transport and e-commerce sectors. The overall message from Agrawal et al. is that AI and its economy-wide adoption will be systemic and disruptive. And overall, its impact will be positive, like the widespread adoption of the last two general technologies – electricity and computers.

    Mr Harari’s views on the civilisational impact of AI are nuanced.

    Harari’ has been surprised by the pace of development of Large Language Models (LLMs) and their rapid penetration into the social and cultural life of human societies.

    At one level, he sees the threat posed by LLMs as a ratcheting up of the threat posed by social media. The design of social media algorithms captures attention and, in the process, creates echo chambers that fuel conspiracies and tribalism. AI entities based on ever-improving LLMs will capture intimacies. If unchecked, they could monopolise an individual’s personal space, weakening and destroying individual relationships and thus weakening the concept of family and friends and hence the very social framework undergirding human society.

    Harari perceives another more subtle threat. Harari hypothesizes, as explained in his books that the life of individuals, societies, cultures and civilizations is circumscribed by stories and myths that are creations of the human imagination. God, religion, nation, money etc., are all myths that have taken deep root and driven society in all its pursuits – politics, economics, art and culture.

    While Agarwal et al. perceives AI as a disruptive “Prediction Machine”, Harari rotates the prism and perceives AI as a threatening “Culture Machine”. He sees AI ( and sometimes he calls it Alient Intelligence) as “eating” and “digesting” all human culture to come to a stage where it can give back images, words, art and stories that are more compelling than any that humans can process. Because these cultural artefacts govern human evolution, this “Alien Intelligence” will take charge of it. Here in his own words, is how he perceives this threat:

    “...But taking what we do know about human history until now, all the, again, stories, images, paintings, songs, operas, theater, everything we’ve encountered and shaped our minds was created by humans. Now, increasingly, we live in a world where more and more of these cultural artifacts will be coming from an alien intelligence. Very quickly we might reach a point when most of the stories, images, songs, TV shows, whatever are created by an alien intelligence. And if we now find ourselves inside this kind of world of illusions created by an alien intelligence that we don’t understand, but it understands us, this is a kind of spiritual enslavement that we won’t be able to break out of because it understands us. It understands how to manipulate us, but we don’t understand what is behind this screen of stories and images and songs.”

    That is a more alarming picture of the AI-age world than any Terminator or Skynet kind of scenario. It is more disturbing because the process is sneaky and sly, and one can see the beginning of it even at the early stages of the LLM revolution.

    While the forces of commerce and the market will ensure that economies reap, with time and effort, the benefits of the “Prediction Machines”, what remedy do we have against the threat of the LLM-based “Culture Machine.”? Harari has a challenging remedy to offer. Harari believes that we humans do not fully understand ourselves. He suggests that for every dollar and hour we spend developing the AI-based culture machine, we also invest a dollar and hour in understanding ourselves better – perceiving the contours of conscious daily reality that exist in our feelings beyond the stories and the myths that confuse and control us. Is that a realistic goal? Will the story of progress that drives our notions of work and happiness allow us to set and accomplish such a goal? Let me put this question to ChatGPT and Bard and see what they say.