Distinguished computer scientists demystify GenAI models showing that they are based on mathematics and a methodology of capturing and storing different kinds of information. Notwithstanding all the scaremongering, what has been fantasised by science fiction about intelligence other than human remains just that: fiction. The real threat continues to be from human beings misusing the confluence of the commercialization of the internet, wireless technology, electronics & sensors, software systems which has accelerated what was always inherent in technology per se: the power to manipulate.
This is not written by a contemporary Luddite.
Anyone stating that the term ‘AI’ is being misused is certain to run the risk of being considered irrelevant but when standard industrial automation which dates back to the 1980s (robots in assembly lines) is being projected as AI, something is seriously wrong. Television companies in Japan were among the earliest (if not the first) to use robots to assemble TV sets. Since the robots were programmed to repeat the set of functions in exactly the same way for every TV set, the result is consistent assembling, not possible with human beings. Each such robot was then expected to replace up to 6 workers, raising the threat of significant levels of unemployment. Japan’s subsequent economic history doesn’t bear this out.
There is no AI
A profound plea from a distinguished computer scientist Jaron Lanier, Prime Unifying Scientist, Microsoft, and a study by a team of eight computer scientist-researchers, show that it is simple mathematics and not some mythical AI which underlies GenAI tools such as chatGPT. Lanier has argued that it is misleading to use the term AI in ‘There is no AI’ (https://www.newyorker.com/science/annals-of-artificial-intelligence/there-is-no-ai), and has requested fellow computer scientists to refrain from using it. I have been arguing how large language models like ChatGPT combine predictable use of language and mathematics to produce consistent results.
The fundamental point to keep in mind is that a computer system can only deal with labelled occurrences and hence the way labelling is done becomes crucial. As Lanier says, “a large language model like GPT-4 contains a cumulative record of how particular words coincide in the vast amounts of text that the program has processed. This gargantuan tabulation causes the system to intrinsically approximate many grammar patterns, along with aspects of what might be called authorial style. When you enter a query consisting of certain words in a certain order, your entry is correlated with what’s in the model; the results can come out a little differently each time, because of the complexity of correlating billions of entries”.
It is all about linear functions
In fact, the larger the universe of words, the better it is for the model because it has a higher probability of finding consistent usage, something that basic statistics teaches us. Dr Nivash Jeevanandam, in ‘AI Insights – unlocking the mystery of large language modelling’, (https://indiaai.gov.in/article/unlocking-the-mystery-how-language-models-retrieve-knowledge), refers to a research paper by eight computer scientists titled ‘Linearity of Relation Decoding in Transformer Language Models’ (https://arxiv.org/pdf/2308.09124.pdf), which observes that LLMs like ChatGPT are based on “a straightforward linear function to retrieve and interpret stored information”.
As Jeevandam explains, “Linear functions are mathematical expressions that represent a direct relationship between two variables without including exponents and with only two variables”. Since the universe of words on the internet is literally limitless, the use of taxonomy (basis of classification) is mandatory to facilitate the retrieval and interpretation process, which means looking for comparable categories of information.
How information is stored
Most of you must now be familiar with neural networks, used to describe LLMs, called transformer models by some, but basically an attempt to compare such models to the human brain which consists of billions of interconnected nodes, or neurons, organized into multiple layers and used to encode and process data. Explaining how such models work, Lanier says: “As a transformer develops expertise, it stores information on a specific subject over numerous levels. If a user inquires about the subject, the model must decode the most relevant fact to respond to the query”.
GenAI models have taken what search engines do to the next level. Google’s famous PageRank algorithm, like others, is based on a certain proprietary taxonomy and indexing, which facilitates faster and relevant results. Competition among GenAI models is effectively a competition among these techniques and modelling.
The future of intelligence
According to Lanier, “the most accurate way to understand what we are building today is as an innovative form of social collaboration. A program like OpenAI’s GPT-4, which can write sentences to order, is something like a version of Wikipedia that includes much more data, mashed together using statistics. Programs that create images to order are something like a version of online image search, but with a system for combining the pictures. In both cases, it’s people who have written the text and furnished the images. The new programs mash up work done by human minds. What’s innovative is that the mashup process has become guided and constrained, so that the results are usable and often striking. This is a significant achievement and worth celebrating—but it can be thought of as illuminating previously hidden concordances between human creations, rather than as the invention of a new mind”. I have quoted at length because it is so lucid and so important.
It is not denigrating to say ‘there is no AI’ but is an emphatic plea to see AI as a tool, not some mythical entity which will, one day, destroy what we know as mankind. Is there risk? Of course. But building a mythology around AI is not going to help anyone except businesses who will cash in on the frenzy and market everything as ‘AI’, inevitably leading to the next bust following Edtech and SaaS. Unfortunately, this is already shaping the debate on regulating AI.
Meanwhile, we have real human risks to worry about. Jeff Horwitz, a Wall Street Journal reporter has written a book ‘Broken Code’, exposing how Facebook (now rechristened Meta) has been manipulating its codes and programs to achieved business goals – https://www.amazon.in/Broken-Code-Inside-Facebook-secrets/dp/1911709038.
I have consistently maintained that we should focus on the dangers from hacking, electronic eavesdropping, remote control of systems, wilful manipulation of systems. If we didn’t, we are writing a program to a future of mishaps which can only hurt the innocent.
