The Intersection of Crypto and AI: Maximizing Positives and Mitigating Negatives
Thoughts on the intersection of crypto and AI
by Ryan Shea
- The crypto doomsters were wrong - winter is over. Crypto is not the only technology basking in the warm spring sunshine. AI, which has itself experienced several winters in the past, is under going a spectacular renaissance.
- What we are witnessing from the likes of ChatGPT and its competitors is orders of magnitude better outputs than earlier “AI” systems were capable of producing just a handful of years ago.
- Many are worried about the social and economic impact from AI, as evidenced by the open letter signed by Elon Musk and other leading AI researchers calling for a pause on “Giant AI Experiments”.
- However, a moratorium on AI development is impossible to enforce. Therefore, we have to prepare for a future where AI increasingly encroaches of our lives in a myriad of different ways.
- The real question is can we as a species (I view this subject as that profound) guide AI’s development in such a way that maximize the positives while mitigating the negatives. This is where crypto comes in.
- Combining AI and blockchain technologies is a nascent and promising area for collaboration and many fruitful projects and businesses will be created as a result. However, the biggest win will be overcoming the centralizing nature of AI.
- Ownership and oversight of the underlying technologies should be broadened out and its evolution not left in the hands of a few private tech companies. Blockchain – the bedrock of crypto – is the means by which to achieve such an outcome.
Both in a meteorological sense, at least in the northern hemisphere, and in terms of crypto assets - winter is over. Year-to-date leading cryptocurrencies like Bitcoin and Ethereum1are up 85% and 76% respectively, gains that represent a substantial outperformance relative to other asset classes, like equities and bonds. Even analogue gold – as opposed to digital gold (AKA Bitcoin) – hasn’t managed to keep pace, up only 12% year-to-date, an increase that nevertheless puts it within spitting distance of record nominal highs.
The crypto doomsters who were loudly proclaiming 2023 a crypto extinction event have – once again – been proved wrong. I say that not just because of the magnitude of the rebound in crypto prices, although that is the most tangible sign, but because crypto prices have, in some sense, been rising for the right reason.
People hold Bitcoin and other cryptocurrencies for a variety of different motives2 but one of the most popular is their ability to provide a fiat money hedge3. Given the emergence of strains within the commercial banking system over the recent months, resulting in the closure or merger of several banks in the US and Europe, it would have been surprising if crypto hadn’t rallied under such conditions. Moreover, given the potential for such financial events to generate economic headwinds – the last FOMC minutes4 showed the Fed’s staff projecting a “mild recession starting later this year”– crypto has also benefited from the reinvigorated expectations of a Fed pivot, with 75bp of rate cuts now priced in by year-end (probably the end of QT too for reasons I outlined in a recent research note5).
Crypto is not the only technology basking in the warm spring sunshine. I am, of course, talking about AI, which having experienced several winters itself in the past6, is undergoing a spectacular renaissance. Fuelled by the explosive popularity of ChatGPT, which having acquired over 100 million users since it launched last November7 is the most successful consumer facing app ever, the media – both traditional and social – are full of articles outlining the latest developments in the space, and for good reason because the results are impressive.
The famous podcaster Joe Rogan recently posted the following tweet, after someone created a fictional podcast8 between himself and the CEO of OpenAI Sam Altman, generated solely by AI language models (presumably ChatGPT) combined with a text-to-speech AI that was able to accurately simulate their voices.
This follows a tweet I came across back in December9 showing the Dutch filmmaker Bob de Jong literally putting words into Morgan Freeman’s mouth – see image below:
More recently, the German artist Boris Eldagsen refused to accept a prize he won in the Sony world photography awards because he created the winning image using a technique that incorporated AI10. His “cheeky monkey” caper was a test to see if prestigious competitions are prepared for AI images to be submitted (it would seem not given his win) and to prompt an open discussion on the issue. Separately, an AI-generated Drake and Weeknd11 song released earlier this month got more than 20 million hits on TikTok12.
AI, at least as it appears to me, is everywhere.
What were are witnessing with such projects is orders of magnitude better outputs than earlier “AI” systems were capable of producing just a handful of years ago. Arguably, the current LLM models that power ChatGPT and others like them, are able to accurately imitate human-created content - the basis for the Alan Turing’s famous test13. AGI, or artificial general intelligence, may still be out of reach (it may be unreachable – who knows) but with machine intelligence already at such a level the social and economic implications will be profound.
Imagine, for instance, the aforementioned AI technologies combined (if you want to go full on Terminator mode add in some of the AI-powered robots produced by Boston Dynamics14). For anyone currently worried about “fake news” this should be extremely disconcerting because as the technology improves – which it does at an exponential not linear rate, hence why ChatGPT 4.0 is substantially better than ChatGPT 3.5 (see image below) - it will become extremely difficult to detect content that is AI-generated versus human-generated. Moreover, such outputs will be produceable at scale for widespread distribution over the internet – one of the primary sources of news for pretty much everyone on the planet15.
