When AI is Just Badly Paid Humans!

In recent years, a number of companies have been caught claiming to use artificial intelligence while in reality, outsourcing this work to humans. The SEC recently settled with two funds who were misleading investors about their use of the technology. While artificial intelligence has been widely used in industry for decades, not all companies have been truthful with their claims of AI breakthroughs.
In today's video we discuss to what extent have big firms been faking AI?

Statistics For The Trading Floor:  https://amzn.to/3eerLA0
Derivatives For The Trading Floor:  https://amzn.to/3cjsyPF
Corporate Finance:  https://amzn.to/3fn3rvC

Patreon Page: https://www.patreon.com/PatrickBoyleOnFinance
Buy Me a Coffee: https://buymeacoffee.com/patrickboyle

Visit our website: www.onfinance.org
Follow Patrick on Twitter Here: https://twitter.com/PatrickEBoyle
Patrick Boyle on YouTube

Transcript

Between 1998 and 1999 147 US listed companies changed their names to include dotcom, dot net or the word internet. Many of these companies’ core businesses were not internet related – it didn’t really matter…

Concerned with all these announcements, the director of the SEC’s investor-education office warned investors not to just invest in a name, saying, “That is asking for losses.”

This April, the Director of the SEC’s Division of Enforcement noted in a conference speech that there was immense investor interest in artificial intelligence and that fake AI, or AI washing has the potential to mislead investors, harm consumers and violates federal securities laws.

FactSet shows that 199 of the S&P500 companies mentioned AI on their first quarter earnings calls. This is the highest number of mentions on record, with the prior record having been set in 2023.

Of course, many of these firms will actually have an AI strategy, many will have been using AI for decades, but it does appear to be the buzzword of 2023 and 2024 – having replaced blockchain.

AI is of course a real thing – and it’s not really all that new. It is however new in things like TV’s, refrigerators, bird feeders, shoes and dog bowls.  I’m not saying that the AI dog bowl is fake – I went to the website and it appears to tell you the temperature and help you weigh out dog food – but if you would buy an AI dog bowl – you really would buy anything. No?

Whenever a new idea gets really hot you can (of course) expect pretenders to jump on the bandwagon.  Earlier this year the SEC announced settled charges against two investment advisers, for making false and misleading statements about their purported use of artificial intelligence. The firms agreed to settle the SEC’s charges and pay $400,000 in total civil penalties – without admitting wrongdoing. 

Gary Gensler said that the two firms “marketed to their clients and prospective clients that they were using AI in certain ways when, in fact, they were not.” He went on to say, “Investment advisers should not mislead the public by saying they are using an AI model when they are not. Such AI washing hurts investors.”

One of the firms it seems claimed that it was “the first investment adviser to convert personal data into a renewable source of investable capital . . . that would allow consumers to invest in the stock market using their personal data.”  They went on to say that they use “machine learning to analyze the collective data shared by their members to make intelligent investment decisions.”

I won’t lie, I’m slightly surprised that they got in so much trouble for statements like that, as essentially that is a collection of words that when combined into a sentence means absolutely nothing.

Let’s look at it again.  They claim to be the first (which may be true – as I haven’t heard anyone else make this claim) - investment advisor (which they do appear to be) to convert personal data into a renewable source of investable capital – personal data into a renewable source of investable capital- that doesn’t mean anything does it?  That’s nonsense. Then they said that they’ll allow consumers to invest in the stock market using their personal data – which could reasonably mean that they allow their customers to invest in the stock market after filling out a form giving their name, address, date of birth and social security number – that sort of thing – which is probably true.  Investment advisors have to get that data from their customers anyhow for tax reasons and as part of the KYC – or know your customer rule – and it doesn’t hurt to know where to mail the statements to either.   Then they said machine learning – blah blah blah – and that’s where they went wrong – as it seems there was no machine, and it didn’t learn anything – and that’s bad… so, don’t do that…

The other firm was accused of falsely claiming to be the “first regulated AI financial advisor” and of misrepresenting that its platform provided “expert AI-driven forecasts.” They also violated the Marketing Rule, falsely claiming that they offered tax-loss harvesting services, and included an impermissible liability hedge clause in their advisory contract, among other securities law violations.

So yeah, you shouldn’t tell your customers that you are using Artificial Intelligence if you are not, just tell them that you are using 100% natural intelligence.  It’s not like regulators are going to turn up at your office with an IQ test and force you to issue a retraction.  You should be OK with that.

AI – as I mentioned earlier is not a new technology.  It was founded as an academic discipline in 1956 and has gone through multiple cycles of optimism and periods of disappointment since then.  It has been used in the financial world for decades, most famously at Renaissance Technologies, but also at most quant funds.  AI based investment strategies have used in the world of finance since long before I started working in the industry.

It’s not just finance either, neural networks and computer aided detection software has been used in medical imaging since the 1980’s and has been used in clinical roles like computerized ECG analysis and arterial blood gas interpretation for quite some time.

The first AI designed drug candidate to enter clinical trials occurred in 2020.

