top of page
Search

Post #19 - Different AI Visions

guyrushton9
Jul 29
4 min read


Reading around the hype about AI and trying to understand the promises, good and bad, that are being made about the technology.



I have had a little time off recently and this has given me the chance to catch up on the mountain of improving books that I keep buying. Given the current hype about AI, I thought I ought to get with the program and understand some of the arguments about it.


It is very easy to be seduced by the enthusiasm. My own very limited experience of using products like CoPilot suggests that it has promise in some areas but that it is fragile, and in some cases downright unhelpful. A case to proceed cautiously and to conduct a rigorous cost-benefit analysis before adopting across major business systems. That was my starting point when I read these two books.

 


AI Snake Oil

Arvind Narayanan & Sayash Kapoor


The pitch for this book is that it encourages healthy scepticism, from people actively and positively engaged with AI research. Well written, and clear, it gives some of the history of AI and explains the basics of the technology and its inherent limitations. It reminds the reader that AI has been around for a long time in various forms and has both had triumphs, and over-promised in the past.


It shows that while some of the applications are working reliably and have been for a long time, such as image recognition and machine learning, others are not living up to the billing. LLMs, while undeniably impressive, are not sentient, they simply match the next most likely token, and that matching process may not be perfect. Their abilities are baked in by their training; they perform very poorly at prediction of complex systems or human interactions. Applying AI to predict a chaotic system is likely doomed to failure, since perfect information is required as a baseline. This makes the desire to use AI to predict behaviour and control societies, hopefully, just a pipe dream.



Genesis

Henry Kissinger, Craig Mundie & Eric Schmidt


The authors of this book are undeniably well informed about AI but are definitely in the evangelist camp. The early part of the book is about the untold benefits of AI and its inevitable progress through our society. They acknowledge the potentially catastrophic effects on society of its adoption if their vision is correct and offer no solutions to those issues. The later chapters postulate that an artificial super intelligence might be used to rule our societies, logically. Bring order and remove human weakness from the business of politics, law and international relations.


I find this vision genuinely chilling in many ways. They have clearly imbibed a lot of science fiction, though possibly Dr Kissinger’s life experiences lead him in this direction in any case. This technological utopia I can only see as a dehumanised dystopia. It would enshrine a priestly class, them and their like, that would service the oracle machine, whose pronouncements would be as unanswerable as the word of God. Unexplained, impossible of analysis, demanding total obedience. It is a vision that, I think, is arrogant and inhuman. Resistance to their vision would not be tolerated, we, the little people, are of no account. The ancient Greeks had a word for this: hubris.

 

Having read a good deal around the subject recently, both on the very basics of the technology and on the potential financial and societal impacts, I am erring on the side of the cautious optimism of the Snake Oil authors. Some of this technology is already working and has been for a long time, without destroying people’s lives. LLM based products will exist in the future and they will be a part of our working lives, they will improve.


AI generated slop, videos and text, might well become unacceptable, or simply unfashionable. Given the huge amounts that have already been produced and the likelihood that AI slop is being used to train the next generation of LLMs it could become self-limiting.


I suspect that the predictions for all conquering LLM based models will not come to pass. Unless we take leave of our senses we will not have ChatGPT21 deciding whether or not we go to war. More prosaically the costs are rising fast and the results are mixed. The enormous investments in trying to be build ever bigger LLMs to try and create an unanswerable pseudo-deity are going to end in appalling losses. Profitability seems to be an ever-receding mirage. It is possible, due to the way that the money has been raised, that these losses will generate another 2008 style financial crisis as lenders find out their investments are worth nothing and the real economy freezes. This will hurt a great many people who have nothing to do with the AI industry. The worst possible outcome will be that the regular citizens, whose jobs and ways of life have been so breezily discounted by the Silicon Valley aristocracy, will be forced to bail out these institutions.



Both of these books are worth your time, please comment on your take on the state of AI and its potential effects.

Comments

bottom of page