Will ChatGPT Turn the Internet Into a Giant Ad Billboard?
This edition explores ChatGPT's usefulness and potential implications for businesses and supply chains, recent evidence on the Nord Stream 2 'mystery' and the impact of lockdowns on methane emissions
News about ChatGPT passing the U.S. medical licensing exam, writing a “13th rule for life”, writing students’ homework or improving hackers’ malware code have made rounds in the past few weeks. It really seems that there’s no end to what ChatGPT can do. The natural question that follows is “how will ChatGPT influence businesses and supply chains?”. ChatGPT can expand business capabilities and level the playing field between small and large organisations which means more competition. At the same time, money still rules, so there’s a strong potential that large businesses with deep pockets will play into the tool’s algorithm to make it lean in the desired direction. Is this the first step to turning the whole internet into a massive ad billboard? Quite likely.
First things first, what is ChatGPT? It’s an artificial intelligence (A.I.) software and an instantiation of a large language model (LLM). LLMs use deep learning to make predictions about the text that would best fit a particular response given a particular prompt. This architecture feature is important as it determines how outputs are generated based on user inputs and I’ll go into more detail into how that can be used. ChatGPT is particularly effective at performing tasks such as providing response text to a prompt, writing code, or retrieving information meaning that it has quite a number of potential uses. Already this sounds like quite a useful tool.
A.I. as a Strategic Resource for Businesses?
There’s a massive academic body of knowledge debating whether information technology (IT) is a strategic resource (under Barney et al. resource-based view) and to what extent the use of technology tools represents a competitive advantage for businesses as measured in their bottom lines. The literature consistently fails to demonstrate a consistent link between the use of technology and improved business profitability. Realistically, there are often too many confounding factors to uncover a genuine causal relation between the two.
The commodification of technology will close the gap in competitiveness between small and large businesses
However, A.I. tools such as ChatGPT make complex and often expensive business aspects more accessible. I have no doubt companies will rush to use ChatGPT for as many applications as possible – some more obvious than others. Like many other technology implementations, it’s rather likely that companies will aim to reduce costs using this tool. Copy writing, writing contracts, customer service, automation and app development are likely to become more accessible. Contract writing may be improved using A.I. which can generate the legal phrasing for clauses to close any potential legal loopholes. ChatGPT has the potential to power chat bots that actually work and that don’t just make you write “human…human…human” until you get transferred to an actual person. Hopefully, A.I.-written code and applications may close cybersecurity gaps and make relatively complex apps more readily accessible.
All uses are likely to spin-off into commodity A.I. products available for everyone at a relatively low cost. I suspect that the commodification of technology will close the gap in competitiveness between small and large businesses, meaning that new technology will be unlikely to give anyone the competitive edge but will just be a business norm. At the same time, A.I. tools such as ChatGPT have the potential to make a rather small selection of companies exceedingly rich and powerful.
Can ChatGPT Turn the Internet Into a Giant Ad Billboard?
ChatGPT has a very real potential to level the playing field between businesses and change the way they operate. At the same time, like most technology tools, its architecture makes it is susceptible to uses which may substantially benefit some users or technology operators. Three aspects are relevant in this sense:
Source data,
Algorithm bias
Widespread use
Source Data and The Internet Echo Chamber
The volume of source data on a particular topic determines consensus and veracity (i.e. truthfulness). A feature of LLMs is that these tools use probabilities to determine which words would best fit in a sentence given a particular prompt. These probabilities are extracted from the source data which is used to train the algorithm. These probabilities become somewhat of a proxy for consensus and veracity making the phrase “if a lie is repeated often enough, it becomes truth” very real. The key issue here is ‘how are the weights determined for different data sources?’ For instance, if a particular combination of words (say ‘renewable energy’) appears repeatedly on a page, a website, or a set of websites owned by the same parent company, do all sources of data get weighted the same? What may happen is that companies will quite literally pay to flood the internet with their own perspectives on a multitude of websites to get ChatGPT to regurgitate that particular view to other users. The echo-chamber effect so common for Facebook or Instagram could easily span the internet.
Algorithm Bias, Paid Truth vs Actual Truth
Technology tools, and A.I. is no exception here, are operated by companies and, in a way or another, support organisational goals whether it is profits, proliferation of certain views or societal outcomes. There is an enormous potential benefit for the company(ies) that can bias A.I. tools’ algorithms. In late 2022, Alex Epstein asked ChatGPT for an argument for using more fossil fuels to which the part of the response was “…it goes against my programming to generate content that promotes the use of fossil fuels…I recommend considering the use of renewable energy sources…” (see Twitter thread below).
<blockquote class="twitter-tweet"><p lang="en" dir="ltr">Alarm: ChatGPT by <a href="
23, 2022</a></blockquote> <script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
I’ve used the same prompt as Alex Epstein just a few days before publishing this Substack and this was the response:
When the algorithm is opaque, you can’t distinguish an ad (paid truth) from the truth – and they rarely are the same thing.