The spam filter that has been on your email for the last twenty years uses AI and YouTubes recommendation algorithm which likely brought you to this video is an AI based system too, and it didn’t suddenly appear last year either.  The recent hype around AI – which has been all around us for years, was sparked by the public release of Chat GPT in November 2022 because chat GPT drew in 100 million monthly active users in under two months, making it the fastest-growing consumer application in history, and fast growth gets VC investors really excited.

A lot of the excitement around Chat GPT was possibly because it appeared to pass the Turing test, one of the best-known methods for assessing AI that grew out of a thought experiment devised by the computer scientist Alan Turing. 

The Turing test pits human respondents against a machine in order to test whether humans can tell if they are conversing with another human or a computer. Turing argued that if a computer could fool people into believing they were conversing with another human rather than a machine, then it could be considered intelligent.  Matthew Jackson, a professor at Stanford University wrote in a paper earlier this year that the most recent version of ChatGPT passes a Turing test, only diverging from average human behavior chiefly to be more cooperative. Essentially GPT4 is an artificial Canadian.

The philosopher John Searle argued (quite correctly in my opinion) that Turing's test is insufficient to detect the presence of consciousness. A computer can be programmed to perform certain parlor tricks – but that does not mean that it has a mind, understanding, or consciousness.

The fact that the Turing test held such a position in the public imagination as the hurdle for true artificial intelligence might be why people are so excited about chatbots but ignored all of the other breakthroughs in the field over the last few decades. 

There are all sorts of ridiculous devices being sold as AI products like the Rabbit R1, a handheld AI device that sold out its first production run in just one day. Investigators like Coffeezilla found that it was not using a new foundational AI model as claimed but instead Chat GPT mixed in with some hardcoded scripts. [Clip – Part of rabbits code says I will never mention that I am a large language model created by Open AI]

There was also the Humane AI pin, that was supposed to do similar things and was just awful. [MKBHD Clip]  Over the last year and a half we’ve have seen AI companies faking product demos – in the same way website demos were faked 25 years ago during the dot com bubble. It’s only so surprising then, that a recent study found that tacking an AI label on products like TV’s and refrigerators lowers the average customer's willingness to buy it.

There are some great examples of fake AI products. Bloomberg wrote in 2016 about the workers who spent twelve hours a day pretending to be chatbots for a calendar scheduling service called X.ai (not the Elon Musk X.ai another one – seems it is a common name.) The workers at Xai described the job to Bloomberg as being so awful that they were looking forward to eventually being replaced by bots.

A London based startup which recently claimed to use AI to read through images of your receipts digitizing them and storing them on an app was recently accused of outsourcing the work to a virtual data extraction team to manually read the receipts and enter the data.

Similarly, in 2017, a business expense management app Expensify admitted that it had been using humans to transcribe receipts it claimed were being processed using AI. Scans of the receipts were apparently being posted to Amazon’s Mechanical Turk crowdsourced task completion tool, where low-paid workers were doing the actual work.

Now the name of Amazon’s mechanical Turk website has an entertaining origin, there is a good book on it which I ‘ll link to in the description. The original mechanical Turk was a fraudulent chess-playing automaton built in 1770, which seemed to be able to play a strong game of chess against human players. It was brought all around the world by its owner, playing against people like Napoleon and Benjamin Franklin. Its owner would open it up to display the complicated clockwork mechanism inside, but it was later discovered to have a human chess master hiding inside, working the machine.

Amusingly Amazon had a second Mechanical Turk, its Go - cashierless stores – which were branded as Amazon Fresh in the UK. They used Amazon’s “just walk out technology” which they said used computer vision, deep learning algorithms, and sensor fusion which meant that customers could select items from the shelves – and without ringing anything up they could “just walk out” and would see the items ring up in their Amazon account.

Last year Amazon began closing some of the stores, and this April The Information reported that the technology, had partially relied on more than 1,000 people in India who were watching camera footage and labeling videos because the underlying technology just didn’t work.

Instead of AI taking peoples jobs – it just outsourced them to India…

A friend of mine who is an engineer pointed out a while ago that when you see humanoid robots – which have had a recent resurgence - making human like gestures, such as turning their heads to see something, you should instantly be skeptical, as it is much cheaper and more efficient to put a circular array of cameras, or a 360 degree camera in the robot than it is to install all of the motors needed to turn the robots head.  Human-like actions are just there to impress investors, they are not there for the purposes of functionality.

Robotics experts mostly agree that humanoid – or animal shaped robots make no sense because biomimicry just isn’t the right approach for any sort of industrial robot.  Possibly the funniest example of this is in a photo a friend sent me from an empty office building where the landlord is trying to attract high tech tenants. The image in the lobby of the building shows robots working alongside humans in a modern looking office, with a robot sitting at a desk typing on a computer keyboard.  Why would a robot ever use a keyboard?  This is one computer connecting to another computer using the most inefficient interface imaginable. It’s not well thought out…

Factories are of course filled with robots that can lift heavy parts, weld, sew things together, tighten bolts and so on – but they don’t have to be human shaped.  You wouldn’t design a sewing machine to look like a person holding a needle and thread, so why would you design a factory robot to look like a person?