Exam Results (ordered by GPT-3.5 performance)
Moreover, this is potentially only the tip of the iceberg. The ability of generative AI to produce text/sound/images that are close to that produced by humans clearly has a lot of people bothered that it will result in widespread job redundancies because machines are generally more cost-effective than carbon-based employees and hence there is a strong profit incentive to switch out labour for capital. Indeed, a recent report by Goldman Sachs suggests that 300 million jobs could be lost due to the rolling out of this technology – an massive negative economic shock by anyone’s standards17.
Reflecting such concerns, the Future of Life Institute last month published an open letter18, signed by thousands including most notably Elon Musk and Steve Wozniak and many leading AI researchers. The letter called for a six-month moratorium on training more powerful AI systems than ChatGPT 4 in order“to jointly develop and implement a set of shared safety protocols for advanced AI design and development that are rigorously audited and overseen by independent outside experts”19.
Hitting the snooze button is clearly one strategy, but the problem with this approach is how does it get enforced? If a large company or nation-state (and I can think of more than a few examples straight off the top of my head, as I am sure many others can) decides to press ahead with AI research and model training, what can be done to stop them? Nothing. Just like Bitcoin, AI development is unbannable. A six month pause would simply give those who choose to ignore it the time to either catch-up or extend their lead in the AI arms race. The logic behind the proposal doesn’t stack up.
Hold The Prozac
That said, not everyone is so pessimistic about the impact of AI. Brian Roemmele, an engineer who has worked with AI and neural nets for decades and has been releasing SuperPrompts20via his website21and twitter account22, for instance believes the Luddite comparison is very pertinent to the current situation. For those few unfamiliar with Luddites, they were a group of weavers in 19thCentury Britain – led by Ned Ludd hence their name - who smashed up machines being introduced into the textile industry amid fears they would lose their jobs. As Brian points out, contrary to their initial beliefs (and the conclusion many people draw from this tale), Ludd and his followers discovered the introduction of machines was not a negative for them. Instead, machines allowed them to do more value-added work, which was not only less mundane, but also generated higher wages, more employment23and – hence – greater prosperity.
I have a great deal of respect for such optimistic views, and certainly history is one their side because the introduction of new technologies tends to precede greater economic wealth creation24. However, I can’t escape the nagging feeling that if – and I admit it is a big if – the current crop of AI models eventually leads to AGI and the creation of a resource competitor to humans then the analogy breaks down25.
That aside, given the impracticality of imposing an effective ban on AI development – temporary or otherwise - it is clear our futures will be increasingly impacted by AI and its various applications. The question is can we as a species (I view this subject as that profound) guide AI in such a way that benefits us all while also mitigating the potential negatives. This is where crypto comes in. But don’t take my word for it.
The AI Says...
In the spirit of embracing the brave new world we are inhabiting I decided to pose the following questions to ChatGPT whose unedited responses are shown in italics:
- Are AI and blockchain complementary technologies?
AI and blockchain are complementary technologies because they both have the potential to revolutionize the way businesses operate. AI can be used to analyze large amounts of data and make predictions about the future, while blockchain can be used to securely store and transmit data. Together, these technologies can be used to create more efficient and secure systems that can help businesses operate more efficiently and securely. AI can also be used to help automate certain processes in the blockchain and make transactions faster and more secure. Additionally, AI can help to create smarter contracts that are more secure and tamper-proof, while blockchain can help to ensure that those contracts are securely stored and cannot be altered or corrupted. Ultimately, the combination of AI and blockchain can help to create a more secure and efficient digital environment.
- Please can you provide specific examples?26
Specific examples of how AI and blockchain can be used together include:
Financial services: AI can be used to analyze customer data and provide personalized recommendations for financial products and services, while blockchain can be used to secure customer information and financial transactions.
Supply chain management: AI can be used to analyze data from multiple sources and provide insights into supply chain efficiency, while blockchain can be used to securely store and transmit data, track assets, and validate transactions.
Healthcare: AI can be used to analyze patient data to detect potential issues and provide personalized treatment plans, while blockchain can be used to securely store patient data and provide tamper-proof records of patient history.
Smart contracts: AI can be used to create smarter contracts that are more secure and tamper-proof, while blockchain can be used to securely store and transmit those contracts.
While I concur with the above, and believe many great businesses will be created to provide these services, to my mind they are not the biggest win from combining these two technologies.
AI models are inherently centralizing, in part due to their development and running costs. ChatGPT-3 cost an estimated $12million to train and $100,000 a day to to run27(I wonder how long it will take the green lobby to start pushing back against the electricity used in training AI models as it has done against cryptocurrencies, even though such criticisms are misplaced28). Even though unit costs of training models is falling, due to larger datasets and the need to increase computing power, some estimate that the costs of training State-of-the-Art (SoA) models could rise to as much as $100-$500million dollars29– funds that obviously not many people have access to. That is the reason AI is really dominated by only a handful of players, such as OpenAI, Google and Meta.
(Hot tip: A number of AI and crypto/AI projects are being launched and promoted on social media. If the project you are considering investing your hard-earned tokens into is simply calling the OpenAI API to access ChatGPT perhaps you should – at the very least – question whether the project is a viable business model given it is reliant on others to produce the inputs upon which its services are derived).