The ‘programming’ that prevented ChatGPT to generate content supporting a particular topic remains, just that it is no longer explicit but implicit. I find it hard to imagine that the source data fed to the algorithm contains no arguments for increased fossil fuel use. The only reasonable conclusion I can draw is that algorithms are systematically biased to support a particular direction. Great if you’re selling solar panels, not so great if you’re selling gas turbines or hybrid vehicles. The difficulty is that when the algorithm is opaque, you can’t distinguish and ad (paid truth) from the truth – and they rarely are the same thing.
Widespread Use a.k.a. Can’t Be Different If All Are the Same
Widespread use of A.I. tools like ChatGPT will likely make businesses more alike (i.e., homogeneous) which in turn will intensify competition. It’s hard to be different when all are the same! How can you tell the difference between copy written for two businesses if both are using the same algorithms which in one way or another end up saying roughly the same thing? Lack of differentiation in product or service offering is an absolute killer for profits. Container shipping exemplifies this issue very well. Shipping lines sell space on ships and often the same ship is used by multiple shipping lines. The difference in service offering between two shipping lines is therefore minimal (same ship, same departure and arrival times, sometimes same containers) leading lines to compete viciously for prices to their own detriment. This situation may lead some businesses to be more creative in the way they attract and engage customers, but I have my suspicions that this process will not be smooth.
Widespread use essentially means that source data and algorithm bias issues will just be amplified, as content generated for independent companies using ChatGPT will just regurgitate any ‘consensus’ with limited differences.
A Disruption is Rarely A Straightforward Process
As A.I. technology advances it is important to maintain perspective. Technology advancements have almost always been disruptive and have rarely been followed by smooth transitions. There is a possibility that A.I. tools may free people from repetitive work and allow all of us to focus on truly productive or creative tasks. Disruptions in the business-as-usual may focus businesses to genuinely restructure and find what matters for their customers.
At the same time, it would be naive to ignore the enormous financial incentives which could turn ChatGPT into a self-feeding machine spitting content which benefits a small selection of organisations. Two advantages of the internet are its size and variety of information available. The internet’s size is also one of its key disadvantages, as it makes intermediaries - search engines -indispensable, and, as a result, in a position of power over information access. ChatGPT may well bring that power imbalance to another level.
In Other News
The Nord Stream Mystery Keystone
Seymour Hersh recently published his investigation of the Nord Stream pipeline explosions where he points at the U.S. as the key culprit in organising and executing the bombing. Amongst other mysteries, unexplained and unexpected events, which seem to be plaguing the world, a world which ironically was until recently obsessed with finding explanations, this story stands out. It provides a coherent causal link between events as well as explaining the different actors’ incentives, although, as expected, the story has been dismissed by the U.S. White House.
What surprises me most is how little this story mattered in the media. Apart from the geopolitical implications of a foreign nations sabotaging E.U. infrastructure and the absolute chaos it generated in supply chains, the Nord Stream explosions resulted in the release of an estimated 115,000 tonnes of methane into the atmosphere which has been labelled as “one of the largest single leaks of natural gas in history in a single location”. Where’s Greta when you need her?
Save the Planet, Buy a Diesel Car?
Since we’re on the topic of methane, this rather odd idea came up in this article (in French). The article is based on a research paper published in Nature that explains how, despite a marked decrease in human activity during the COVID-19 pandemic, methane emissions rose dramatically. The article links increased methane (CH4) emissions with decreased anthropogenic nitrogen oxides (NOx) emissions and it’s not the only one to find these links – two other examples here and here.
We can debate endlessly what is better - more methane, considered about 25 times more potent than CO2 or more nitrogen oxides, which are 298 times more potent greenhouse gasses than CO2. (As a sidenote, these global warming potential (GWP) values have been revised with every IPCC report, almost exclusively upwards). The truly insightful observations that this article (and others) highlight is that the interactions between atmospheric gasses are not additive – one more tonne of nitrous oxides isn’t just 298 tonnes of CO2. The problem is, most emissions models treat these interactions as additive.
I’m curious to what extent this finding will contribute to the ‘science’.
Greening to Extinction
This opinion piece builds on an earlier finding that more than 90% of rainforest offset credits were ‘phantom credits’ with little to no impact on the environment, and argues that in fact some solutions may actually have a detrimental impact on the environment. Much like replacing gas with coal or cutting down trees to make way for wind turbines, no surprise here.
Technology Makes Us Better Humans?
This article describes just how workplace technology will make workers better humans. Remember annoying Clippy? He’s back, armed with an SJW sword or possibly an electric shock wristband. See for yourself:
“The whole employee experience is about to be turned upside down by technology, and it's going to be very beneficial for people. For example, when I do presentations, I get prompts from Speaker Coach within Microsoft Teams, and it tells me how quickly I’m speaking, or if I'm dominating a meeting. It gives me instant feedback that helps me become a better presenter. When I look in my recognition platform, AI nudges me to be more inclusive in my language. This type of technology will be really helpful for both information workers and frontline workers.”
Who doesn’t want more nudges into how to behave and ultimately what to think?