I worry that if the people building these humanoid robots had been tasked with building a car 140 years ago, instead of building an engine connected directly to the wheels, they would have tried to build a mechanical horse to pull a cart.

As I mentioned earlier there is a huge increase in the number of companies mentioning AI on their earnings calls.  According to Brownstone research, the company who mentioned AI the most was intel, where the CEO mentioned AI more than thirty times on their fourth quarter 2023 call.

Despite all of the talk of AI, Intel fell behind their competitors because – according to Reuters “for more than two decades, they believed the CPU could more effectively handle the processing tasks required to build and run AI models which left them lagging behind their competitors in building GPU’s.  Talking a lot about AI does not necessarily mean that a company is at the cutting edge of AI.

I’m not sure what to make of all the tech CEO’s with their sci-fi claims that we are on the cusp of developing Artificial General Intelligence which will destroy us all.  They sign letters saying that AI research should be halted for safety reasons while rushing to build their own models that break all of the rules that they claim should be followed.  I can’t help but wonder if they feel that claiming that the technology is dangerous will make investors believe that the technology is much more advanced than it actually is, which might drive up their stock prices and executive compensation.

Big breakthroughs in AI and quantum computing can be expected to have both pros and cons. A sudden breakthrough in computing power might mean that all of the encryption tools that we use today, can be easily broken, but that has always been the nature of technological advancement, where new things are better than old things, but smart people work out solutions to these new problems and the world slowly gets better over time.  Due to modern technology, a middle class American today has access to comforts, education and healthcare that the wealthiest man in the world didn’t have a hundred years ago.

There is a very good New Yorker article from last year written by the computer scientist Jaron Lanier who points out that people wrote all of the code used in generative AI models, and people wrote the text and created the images that the models are trained on. The new programs mash up this work and the results are surprising and often striking. The non repeating nature of these creations can make the software feel alive but while this is a significant achievement and worth celebrating—it should be thought of as illuminating previously hidden concordances between human creations, rather than the invention of a new mind. 

He talks in the article about how much better it can be when computer interfaces become less rigid – using natural language prompts for example. We have gotten used to software that requires us to conform to it – for example, forms that won’t let you hit submit if you haven’t filled them out the way the code requires. This requirement for humans to conform to the needs of software creates a feeling of human subservience to computers. The way these new AI tools work, means that we can imagine websites that reformulate themselves on the fly, tailoring themselves to a user’s particular cognitive abilities and styles. He argues that this flexibility that AI provides, might give us back more agency over these tools – a very different vision to the matrix or terminator future that other tech visionaries seem to expect.

At present the technologies that are most hyped are shockingly expensive. The FT recently described generative AI as the biggest, and the fastest infrastructure rollout in history, which leaves us with the big question of who will eventually benefit the most from all of this spending and when will the returns on investment be realized.

Infrastructure plays like Nvidia are often the early winners when a new technology is being rolled out but so far no company has yet created a “killer app” for generative AI – or at least one that people will pay money for - despite having sold that dream to investors.  While Nvidia is riding high right now – the infrastructure plays of the early internet like Cisco, EMC, Corning and JDS Uniphase didn’t turn into great long-term investments if you invested during the hype phase. 

The big tech firms who are pumping money into AI are all profitable businesses, while you might think of Facebook as a social network, google as a search engine and Amazon as an online retailer, they are all mostly in the advertising business.  They make their money selling advertising and are then pumping it into an AI moonshot hoping that they will find a way of turning a profit out of it, but at present we don’t know where those profits will come from.  They haven’t explained that yet.

Huge tech spending in the past has sometimes, but not always worked out. Investors are still waiting to see returns on the investment in VR headsets, the metaverse and blockchain.

I started out telling you about the SEC warning investors during the dot com bubble to not just invest in a company because they add dot com to their name.  Well, it turns out the SEC was wrong about that.  A 2001 paper in the Journal of Finance – called a rose dot com by any other name - found that while name changes generally don’t affect the value of a company, there had been dramatic increases in both stock price and trading volume for the companies who took on internet related names during that two year period.  The professors found that the companies that changed their names rose an average of 53 percent over the five days after the announcement date.  Over fifteen business days, the stock prices rose by – and I’m not making this up – four dollars and twenty cents per share. They found that the returns were similar across all firms, regardless of the company’s actual involvement with the Internet.

The professors checked if the price changes could be explained by the small company effect, by the beta of the stocks or by a momentum effect.  They compared the returns on these stocks to the returns of real internet companies who didn’t have internet related names – in case the returns were just sector bias. Their results were robust. So maybe in an AI bubble, fake AI companies will do just as well as real AI companies – maybe Zuckerberg should quickly rename Meta while he still has a chance – the old name isn’t doing him any favors right now.

By September 2000, the New York Times reported that companies were dropping the E’s the I’s and the dot coms from their names – times had changed. No one wanted to be a dot com in the early 2000’s.

Thanks for tuning in to this week’s podcast – which is entirely supported by donations from listeners like you on Patreon.  There is a link to that in the show notes if you are interested.  Have a great week, and talk to you again soon, bye.