Is it really such a good idea to have a technology that will, without question, increasingly encroach upon on everyone’s daily lives in a myriad of different ways be controlled by only a few, private, organisations30?
Blockchains provide a means to make AI more transparent and more accountable because they are able to store and process the data used by AI systems in a decentralized fashion. They also provide the means to have decentralized governance based on well-known consensus mechanisms and can be deployed in a trustless, censorship-resistent manner. Sure, it will not overcome all the potential hurdles, especially if AGI ever comes to fruition, but at least it stops the power of AI from becoming concentrated in the hands of the few with little or no public oversight.
Decentralized blockchain based AI is not just a pipe dream. There are already several projects seeking to build out such systems. One example, that I came across in preparing this research note was SingularityNet31which is a decentralized AI marketplace running on Cardano that allows AIs to cooperate and coordinate. Another is Bittensor32, which is a permissionless peer-to-peer machine learning network that trains AI models collaboratively.
Another ancillary benefit of combining AI and blockchain technologies is that it can help tackle the “(deep) fake news” problem that is is likely to grow substantially in the not-too-distant future. By virtue of being on the blockchain, an independently verifiable history is created making it possible to determine when some content (video/voice/text) was created and/or when it come into someone’s possession (chain-of-custody). This will help by making it harder for malicious actors to create and distributed deepfakes undetected.
Certainly, I (and I suspect many others will) find this preferable to an alternative solution being put forward by Worldcoin, a Layer 2 protocol on the Ethereum blockchain co-founded by none other than OpenAI CEO Sam Altman. According to a recent blog post on their website33, Worldcoin is creating an open and permissionless identity protocol (World ID) to allow people to verify their “humanness” (Proof of Personhood in their lexicon) by collecting biometric data, iris patterns in their case. (Users submit their biometric data by staring into one of their orbs34. PS. Not sure the cat ears and whiskers make it any more appealing or user friendly.)
Worldcoin’s Eye Of Sauron
So there you have it. Winter is over, spring has finally arrived both for crypto and AI, two technologies with the potential to radically transform the world in the near future.
In many ways I agree with comments made by Gary Gensler who in his recent testimony before the US House of Representatives expressed his belief that “The technology that I think is most transformative in our time is artificial intelligence”. He followed that up by adding “It’s not crypto”. On the latter I disagree. For the reasons outlined in this research note, in my judgement to get the most benefits and, as importantly, mitigate the potential negatives, ownership and stewardship of the underlying AI technologies should be broadened out and its evolution not left in the hands of a few private tech companies. Blockchain – the distributed bedrock of crypto – is the means by which to achieve such an outcome. In doing so, perhaps we can turn two transformative technologies into something worth more than the sum of its parts, or as the title of this note suggests make 1 + 1 = 3!
Until next time.
Ryan Shea, crypto economist
1The successful roll-out of the Shapella fork earlier this month which allowed ETH staked on the Beacon chain to be withdrawn provided a idiosyncratic boost – see: https://www.theblock.co/post/226155/ethereum-shapella-price-2000?utm_source=twitter&utm_medium=social
3Namely the digital gold narrative, which I outlined in a previous research note - see: https://blog.trakx.io/interest-ing-times-part-i/
11I have no idea who these people are either.
15Either directly or indirectly given that almost all accredited journalists rely on online information sources when writing articles.
19Others – such as Eliezar Yudkowsky - have gone even further and called for the dismantling of data centres that power these AI models. Perhaps his strength of opposition is due to concerns about Roko’s basilisk, which was a thought experiment about a future AI that would be incentivized to create a virtual reality simulation to torture people who were previously aware of it potential existence but do not facilitate it’s creation. The thought experiment, which was posted back in 2010 on the LessWrong website co-founded by Eliezar Yudkowsky, was provocative to say the less. Go google it - see: https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/
20SuperPrompts – Brian’s trademark phrase - are text statements that can be pasted into ChatGPT and other text based AI systems to enhance the outputs of such systems.
21See: https://readmultiplex.com (FYI – it is a subscription service).
22He has one of the best accounts (@BrianRoemmele) on twitter in my opinion.
24The structural composition of the economy shifted from one predominantly based on agriculture, to industry and finally to services. During these transitions those workers who were unable, or unwilling, to retrain to match the skills required by the new industries lost out, but overall living standards still rose.
25When I raised this issue with Brian he, naturally given his perspective, disagrees believing it will instead give people far more personal power.
26Always be polite when asking ChatGPT questions because it improves the results.
30For anyone under the illusion that humans would never use AI for nefarious reasons please check out - https://www.youtube.com/watch?v=g7YJIpkk7KM and https://twitter.com/chaos_gpt
33Worldcoin is also envisaged as being a global network for the distribution of UBI (universal basic income) – a popular idea in some tech circles to deal with the employment destroying impact of AI - see: https://worldcoin.org/blog/engineering/humanness-in-the-age-of-ai