Artificial Intelligence – The madness of crowds

Generative AI has bubble written all over it.

• The signs are everywhere from ridiculous valuations to outlandish use cases and every man and his dog now wanting to build one, but the technology has no chance of living up to the hype meaning that sooner or later the house of cards will fall.
• That does not mean that there is not money to be made or advantage to be won, but it needs to be executed from the point of view that the bonanza is temporary.
• This kind of excitement, hype and speculation is nothing new but is in fact a manifestation of a well-known and documented phenomenon referred to as the madness of crowds.
• In his 1841 book on the subject (see here), Charles MacKay describes this as crowd psychology that can create an emotional feedback loop whereby dissent may be stifled as the crowd, not wanting to miss out, hears only what they want.
• The first well-documented case occurred in The Netherlands between 1634 and 1637 and was concerned with speculation in the price of tulip bulbs.
• Another was the South Sea Bubble in the UK which occurred in 1720 surrounding the South Sea Company which was granted a monopoly to trade in South America from which unimaginable riches were supposed to flow.
• More recently we had the Internet Bubble of 1999 and 2000 as well as the cryptocurrency craze which came a cropper just last year.
• The circumstances of these bubbles are always different but there are some characteristics that are common to all of them.
• These include popular awareness and interest, investments made with no regard for fundamental analysis or valuation, unrealistic expectations and more recently, every company deciding to use the technology in question for fear of being left behind.
• Andreessen Horowitz’s leading of a $200m round in Character.ai at a $1bn valuation when the company has no revenues and no product is just the latest example and follows Google’s $300m investment in Anthropic which could easily be nothing more than vapour.
• ai is a generative ai company that is just 6 months old which joins the quickly expanding field of new companies popping up to soak up the free money on offer.
• This is exactly what happened in the last AI craze about 5 years ago and led to RFM’s categorization of most AI companies as tricksters who don’t use AI at all but say they do in order to raise money at a better valuation.
• The problem with all of this excitement is that it is setting expectations for the capabilities of generative AI that rigorous scientific testing confirms are unrealistic.
• Generative AI is exceptionally good at giving the impression that it is aware which is leading to the conclusion that general artificial intelligence is just around the corner.
• In simple terms, general artificial intelligence is the ability of a machine to take what it has learned from one task and apply it to another which requires causal understanding of the task to be present.
• All of the scientific literature demonstrates that AIs that are created using neural networks of any size have no causal understanding of what it is that they do.
• Instead, what they do is recognise statistical patterns in data or in the case of large language models, calculate the probabilities of words occurring next to one another in a sentence.
• This is why all of the chatbots around today make the most horrendous mistakes and have no concept of what is acceptable and what is not.
• This is also why they need human overseers to ensure that they do not go off the rails.
• This weakness is inherent in all of these systems that are powered by large neural networks which is what drives my opinion that the route to general artificial intelligence does not lie down this route.
• This means that general AI is decades away which is not the message that is being pushed out through the media and on social media.
• Consequently, when general AI fails to meet expectations (as it has been doing pretty consistently for the last 60 years), the money will dry up and the valuations will collapse.
• The result will be the 4^th AI Winter which will look very similar to what crypto is currently experiencing and what the Internet experienced between 2001 and 2007.
• This gloomy assessment does not preclude one from making money, but one needs to look at this from the point of view that investment or sales strategies need to be temporary in nature.
• These generative AI models consume vast resources meaning that those that supply these resources look set to receive a good portion of the money that is being thrown at the sector right now.
• This includes Nvidia and AMD who supply the chips that all of these models are trained on and further afield, the likes of Qualcomm, MediaTek and yes, even Intel.
• Qualcomm provided a very able demonstration of a 1bn parameter model running on the device (see here) and is now in a position to use this as a selling point of its latest Snapdragon 8 Gen 2 chipset regardless of whether anyone makes use of it.
• Intel, despite its issues, is also a major supplier of silicon to the data centre and this latest craze could trigger a pick-up in its fortunes albeit for a short period of time.
• This kind of madness is enough to drive Nvidia and AMD well past their recent highs implying that there is short-term upside despite the obvious valuation issues.
• This is one for traders and speculators, not investors who may be better served by picking through the crypto wreckage as all the silly money is pulling out and going into generative AI instead.
• The currencies remain extremely risky but there are proper use cases for the blockchain technology once it has been fixed and the assets are now properly on sale.
• The 4^th AI winter beckons.